Exploiting artificial intelligence in precision oncology: an updated comprehensive review.

Cancer precision and personalized medicine

Cancer precision medicine tailors treatment based on the patient’s tumor, the patient’s genetic makeup, medical record, patient’s characteristics, lifestyle and environmental factors [ 1 – 3 ]. Precision oncology aims at optimizing the clinical outcome of anticancer therapy, preventing the adverse effects, and making healthcare more predictive, personalized, and cost-effective [ 3 – 5 ]. The pros and cons of precision medicine are extensively reviewed elsewhere [ 3 – 7 ]. In this review, we highlight the gap between the expected and real-life scenarios of exploiting targeted therapies and pharmacogenomics in cancer therapy. Artificial intelligence (AI) mimics human intelligence and has broad applications in various fields [ 8 – 13 ]. Machine learning (ML), a subset of AI, trains machines to automatically learn from the data by recognizing patterns and then make predictions based on the learnt patterns [ 11 , 14 ]. There are several ML algorithms exploited in medicine including linear regression algorithms, logistic regression algorithms and decision trees and random forests algorithms [ 14 , 15 ]. Artificial neural network (ANN), a ML algorithm, resembles the architecture and function of the biological neural networks of the brain. ANNs comprise nodes (i.e. simulating the neuronal cell bodies) linked via weighted connections (i.e. mimicking the axons and dendrites) [ 14 ]. The nodes of ANN receive inputs (data such as text or images), identify patterns and promote decision making [ 2 , 14 ]. ANNs have a single hidden layer which connects the input layer with the output layer. In contrast, deep neural networks (DNNs), another subset of ML, consist of multiple hidden layers (i.e. deep) which significantly enhance their accuracy and capability in terms of detecting features [ 14 , 16 ]. AI is deemed to transform the precision medicine of cancer patients [ 8 , 17 ]. Herein, we critically discuss the potential and limitations of AI-driven tools within: i) preclinical cancer research as well as ii) clinical oncology settings to fulfill the promise of cancer precision and personalized therapy.

Targeted therapy: Does it fulfill the promise of cancer precision medicine?

Exploiting the state-of-the-art technologies deepened our comprehension of the molecular mechanisms underlying the genetic and non-genetic aberrations implicated in tumorigenesis and cancer resistance [ 18 – 24 ]. Indeed, the treatment paradigm of diverse cancer types shifted from conventional chemotherapy to targeted therapy, a form of precision medicine [ 25 – 27 ]. In 2001, imatinib, the first-in-class tyrosine kinase inhibitor (TKI), was approved by the Food and Drug Administration (FDA) for treatment of mutated BCR-ABL (breakpoint cluster region (BCR) and abelson murine leukemia viral oncogene homolog 1 (ABL1)) chronic myeloid leukemia (CML) [ 28 , 29 ]. Although imatinib revolutionized the treatment of BCR-ABL CML patients, resistance to imatinib has been reported [ 28 , 30 – 32 ].

Several potent small molecules were developed to selectively target tyrosine kinase receptors such as epidermal growth factor receptor (EGFR), human epidermal growth factor receptor 2 (HER2) and FMS-like tyrosine kinase 3 (FLT3) which are overexpressed or constitutively activated in distinct types of cancer including breast cancer, lung cancer, bladder cancer, stomach cancer and acute myeloid leukemia (AML) [ 24 , 33 – 36 ]. Nonetheless, cancer resistance to TKIs as well as TKI-associated toxicities largely restrained their clinical efficacy [ 21 , 37 – 42 ].

Antibody-based therapy is another type of targeted cancer therapy [ 40 , 43 – 49 ]. Bevacizumab is an anti-vascular endothelial growth factor (anti-VEGF) developed to suppress angiogenesis [ 40 , 50 ]. Immune checkpoint inhibitors (ICIs) are monoclonal antibodies directed against the inhibitory brakes of the immune system including the cytotoxic T-lymphocyte-associated protein 4 (CTLA4), programmed cell death protein 1 (PD1) and programmed cell death ligand 1 (PDL1) [ 44 , 51 – 53 ]. Notably, only a modest proportion of the ICI-treated cancer patients undergo complete remission and the vast majority of the treated cancer patients suffer from the autoimmune related adverse effects of ICIs which are dose-limiting and mandate treatment interruptions [ 44 , 52 , 54 , 55 ].

Antibody-drug conjugates (ADCs) permit preferential delivery of cytotoxic agents to tumor cells [ 49 , 56 ]. Trastuzumab emtansine and trastuzumab deruxtecan are HER2-targeted ADCs which elicited clinical benefit in patients with HER2 low and HER high breast cancer [ 49 , 56 , 57 ]. Nonetheless, resistance to HER2-targeted ADCs was reported. Trastuzumab deruxtecan and to a lower extent trastuzumab emtansine were associated with drug-related interstitial lung disease or pneumonitis [ 58 ].

Thus, given their selectivity and relatively fewer adverse effects compared to conventional chemotherapy, targeted therapies were foreseen to render cancer “a curable disease”. Nonetheless, intrinsic and acquired cancer resistance as well as the specific dose-limiting toxicities of targeted therapies restrained their anticancer capability.

Pharmacogenomics: Does it fulfill the promise of cancer precision and personalized medicine?

Pharmacogenomics emerged as a fundamental field which has the potential to unleash – at least partially – the therapeutic potential of precision medicine [ 59 – 61 ]. Pharmacogenomics investigates how the genetic makeup modulates patient’s response to therapy in terms of pharmacokinetics (e.g. metabolism), pharmacodynamics, and toxicokinetics/toxicodynamics [ 59 , 62 ].

Methemoglobinemia and hemolytic anemia were reported in cancer patients with glucose-6-phosphate dehydrogenase (G6PD) deficiency following their treatment with rasburicase [ 59 , 63 , 64 ]. The latter is contraindicated in G6PD deficient patients [ 59 , 63 , 64 ]. Rasburicase – an oxidizing drug – is a uric acid oxidase (uricase) which converts uric acid to the more soluble metabolite “allantoin” and is used in the treatment of tumor lysis syndrome [ 59 , 63 , 64 ]. The Clinical Pharmacogenetics Implementation Consortium (CPIC) has published recommendations for adjusting the doses of three thiopurines (azathioprine, mercaptopurine, and thioguanine) based on thiopurine methyl transferase (TPMT) and nudix hydrolase 15 (NUDT15) genotypes [ 65 ]. Variant alleles of TPMT and NUDT15 with low enzymatic activities are associated with higher levels of thiopurines and/or their cytotoxic metabolites and hence myelosuppression [ 59 , 65 ].

Cancer pharmacogenomics evolved from the traditional perception of pharmacogenetics linked to specific drug-gene pairings, to include the state-of-the-art next generation sequencing (NGS) technologies such as whole exome sequencing (WES) or whole genome sequencing (WGS) which offer a more holistic “non-biased” analysis and hence could identify novel common and rare variants [ 1 ]. Nonetheless, WES and WGS are still challenged by their relatively high costs and problems linked to the storage, processing and interpretation of large amounts of data generated [ 66 , 67 ].

Exploiting artificial intelligence (AI)-driven tools in precision oncology

Given its multimodal architecture, AI is expected to transform precision medicine [ 2 , 8 , 68 – 74 ]. AI-driven algorithms have been integrated into NGS technologies including genomics, epigenomics, variant calling, transcriptomics, proteomics and single-cell sequencing [ 9 ]. AI-powered tools such as ML and DL optimized each step of the NGS workflow starting from the experimental design, library preparation as well as bioinformatics analysis [ 9 , 75 ]. AI-driven tools also processed and analyzed large datasets including that of electronic health records, diagnostic imaging and biomarkers [ 2 , 8 , 68 , 76 ]. AI-assisted gene editing tools as Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)-Cas9 promoted efficient targeted genome modifications [ 75 ]. The AI-driven AlphaFold neural network model integrated physical and biological knowledge about protein structure to accurately predict the three-dimensional (3D) structure of proteins with no similar known structures [ 77 ]. The technical algorithms and workflows of AI-based tools are extensively reviewed elsewhere [ 2 , 9 ]. Herein, we discussed the lessons learnt from exploiting AI-driven tools in both preclinical cancer research as well as clinical oncology.

Preclinical applications of AI-driven tools in basic and translational cancer research

Cancer driver genes are the genes whose mutations confer proliferative advantage to the affected cell [ 78 ]. Comprehensive PanCancer analysis using AI-driven algorithms identified 299 driver genes and more than 3000 putative mis-sense driver mutations across diverse types of cancer [ 78 ]. Identification of actionable driver mutations - for which targeted therapies exist - is of paramount importance for guiding clinicians towards precision medicine.

PandaOmics is a cloud-based platform which exploits AI and bioinformatic tools to analyze multi-omics datasets as well as biomedical text datasets to ultimately promote the discovery of biomarkers and therapeutic targets [ 79 , 80 ]. It also comprises an inClinico platform which predicts the probability of the transition of a Phase 2 clinical trial to Phase 3 [ 79 ]. Via exploiting PandaOmics, Chueng and colleagues pinpointed that voltage-dependent calcium channel subunit alpha-2/delta-1 (CACNA2D1) might act as a potential therapeutic target in Epstein-Barr virus positive (EBV + ) nasopharyngeal cancer [ 80 ]. Tethering of the EBV episome (or genome) increased H3K27 acetylation – which is a marker for active enhancers and promoters – in proximity to the promoter of CACNA2D1 gene in EBV + nasopharyngeal cancer cells. Indeed, nasopharyngeal cancer expressed higher CACNA2D1 transcript and protein levels compared to normal nasopharyngeal samples. Genetic depletion of CACNA2D1 in EBV + nasopharyngeal cancer cell line evidently reduced its proliferation in vitro and in vivo [ 80 ].

Within the context of structure-based drug discovery, AI-powered tools were used for ligand identification, de novo drug design and prediction of the ligand-protein binding pose and potency [ 75 , 81 ]. MolProphet is a cloud-based AI-driven drug discovery platform which promoted target pocket prediction, structure-based and ligand-based virtual screening, hit discovery, lead optimization, de novo molecule generation and scaffold hopping [ 82 ]. Pandey utilized MolProphet platform to generate novel small molecules targeting aurora kinase A (AURKA) which is overexpressed in diverse types of cancer and acts as a synthetic lethal partner of several tumor suppressor genes [ 83 – 85 ]. Molecular simulations, molecular dynamics analysis, and binding energy calculations suggested that – out of 200 novel compounds − 4-(5-fluoro-6-[(1Z)-3-hydrazinyl-3-oxo-2-phenylprop-1-en-1-yl]pyridin-2-yl)benzoic acid could be a candidate AURKA inhibitor. However, this study relied only on in-silico simulation and in vitro validation is warranted [ 83 ]. AI-driven tools also have the potential to advance the field of polypharmacology (which refers to small molecules interacting with multiple targets simultaneously) through predicting their selectivity and off-target effects. Nonetheless, experimental validation methods are needed to confirm the selectivity and absence of off-target effects [ 86 ].

Artificial intelligence is also speculated to optimize the therapeutic outcome of antibody-based cancer therapy [ 2 , 87 ]. Within the context of chimeric antigen receptor (CAR) T-cell therapy, AI-assisted platforms are foreseen to optimize target identification, vector design, manufacturing and personalized data-driven clinical decisions [ 87 ].

Deep neural network Integrating Prior Knowledge (DIPK) - a DL-based framework - was developed to predict the susceptibility of cancer cell lines to pharmacological agents (Fig. 1 A). DIPK comprised: i) cell modeling: which was acquired from the gene interaction (or interactome features) as well as gene expression (or transcriptome features), ii) drug modeling: the molecular graphs of drugs were obtained from the atomic and chemical bonding details of the drugs extracted from their Simplified Molecular Input Line Entry System (SMILES) string, iii) fusion of the interactome features, transcriptome features and atomic features and iv) output which was the prediction of the logarithm of the median inhibitory concentration (LN IC50) [ 88 ]. The performance of DIPK was evaluated using the Genomics of Drug Sensitivity in Cancer (GDSC) dataset. GDSC is a comprehensive resource that includes the gene expression profiling of cancer cell lines together with their IC50 to distinct drugs [ 92 ]. Notably, there was a strong correlation between the predicted and observed LN IC50 values [ 88 ]. DIPK also predicted drug response at a single-cell level. Moreover, DIPK forecasted better response of breast cancer patients to paclitaxel within the pathologic complete response cohort compared to the residual disease cohort. Nevertheless, further systematic large-scale prospective studies are needed to evaluate the potential of incorporating DIPK in decision-making and devising personalized treatment strategies for the people of cancer [ 88 ]. Fig. 1 Simplified illustration of the AI-powered frameworks exploited in preclinical cancer research. A) Deep neural network Integrating Prior Knowledge (DIPK) framework was constructed to predict the drug responsiveness of cancer cell lines [ 88 ]. DPIK integrates: i) cell modeling: the interactome features were captured from the gene interaction network and the transcriptome features were obtained from gene expression profiles of cancer cell lines and ii) drug modeling: the atomic features were derived from the atomic and chemical bonding details of a drug. The output of DIPK was the prediction of the logarithm of the median inhibitory concentration (LN IC50).B) Schematic representation of the machine learning (ML) and deep learning (DL) models which integrated multi-omics datasets (including genetic mutation, gene expression, DNA methylation and copy number variation (CNV), drug chemical properties and drug sensitivity (LN IC50) of cancer cell lines treated with drugs (obtained from Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE)) as learning data for predicting anticancer drug responsiveness (IC50) [ 90 ]. C) “XGraphCDS” comparative learning framework was developed to predict differential cancer responses to pharmacotherapy. This model was based on cancer gene expression profile and the chemical structure knowledge of the drugs. The output was quantitative (LN IC50) or qualitative (sensitivity or resistance). An in vivo prediction model for cancer patients was constructed from the in vitro data [ 91 ]. D) Workflow of “DrugGene” framework. This model combined two types of neural networks: i) visual neural network (VNN) which integrated the mutational landscape, CNV and gene expression profiles of cancer cells and ii) artificial neural network (ANN) which input the chemical structural features of anticancer drugs. The output of DrugGene was the predicted area under the curve (AUC) where 0 and 1 reflected complete cell killing and no effect respectively [ 92 ]. E) Workflow of the CAN-Scan platform. Primary patient-derived colorectal cancer cells were characterized for their drug response using clinically relevant doses of FDA-approved drugs, whole exome sequencing (WES) and whole transcriptome sequencing (WTS). ML algorithms were used to train drug-specific predictive models to identify prognostic biomarkers. This figure was prepared at least partially using Biorender software

To de-noise and mitigate the inconsistencies observed with drug sensitivity data matrices including that of Cancer Therapeutics Response Portal (CTRP) and GDSC, Hu and colleagues introduced a DL framework called residual threshold deep matrix factorization (RT-DMF) [ 93 ]. RT-DMF yielded corrected datasets which better correlated with each other and conferred better predictive potential than the original datasets [ 93 ].

Regression-based ML and DL models were constructed to predict anticancer drug response (LN IC50) by using the IC50 of cancer cell lines treated with drugs (from the GDSC and the Cancer Cell Line Encyclopedia (CCLE)), multi-omics datasets (including that of genetic mutation, gene expression, DNA methylation and copy number variation (CNV)) and drug chemical properties (Fig. 1 B). The performance of DL models outperformed that of ML [ 89 ]. DL frameworks comprised two convolutional neural network (CNN) models: i) Cancer Drug Response profile scan (CDRScan) and ii) residual neural network (ResNet). CDRScan was based on a drug screening assay supported by the genomic somatic mutational landscape of 787 human cancer cell lines and structural profiles of 244 drugs [ 94 ]. DNNs are challenged with the vanishing gradient problem which takes place when gradients computed during backpropagation become very small as they move backward across layers. This hinders the learning capability of the early layers which are closer to the input and thus, restrains effective training of the network. Notably, ResNet introduced the so-called “residual connections” which ameliorated the gradient vanishing problem, captured more complex features and permitted more efficient training of deeper networks. Indeed, the predictive capability of anticancer drug responsiveness using ResNet was better or comparable to that of CDRScan [ 89 ].

DeepDR, a DNN model, was developed to predict tumor response to drug therapy [ 95 ]. The model was trained using mutational and transcriptomics data of human cancer cell lines deposited in the GDSC database and ultimately applied to The Cancer Genome Atlas (TCGA) data to predict a patient tumor’s response to anticancer drugs [ 92 , 95 ]. Indeed, DeepDR forecasted the IC50 values of 265 drugs and the drug responses of 33 cancer types (e.g. EGFR inhibitors in non-small lung cancer and. vinorelbine for titin-mutated tumors) [ 95 ].

Wang and colleagues formulated another neural network known as “XGraphCDS” to predict differential cancer responses to pharmacotherapy [ 90 ]. This comparative learning framework was based on cancer gene expression profile (from the GDSC and CCLE) and the chemical structure knowledge of the drugs. The output was quantitative (LN IC50) or qualitative (sensitivity or resistance) (Fig. 1 C). An in vivo prediction model for cancer patients was constructed from the in vitro data [ 90 ].

Although IC50 is widely used as a prediction variable in regression models, more concrete measures such as the area under the curve (AUC) and area above the dose response curve (AAC) were proposed to be more robust [ 96 ]. Relying on continuous values rather than discrete (single) values of the IC50 was also recommended [ 96 ]. “DrugGene” - a DL model - was generated to predict the drug sensitivity via combining two types of neural networks: i) visual neural network (VNN) which modelled the hierarchical molecular subsystems within cancer cells via integrating their mutational landscape, CNV and gene expression profiles and ii) artificial neural network (ANN) which input the chemical structural features of anticancer drugs (Fig. 1 D). The output of the predicted anticancer drug sensitivity was in the form of normalized AUC where 0 and 1 reflected complete cell killing and no effect respectively [ 91 ].

Kuenzi and colleagues developed a DL model called “DrugCell” which comprised two neural networks [ 97 ]. The first neural network simulated the intracellular hierarchical organization and identified which cellular subsystems (e.g. pathways and organelles) contributed to the anticancer responses (expressed as AUC) taking into consideration the mutational profile of cancer cells [ 97 ]. The second neural network integrated the chemical structure of the drugs [ 97 ]. It is worth mentioning that “DrugCell” predicted potentially effective comboregimens for AML therapy as lenalidomide and DNA methyl transferase (DNMT) inhibitors. Although meta-analysis suggested that lenalidomide + azacitidine (DNMT inhibitor) might be superior to azacitidine monotherapy in AML therapy, randomized clinical trial (RCTs) negated the additional beneficial effect of this combination [ 98 , 99 ]. Thus, identifying predictive biomarkers are needed to identify the subset of the AML patients who are most likely to benefit from this comboregimen.

Many preclinical cancer therapies fail during clinical trials [ 100 ]. To better mimic clinical scenarios, Chia and colleagues presented CAN-Scan, which exploited ML to integrate the drug response signatures of primary patient-derived colorectal cancer cells tested against clinically relevant doses of 84 FDA-approved drugs with their multi-omics characterization (e.g. genetic alterations and transcriptomic signatures) [ 101 ]. Importantly, CAN-Scan linked overexpression of cell adhesion gene, increased BRAF/mitogen-activated protein kinase (MAPK) signaling as well as focal copy-number gain/amplification in chromosome 7 to the resistance of colorectal cancer to 5-fluorouracil [ 101 ]. Although CAN-Scan did not consider the potential impact of the tumor microenvironment (e.g. immune cells and fibroblasts) on cancer response, it accurately predicted the response of four cohorts of ethnically distinct colorectal cancer patients [ 101 ]. Drugs such as regorafenib and vemurafenib were proposed as a potential therapeutic option for treating BRAF + CRC patients who are irresponsive to 5-fluorouracil [ 101 ]. Nonetheless, large-size multicenter prospective studies are warranted to validate this postulation.

Clinical applications of AI-powered frameworks in cancer management

To obtain a broader overview of the applications of AI-driven tools in clinical oncology, we systematically searched the registered ClinicalTrials.gov studies which exploited AI-assisted tools (Supplementary Table 1 ). After omitting the false-positive search trials, 397 clinical trials were spotted; more than 50% of which evaluated the potential utility of AI-assisted frameworks in cancer diagnosis (Fig. 2 A). Approximately 28.6 and 12% of the 397 clinical trials aimed to assess the potential utility of AI-powered tools for non-diagnostic purposes and for guiding treatment decisions respectively (Fig. 2 B). The vast majority of the AI-assisted clinical trials were carried out in China (~33.5%) followed by the US (~9%) and Italy (~8%) (Fig. 2 C). In Table 1 , we shed light on selected AI-driven tools approved by the FDA to be used for cancer patients [ 17 ]. This section reviewed the real-world data of AI-powered models deployed in the screening, diagnosis, prediction of drug response, clinical trial matching, guiding informed treatment decisions as well as supportive care of cancer patients. Fig. 2 Applications, cancer types and locations of artificial intelligence (AI)-assisted clinical trials registered on ClinicalTrials.Gov database. ( A ) Applications of AI-powered tools in clinical trials ( n = 397): diagnostic biomarker (~54%), non-diagnostic biomarker (28.6%), treatment decisions (~12%), supportive care (~4%) and other applications. ( B ) Cancer types in AI-assisted clinical trials [colon cancer (~22.7%), lung cancer (9%), head and neck cancer (8%), breast cancer (7.8%), skin cancer (6.3%), prostate cancer (6.3%), oesophageal cancer and gastric cancer (6.3%), liver cancer (~5%), cervical cancer (~3.5%) and bladder cancer (~3.3%)]. ( C ) Top ten countries with the largest number of clinical trials registered on ClinicalTrials.Gov database which evaluated the utility of AI-driven tools in cancer management Table 1 Selected list of artificial intelligence (AI)-enabled tools approved by the Food and Drug Administration (FDA) for marketing in the United States Device Description Computer-assisted diagnostic software for lesions suspicious for cancer It characterizes lesions based on the features extracted from the images of magnetic resonance, mammography, radiography, or computed tomography. Radiation Planning Assistant (RPA) An AI-assisted tool that provides automated contouring and planning solutions for the treatment of several tumor and non-tumor specific radiotherapy indications. It is specifically designed for use in radiotherapy providers in low-income and middle-income countries. SpotLight/SpotLight Duo with Low Dose Lung Cancer Screening Option The low dose computed tomography (CT) X-ray screening option uses low dose CT for the screening and diagnosis of lung cancer. Pathwork® Tissue of Origin Test Kit-FFPE (Formalin-Fixed Paraffin-Embedded) An in vitro diagnostic test which assesses the degree of similarity between the gene expression patterns in a patient’s FFPE tumor and the gene expression profiles in a database of fifteen tumor types (including bladder cancer, breast cancer, colorectal cancer, gastric cancer, testicular germ cell cancer, hepatocellular cancer, kidney cancer, non-small cell lung cancer, non-Hodgkin’s lymphoma, melanoma, ovarian cancer, pancreatic cancer, prostate cancer, sarcoma, and thyroid cancer). Avenda Health AI Prostate Cancer Planning Software “Unfold AI” An AI-based software which integrates patient data and deep learning algorithms to enhance the accuracy of the diagnosis and treatment planning of prostate cancer. Opulus™ Lymphoma Precision An AI-powered software which analyzes whole-body fluorodeoxyglucose positron emission tomography/computed tomography (FDG-PET/CT) scans of patients with FDG-avid lymphomas (i.e. with high levels of absorbed FDG and increased metabolic activity). TumorSight Viz An AI-driven 3D visualization and analysis tool of breast MRI images from early stage or locally advanced breast cancer patients. ArteraAI Prostate test An AI-driven risk stratification tool for patients with non-metastatic prostate cancer. QUIBIM Precision Prostate (qp-Prostate) An AI-assisted tool used to enhance the detection and diagnosis of prostate cancer. GI Genius™ system An AI-assisted colonoscopy tool designed to help endoscopists detect colorectal polyps and colon adenomas. AI: artificial intelligence, CT: computed tomography, FDG-PET: fluorodeoxyglucose positron emission tomography, FFPE: formalin-fixed paraffin-embedded, RPA: radiation planning assistant

Exploiting AI-driven tools as biomarkers in cancer patients

There are distinct types of biomarkers including diagnostic, monitoring, pharmacodynamic/response, predictive, prognostic, safety, and susceptibility/risk biomarkers [ 102 – 104 ]. Most of the Clinicaltrials.gov studies evaluated the potential added value of co-implementing AI-powered tools in cancer screening and diagnosis with a peculiar interest in colorectal cancer (Supplemental Table 1 ). Herein, we highlighted the lessons learnt from exploiting AI-supported frameworks as biomarkers.

Radiomics refers to the extraction of quantitative phenotypic features from radiological images in an automated, high-throughput manner [ 105 ]. These features are then incorporated into AI-driven tools or statistical tools to perform a classification task to promote risk stratification, prognosis prediction or response prediction [ 105 ]. A ML model was constructed based on the clinical data and “radiomics signature” stemming from the imaging features from pelvic magnetic resonance imaging (MRI) to predict inguinal lymph node metastasis in pre-operative patients with vulvar cancer [ 106 ]. ML models integrated computed tomography (CT) image measurements and clinicopathological data (including blood tests and primary tumor pathology) from pre-operative colorectal cancer patients to ultimately predict the risk of mesenteric lymph node metastasis (Fig. 3 A) [ 107 ]. Similarly, ML models were developed to predict pelvic lymph node metastasis in pre-operative early-stage cervical cancer patients via combining the pre-operative CT measurements of the whole abdomen and pelvis and clinicopathological data (Fig. 3 B) [ 108 ]. External validation of these models is warranted to ensure their generalizability. Fig. 3 Schematic diagram of distinct applications of artificial intelligence (AI)-powered models which integrated radiomics features. ( A ) Machine learning (ML) models were based on integrated computed tomography (CT) image measurements and clinicopathological data (including blood tests and primary tumor pathology) from pre-operative colorectal cancer patients to predict the risk of mesenteric lymph node metastasis [ 107 ]. ( B ) ML models were developed to predict pelvic lymph node metastasis in pre-operative early-stage cervical cancer patients via integrating the pre-operative CT images of the whole abdomen and pelvis and clinicopathological data [ 108 ]. ( C ) ML models were developed to predict the location of recurrences of glioblastoma from post-operative MRI scans of glioblastoma patients [ 109 ]. ( D ) ML-assisted framework integrated basic non-small lung cancer, patient characteristics, histology, PDL1 tumor proportion score, driver oncogene status, patient’s anticancer treatment histories, blood laboratories findings and radiomics features of the primary tumor from contrast-enhanced CT images to predict personalized overall survival [ 110 ]. This figure was prepared at least partially using Biorender software

Meta-analysis of 9 studies exploiting AI-driven models in the diagnosis of soft tissue sarcoma revealed their superior accuracy and sensitivity [ 111 ]. Since the analyzed studies were retrospective, large size multicenter prospective studies are needed to test the generalizability of the results.

Kang and colleagues developed a DL model which accurately measured the volume of meningioma based on contrast-enhanced MRI scans [ 112 ]. Discriminating edema from glioblastoma is challenging [ 109 ]. To this end, a multi-institutional retrospective study by Cepeda and colleagues developed a ML model which distinguished glioblastoma from edema and predicted the location of tumor recurrences from post-operative MRI scans of glioblastoma patients (Fig. 3 C). However, this study lacked anatomopathologic confirmation of the so-labeled recurrences areas and hence cannot exclude pseudoprogression [ 109 ].

Radiomics models supported by ML algorithms were also constructed to predict disease-free survival and overall survival in patients with locally advanced cervical cancer after concurrent chemoradiotherapy [ 113 ]. Despite an increasing number of studies evaluating the use of AI-driven algorithms to promote the response predictive capability of radiomics, these tools are still not widely adopted in clinical practice. This could be – at least partially - due to the lack of large-scale multicenter studies and standardized methods that do not enable a meaningful meta-analysis [ 114 ].

Koyama and colleagues developed an AI-based personalized survival prediction model according to treatment selection for patients with advanced non-small cell lung cancer (NSCLC) (Fig. 3 D). This ML-assisted framework integrated baseline patient characteristics including age, sex, stage, histology, PDL1 tumor proportion score, driver oncogene status (as EGFR mutation, ALK fusion, ROS proto-oncogene 1 receptor tyrosine kinase (ROS1) fusion, and B-Raf proto-oncogene serine/threonine kinase (BRAF) mutation), patient’s anticancer treatment histories, laboratories findings and radiomics features of the primary tumor from contrast-enhanced CT images [ 110 ].

Probe based confocal laser endomicroscopy promotes accurate diagnosis of esophageal squamous neoplasms [ 115 ]. Nonetheless, analysis of probe based confocal laser endomicroscopy images mandates extensive experience. An AI-aided confocal laser endomicroscopy system displayed higher sensitivity compared to expert endoscopists and boosted the accuracy and sensitivity of non-expert endoscopists in the diagnosis of esophageal squamous neoplasms [ 115 ]. A RCT (trial registration ChiCTR2100044126) demonstrated that AI-assisted endoscopy (using DL algorithms) significantly boosted the detection rate of esophageal cancer and precancerous lesions compared to unassisted endoscopy and was safe [ 116 ].

Recognition of the intra-abdominal metastasis in advanced gastric cancer by the AI-aided laparoscopic exploration system matched that by novice (inexperienced) surgeons, but outperformed novice surgeons in identifying small lesions [ 117 ]. An international multicenter study revealed that real-time polyp assessment during colonoscopy using AI-aided diagnosis did not significantly increase the sensitivity for small neoplastic polyps compared with visual inspection [ 118 ].

Changes in the tongue features may reflect GIT diseases especially in the stomach [ 115 ]. Tian and colleagues reported that integrating ML with tongue diagnosis analysis might serve as an non-invasive tool for early detection of gastric cancer specifically in resource-limited healthcare facilities [ 115 ]. ML models were trained using baseline clinical data, tongue features (e.g. color, coating and texture), and tumor marker levels. Nonetheless, external validation and technical standardization of this AI-assisted tool are warranted.

To diagnose bladder cancer, urologists rely on two fundamental tests: cystoscopy and cytology [ 119 ]. Many patients abandon cystoscopy given its invasive nature. Although voided cytology – which involves cell morphological analysis of urine biospecimens - is simple, non-invasive and thereby widely accepted by the patients, it mandates highly experienced uropathologists and has low sensitivity in diagnosing urothelial carcinoma [ 119 ]. To this end, a multicentre trial evaluated the diagnostic potential of VisioCyt®, a digital AI-assisted medical device, in voided urine samples obtained from patients with urothelial carcinoma and from negative controls (Fig. 4 A) [ 119 , 120 ]. Notably, VisioCyt® exhibited higher sensitivity compared to conventional cytology carried out by experienced uropathologists [ 119 , 120 ]. Fig. 4 Simplified illustration of the workflow of artificial intelligence (AI)-driven frameworks used for the detection of cancer and its grade. ( A ) Simplified illustration of the workflow of VisioCyt®, an AI-assisted medical device. This deep learning (DL) algorithm automatically analysed digital cytology slides of voided urine samples to ultimately promote the detection of urothelial bladder cancer [ 119 , 120 ]. ( B ) Spatial transcriptomics, pseudotime trajectory and machine learning (ML) algorithms were exploited to identify biomarkers which predicted the grade of prostate cancer. This figure was prepared at least partially using Biorender software

Multiple myeloma (MM) is characterized by the proliferation of malignant plasma cells in the bone marrow [ 121 ]. Monoclonal Gammopathy of Undetermined Significance (MGUS) is a premalignant plasma cell dyscrasia associated with elevated levels of IgG immunoglobulin without end-organ damage [ 121 , 122 ]. To distinguish MM from MGUS at the molecular level, Ruhela and colleagues developed a targeted sequencing panel (295-gene panel for MM) using an AI-driven attention-based model called “Bio-Inspired Graph Network Learning-based Gene-Gene Interaction (BIO-DGI)” [ 122 ]. This framework integrated variant profiles such as single nucleotide variants (SNVs), CNVs, structural variants (SVs) and loss-of-function (LOF) variants from WES and WGS datasets of MM and MGUS patients to ultimately identify genes relevant to MM. These combined profiles identified 354 candidate genes which were then filtered to 346 genes after Geo2R validation. After omitting MGUS-specific genes, a list of 282 genes emerged. These genes were categorized into: 70 disease-initiating genes (which were altered in MM and MGUS) and 212 disease-transformative genes (which were specifically altered in MM). Another 13 well-established biomarkers obtained from literature as well as existing MM-related panels were added to formulate the 295-gene panel [ 122 ].

A 5-fluorouracil resistance-related signature (5-FRSig) model was constructed to predict the prognosis of colorectal cancer patients and identify the most likely to respond for chemotherapy and immunotherapy [ 123 ].

A deep residual learning model predicted microsatellite instability from the H&E stained histology slides of gastrointestinal cancers such as stomach adenocarcinoma and colorectal cancer [ 124 ]. Consistently, a DNN identified different tissue types including stromal tissue in the H&E-stained histological images of colorectal cancer which ultimately improved the prediction of the survival of colorectal cancer patients [ 125 ]. HIBRID: a DL-based analysis of H&E stained whole slide images combined with circulating tumor DNA (which detects molecular residual disease) enhanced risk stratification in patients with stage II-IV colorectal cancer [ 126 ]. Nonetheless, large-scale multicenter prospective validation studies are warranted to support the routine clinical use of this prognostic biomarker.

Although serum prostate-specific antigen (PSA) is commonly used for screening prostate cancer, its compromised sensitivity and specificity lead to unneeded biopsies or false-negative diagnosis [ 127 , 128 ]. Smelik and colleagues speculated that reliable biomarkers could be spotted after taking into consideration the following (Fig. 4 B): i) the distinct grades of malignant transformation within the same tumor. Thus, spots of normal glands, prostatic intraepithelial neoplasia, and five grades of prostate cancer on H&E–stained whole-slide images were annotated and included in the analysis, ii) distinct grades of cellular progression at the molecular level, could be captured and characterized using spatial transcriptomics, and iii) linear trajectory could be constructed using Pseudotime and based on the similarity of the expression profiles of these cells [ 129 ]. Notably, 45 genes positively or negatively correlated with the trajectories of Pseudotime. The diagnostic potential of these genes were further assessed in the sera, prostate tissue and urine samples of 2,000 prostate cancer patients and healthy controls. These biomarkers outperformed serum PSA levels in predicting cancer grades. Nonetheless, large-size prospective multicenter studies are urged to validate the sensitivity and specificity of these biomarkers [ 129 ].

Exploiting AI-driven tools in devising treatment protocols of cancer patients

Frameworks powered by AI were used to guide medical oncologists towards rational treatment decisions [ 130 ]. Indeed, IBM Watson for Oncology (WFO), a cognitive computing system trained by the Memorial Sloan Kettering Cancer Center (MSKCC), exploited natural language processing to find specific attributes in a patient’s record and ranked evidence-based treatment options into three categories: recommended, for consideration and not recommended (Fig. 5 A) [ 130 ]. Natural language processing algorithms process and interpret human language [ 132 , 133 ]. Indeed, WFO has been tested in many countries including the US, China, India and Netherlands. Using retrospective data from 362 cancer patients, Zhou and colleagues examined whether there was a concordance between the treatment decisions proposed by the WFO and that of the multidisciplinary tumor board [ 130 ]. Intriguingly, different concordance rates were observed with different cancer types: ovarian cancer (96%), lung cancer and breast cancer ( > 80%), rectal cancer (74%), colon cancer and cervical cancer (64%) and gastric cancer (12%). This discrepancy was speculated to be secondary to different incidence rates and available therapies in the US and China [ 130 ]. Fig. 5 Clinical applications of artificial intelligence (AI)-supported tools in treatment making decisions as well as supportive care of cancer patients. ( A ) Simplified schematic diagram of the workflow of IBM Watson for oncology (WFO), a cognitive computing system which integrated the guidelines of the National Comprehensive Cancer Network (NCCN) for cancer therapy and was trained by the Memorial Sloan Kettering Cancer Center (MSKCC) which exploited natural language processing to find specific attributes in a patient’s record and aimed at guiding the physicians towards evidence-based treatment options ranked into three categories: recommended, for consideration and not recommended [ 130 ]. ( B ) Workflow of the knowledge model for potential interaction and potential consideration screening (KM2PS)” [ 131 ]. This approach integrated both genetic and non-genetic factors to improve four types of screening before prescribing medication: i) drug-gene interactions (DGI), ii) drug-drug interactions (DDI), iii) drug-herbal interactions (DHI), and iv) drug-social lifestyle interactions (DSI) [ 131 ]. This framework also considered the liver function (LF) and renal function (RF). ( C ) Illustrative figure of chemotherapy naïve cancer patients with anxiety and depression who received ChatGPT-based digital counselling to relieve their anxiety and depression. This figure was prepared in part using biorender software

Since cancer cells depend on multiple driver alterations whose oncogenic effects can be abrogated using combination regimens, Li and colleagues developed a comprehensive resource called “Recurrent Features Leveraged for Combination Therapy (REFLECT)” which integrated ML and cancer informatics algorithms to propose precision combination regimens tailored to target recurrent oncogenic co-alterations across patient cohorts [ 134 ].

To avoid potential harm from polypharmacy and foster personalized prescription, Jamrat and colleagues developed the so-called “knowledge model for potential interaction and potential consideration screening (KM2PS)” (Fig. 5 B) [ 131 ]. To construct KM2PS, knowledge related to potential drug-gene interactions were obtained from distinct tertiary drug information databases including Clinical Pharmacogenetics Implementation Consortium, Pharmacogenomic Knowledge Base, and Table of Pharmacogenomic Biomarkers in Drug Labeling by the US FDA. Knowledge linked to drug-drug interactions, liver functions and renal functions were acquired from Lexicomp, DailyMed, and Drugbank. Drug interactions involving cytochrome P450 were obtained from the Drug Interactions Flockhart Table [ 131 ]. Drug-herbal interactions knowledge was obtained from the Medplant database (Mahidol University, Thailand). The input included patient’s prescriptions, medical record, pharmacogenomics data, liver and renal function, the herbal supplements used and social lifestyle (e.g. consumption of caffeine, alcohol and smoking). Thus, this approach integrated both genetic and non-genetic factors to improve four types of screening before prescribing medication: i) drug-drug interactions, ii) drug-gene interactions, iii) drug-herbal interactions, and iv) drug-social lifestyle interactions [ 131 ]. This model was proposed to promote risk-benefit analysis and informed clinical decision-making to ensure that the patient receives the optimum personalized treatment (i.e. maximum efficacy and minimum harm) [ 131 ].

Ammo and colleagues assessed the decision-making capability of ChatGPT-4o in five simulated cases of sarcoma [ 135 ]. Moderate effectiveness of ChatGPT-4o’s recommendations was reported following the evaluation of a panel of an orthopedic surgeon, plastic surgeon, radiation oncologist, radiologist, and pathologist [ 135 ].

Within the context of supportive care, a RCT ( NCT06854315 ) assessed the potential effect of a ChatGPT-based digital counselling intervention versus standard clinician-led education group on anxiety and depression in 160 chemotherapy-naïve cancer patients (Fig. 5 C) [ 136 ]. Intriguingly, the ChatGPT-based counseling significantly mitigated the anxiety and depression in pre-chemotherapy cancer patients compared to the controls [ 136 ]. Indeed, further research is required to validate these findings in large-sized multicentre prospective RCTs and refine AI adoption in the supportive care of cancer patients.

Given heightened demand for genetics services and limited availability of certified genetic counselors, women with stage 0-III breast cancer who did not meet the NCCN criteria for genetic testing were randomized to pre-test counselling with a Chatbot or a certified genetic counselor [ 137 ]. Of note, there was no significant difference in the median knowledge score and the patient satisfaction score between in-person counselling and automated Chatbot cohorts which would enable better access to healthcare services [ 137 ].

Many clinical oncologists do not have advanced programming expertise to carry out bioinformatic analyses of large datasets of genomic data, patients’ clinical metadata and treatment outcomes (e.g. survival) [ 138 ]. To solve this problem, the Artificial Intelligence Agent for High-Optimization and Precision Medicine (AI-HOPE), an AI-driven conversational agent system was developed (Fig. 6 ). In this framework, users stated their research question, and AI-HOPE converted natural language into structured analytical workflows. For instance, if the research question was “Do AML patients with FLT3 mutations have different progression-free survival?”, AI-HOPE would initiate a case-control study. On the other hand, if the research question was “Is BRAF mutation in breast cancer linked to brain metastasis?”, AI-HOPE would pursue a data association analysis [ 138 ]. AI-HOPE then carried out integrative data analysis of genomics and metadata to generate outputs such as odds ratio, hazard ratios and Kaplan-Meier survival curves. The closed system of AI-HOPE prevented the leakage of clinical data. AI-HOPE pinpointed that TP53 mutations were more prevalent in late-stage (stage III/IV) versus early-stage (stage I/II) colorectal cancer patients [ 138 ]. Fig. 6 Simplified schematic diagram of the workflow of the artificial intelligence agent for high-optimization and precision medicine (AI-HOPE) [ 138 ]. This figure was prepared at least partially using biorender

Thus, AI-driven tools have the potential to promote precision cancer medicine via enhancing the diagnosis, screening, biomarker discovery, clinical trial matching, treatment decision and supportive care of cancer patients. Nonetheless, large-size multi-center prospective studies are required to validate the utility of these AI-powered frameworks.

Limitations of exploiting AI-driven tools in precision oncology and proposed roadmap

Despite the potential of AI-driven tools, there are several challenges which restrain their successful clinical implementation [ 10 , 15 , 139 ]. To accurately and precisely classify or predict tasks, ML and DL algorithms mandate sustainable supply of large datasets [ 15 , 140 ]. Nonetheless, large datasets are not always readily available and might not be continuously provided to further optimize/train the AI-driven models [ 15 ]. This could be in part due to the reluctance of some healthcare institutions to share patients’ data, data privacy issues, compromised data quality and short-term storage of datasets. Modest awareness of the physicians (~25%) about employing AI in medicine was spotted among the challenges of AI in the healthcare industry [ 141 ]. Thus, promoting global awareness about the importance of sharing patients’ datasets to train AI-driven models is needed to ultimately improve the clinical outcome of cancer patients. Obtaining patient consent, data encryption and modifying the regulations and rules governing the processing of personal data could also solve data privacy and security issues. The quality of large datasets is usually suboptimal (i.e. they may include missing or inaccurate data). Indeed, accurate assessment of the data labels fed into AI models require multiple subjects to ensure correct data-annotation [ 140 ]. Standardization of the electronic health records and interoperability have been proposed to integrate datasets from different healthcare institutes [ 15 ]. The relatively short lifespan of stored datasets could be overcome by the cloud computing servers.

Biases (e.g. racial biases) may take place during data collection and hence could compromise the predictive performance of AI-powered models which were trained based on “biased dataset” [ 15 ]. To overcome the suboptimal performance of AI-driven algorithms applied to populations different from the training dataset, many researchers diversified their training datasets to be more inclusive and promote health care equity and/or used other technical solutions of generalizability [ 142 ]. Low-income and middle-income countries do not have enough technological infrastructure and trained health care providers to access and implement AI-driven technologies [ 143 ]. To minimize the AI gap, Khan and colleagues proposed leapfrogging and absorptive strategies for low-income countries with and without strong foundations respectively [ 144 ]. AI-leading economies and international organizations such as UNESCO could also help low-income countries catchup with the latest AI technologies through technological cooperation grants, technology transfer, and technical assistance [ 144 ].

The so-called “black box” process of AI-driven models also restrained their clinical adoption because it is not always known why these algorithms give a particular prediction, classification or recommendation [ 15 ]. The “black box” process of AI-driven model is even more critical when patient-centered AI-driven models goes wrong and the accountability of the physician versus the AI-model developer is questioned. Explainable AI techniques such as Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) emerged to clarify how input features contribute to model prediction [ 145 ].

The promising results of most AI-assisted tools have been demonstrated in preclinical studies, retrospective studies and in relatively small sized clinical trials [ 79 , 80 , 94 , 110 ]. Hence, more large-sized multicentre prospective randomized clinical trials are urged to evaluate the safety and efficacy of AI-assisted tools in precision oncology.

Under-fitting or over-fitting are among the problems of AI-powered models [ 15 ]. Under-fitting occurs when the algorithm does not perform well on both the training dataset and validation or unseen data [ 146 ]. This could be due to inadequate training or absence of relevant predictive features. Conversely, over-fitting takes place when an algorithm performs well on training data but not on unseen or external data [ 146 ]. This happens when the model learns irrelevant relationships between the inputs (e.g. patient variables) and the outcomes or when the model predicts inaccurate future outcomes using unsuitable features (for instance when there are many variables compared to the outcomes) [ 15 ]. Validation of AI-driven models could be carried out using external datasets and large-size prospective trials involving national and international organizations [ 15 , 140 ]. Thus, AI-driven training algorithms should be validated in different sites to enhance their generalizability.

AI-driven platforms are not formulated to replace clinicians but rather to spare their efforts and time to do more sophisticated human intellectual tasks which cannot be carried out by machines. Thus, developers of AI-driven algorithms should take into consideration that these models would be ultimately integrated into the workflow of medical oncologists; the majority of which may have limited programming skills [ 15 ]. Thus, the user interface of these tools should be user friendly, time-saving and easily interpretable.

Implementing AI-driven tools in clinical practice mandates systematic coordination between several stakeholders including governments, public sector agencies, AI-developing industries, researchers, and implementers. The World Health Organization (WHO) has published a comprehensive guidance on the ethics and governance of AI for health [ 147 ]. This guidance presented six principles: i) protecting autonomy; ii) advocating human well-being and safety; iii) promoting transparency, explainability and intelligibility; iv) promoting responsibility and accountability; (v) fostering inclusiveness and equity; and vi) ensuring the responsiveness and sustainability of AI [ 147 ].

Approximately 1000 AI-powered medical devices have been authorized by the US FDA which continue to receive hundreds of regulatory submissions for small molecules which were discovered and developed using AI-powered tools [ 148 ]. The US FDA released a guidance entitled “Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products Guidance for Industry and Other Interested Parties” [ 149 ]. In this draft, the FDA presented a risk-based credibility assessment framework that could be exploited for assessing the credibility of AI-driven models. This guidance reviewed the deployment of AI-powered algorithms in the drug product life cycle to support regulatory decision-making linked to the safety or effectiveness of drugs [ 149 ]. This includes using AI-powered models in: i) minimizing the number of in vivo pharmacodynamic, pharmacokinetic and toxicological studies, ii) predicting the clinical pharmacokinetics and driving exposure-response analyses, iii) multimodal data integration from distinct sources, iv) defining and assessing post-marketing adverse drug experience information, and v) identifying optimum manufacturing conditions. It is worth noting that the US FDA underscored the importance of maintaining the credibility of the AI-driven models because their performance can vary over time [ 148 ]. Kore and colleagues shed light on the significance of monitoring the clinically deployed AI-driven models for data drift [ 142 ]. The term “data drift” indicates systematic alteration in the distribution of the input features (e.g. changes in the demographics of a population following immigration) which could compromise the performance of the deployed AI-driven model [ 142 ]. Data drifts were affected by the sample size and patients features [ 142 ]. Monitoring deployed AI models for data drifts could help act before the safety of the patient is compromised and decide whether the existing model should be re-assessed, re-trained, suspended or replaced [ 142 ].

The UK National Health Service (NHS) also developed the ‘Code of Conduct for Data-Driven Health and Care Technologies’ in which they provided principles linked to data ethics framework, usability and accessibility, technical assurance, clinical safety, data protection and transperancy, cybersecurity, regulatory requirements, interoperability and open standards [ 150 ].

To sum, our expectations regarding the potential of AI-driven tools in precision oncology should be realistic. AI is dynamic in nature with more precise and deeper models being developed overtime. Thus, adoption of AI-powered algorithms in clinical practice merits more time and solid evidence proving their safety and efficacy.

Conclusions and future directions

“One-size-fits-all” therapies failed to induce complete remission in a significant proportion of the treated cancer patients. Hence, precision and personalized medicine is foreseen to revolutionize cancer therapy. Precision oncology necessitates sophisticated analyses. AI-driven tools have the potential to fill the gap by enabling integrative analysis of massive data including that of multi-omics data, clinical metadata, radiomics features and chemical features of anticancer drugs. The multi-spectrum potential of AI-driven tools in optimizing almost all stages of the preclinical cancer research starting from discovery of druggable cancer targets, target pocket prediction, structure-based and ligand-based virtual screening, hit discovery, lead optimization, generation of novel small molecules as well as scaffold hopping and prediction of cancer drug response. There are many small molecules which demonstrated potent anticancer activity yet still have unknown direct target(s) [ 104 , 151 , 152 ]. The process of identification of the precise target(s) is sophisticated, time-consuming and costly [ 153 ]. By multimodal analysis of drug response data, drug modeling data and cancer cell modeling data, AI-driven models could help predict the potential target(s) of the small molecules. Within the clinical oncology context, AI-based frameworks integrating patient’s medical record, image features and multi-omics data promoted the identification of predictive, prognostic, susceptibility and diagnostic biomarkers, supported clinical trial matching, guided informed treatment decision making and provided supportive care for cancer patients. However, large-size multicenter prospective RCTs are needed to confirm the potential added value of AI-powered tools in managing diverse types of cancer, especially rare cancers and across all ethnicities. Other limitations include data security and privacy, potential bias in the data used, model accuracy, unfair accessibility to precision and personalized medicine, need for advanced programming skills, difficulty of interpreting AI-driven data, and need for clinical real-world validation. Indeed, consensus regulatory standards and ethical guidelines for using AI-driven models are warranted to prevent data leakage and ensure responsible and effective use of AI-powered tools to benefit all cancer patients rather than the privileged subset only. Notably, more “user-friendly” and “publicly available” AI-assisted frameworks covering their different applications are needed to reduce the technical and financial barriers and ultimately help scientists, medicinal chemists, pharmacologists and medical oncologists to unleash the full potential of AI-assisted algorithms in preclinical cancer research and in clinical practice. This would be of paramount importance especially for research institutions and healthcare systems in low-income and middle-income countries with limited resources. Thus, we are currently at the tip of the iceberg of “AI-powered precision cancer medicine” and more systematic efforts are needed to fully exploit its capabilities.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Change history

12/30/2025

This article is updated to correct the error in author name.

Acknowledgements

None.

Author contributions

RYG and AKA: Investigation, Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing.

Funding

Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). This research was funded by Ain Shams University Strategic Research Grant, Cairo, Egypt which was awarded to AKA.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Abbreviations

Area above the dose response curve

Abelson murine leukemia viral oncogene homolog 1

Antibody-drug conjugate

Artificial intelligence

Artificial intelligence agent for high-optimization and precision medicine

Acute myeloid leukemia

Artificial neural network

Area under the curve

Aurora Kinase A

Breakpoint cluster region

Voltage-dependent calcium channel subunit alpha-2/delta-1

Chimeric antigen receptor

Cancer Cell Line Encyclopedia

Cancer Drug Response profile scan

Chronic myeloid leukemia

Convolutional neural network

Clinical pharmacogenetics implementation consortium

Clustered Regularly Interspaced Short Palindromic Repeats

Cytotoxic T-lymphocyte-associated protein 4

Cancer Therapeutics Response Portal

Deep learning

DNA methyl transferase

Deep neural network

Deep neural network Integrating Prior Knowledge

Epstein-Barr virus

Epidermal growth factor receptor

Food and Drug Administration

FMS-like tyrosine kinase 3

Glucose-6-phosphate dehydrogenase

Human epidermal growth factor receptor 2

Immune checkpoint inhibitor

Knowledge model for potential interaction and potential consideration screening

Loss of function

Mitogen activated protein kinase

Monoclonal Gammopathy of Undetermined Significance

Machine learning

Multiple myeloma

Magnetic resonance imaging

Next generation sequencing

Non-small cell lung cancer

Nudix hydrolase 15

Programmed cell death protein 1

Programmed cell death ligand 1

Prostate-specific antigen

Randomized controlled trial

Residual threshold deep matrix factorization

Structural variants

The Cancer Genome Atlas

Tyrosine kinase inhibitor

Thiopurine methyl transferase

Vascular endothelial growth factor

References

1. Patel JN Cancer pharmacogenomics, challenges in implementation, and patient-focused perspectives Pharmgenomics Pers Med 2016 9 65 77 27471406 10.2147/PGPM.S62918 PMC4948716 Patel JN. Cancer pharmacogenomics, challenges in implementation, and patient-focused perspectives. Pharmgenomics Pers Med. 2016;9:65–77. 27471406 10.2147/PGPM.S62918 PMC4948716

2. Dewaker V Morya VK Kim YH Park ST Kim HS Koh YH Revolutionizing oncology: the role of artificial intelligence (AI) as an antibody design, and optimization tools Biomark Res 2025 13 52 10.1186/s40364-025-00764-4 40155973 PMC11954232 Dewaker V, Morya VK, Kim YH, Park ST, Kim HS, Koh YH. Revolutionizing oncology: the role of artificial intelligence (AI) as an antibody design, and optimization tools. Biomark Res. 2025;13:52. 40155973 10.1186/s40364-025-00764-4 PMC11954232

3. Edsjö A Holmquist L Geoerger B Nowak F Gomon G Alix-Panabières C Precision cancer medicine: concepts, current practice, and future developments J Intern Med 2023 294 455 81 10.1111/joim.13709 37641393 Edsjö A, Holmquist L, Geoerger B, Nowak F, Gomon G, Alix-Panabières C, et al. Precision cancer medicine: concepts, current practice, and future developments. J Intern Med. 2023;294:455–81. 37641393 10.1111/joim.13709

4. Malani D Kumar A Brück O Kontro M Yadav B Hellesøy M Implementing a functional precision medicine tumor board for acute myeloid leukemia Cancer Discov 2022 12 388 401 10.1158/2159-8290.CD-21-0410 34789538 PMC9762335 Malani D, Kumar A, Brück O, Kontro M, Yadav B, Hellesøy M, et al. Implementing a functional precision medicine tumor board for acute myeloid leukemia. Cancer Discov. 2022;12:388–401. 34789538 10.1158/2159-8290.CD-21-0410 PMC9762335

5. Lipinski KA Barber LJ Davies MN Ashenden M Sottoriva A Gerlinger M Cancer Evol And The Limits Of Predictability In Precis Cancer Med Trends In Cancer 2016 2 49 63 10.1016/j.trecan.2015.11.003 PMC4756277 26949746 Lipinski KA, Barber LJ, Davies MN, Ashenden M, Sottoriva A, Gerlinger M. Cancer evolution and the limits of predictability in precision cancer medicine. Trends Cancer. 2016;2:49–63. 10.1016/j.trecan.2015.11.003 PMC4756277 26949746

6. Rulten SL Grose RP Gatz SA Jones JL Cameron AJM The future of precision oncology Int J Mol Sci 2023 24 12613 10.3390/ijms241612613 37628794 PMC10454858 Rulten SL, Grose RP, Gatz SA, Jones JL, Cameron AJM. The future of precision oncology. Int J Mol Sci. 2023;24:12613. 37628794 10.3390/ijms241612613 PMC10454858

7. Doroshow DB Doroshow JH From the broad phase II trial to precision oncology: a perspective on the origins of basket and umbrella clinical trial designs in cancer drug development Cancer J 2019 25 245 53 10.1097/PPO.0000000000000386 31335388 PMC6658138 Doroshow DB, Doroshow JH. From the broad phase II trial to precision oncology: a perspective on the origins of basket and umbrella clinical trial designs in cancer drug development. Cancer J. 2019;25:245–53. 31335388 10.1097/PPO.0000000000000386 PMC6658138

8. Tran KA Kondrashova O Bradley A Williams ED Pearson J V Waddell N Deep learning in cancer diagnosis, prognosis and treatment selection Genome Med 2021 13 152 10.1186/s13073-021-00968-x 34579788 PMC8477474 Tran KA, Kondrashova O, Bradley A, Williams ED, Pearson J V, Waddell N. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med. 2021;13:152. 34579788 10.1186/s13073-021-00968-x PMC8477474

9. Athanasopoulou K, Michalopoulou S VI, Adamopoulos A, PG. Integrating artificial intelligence in next-generation sequencing: advances, challenges, and future directions. Curr Issues Mol Biol. 2025;47. 10.3390/cimb47060470 PMC12191491 40699869

10. Dlamini Z Francies FZ Hull R Marima R Artificial intelligence (AI) and big data in cancer and precision oncology Comput Struct Biotechnol J 2020 18 2300 11 10.1016/j.csbj.2020.08.019 32994889 PMC7490765 Dlamini Z, Francies FZ, Hull R, Marima R. Artificial intelligence (AI) and big data in cancer and precision oncology. Comput Struct Biotechnol J. 2020;18:2300–11. 32994889 10.1016/j.csbj.2020.08.019 PMC7490765

11. MacEachern SJ Forkert ND Machine learning for precision medicine Genome 2021 64 416 25 10.1139/gen-2020-0131 33091314 MacEachern SJ, Forkert ND. Machine learning for precision medicine. Genome. 2021;64:416–25. 33091314 10.1139/gen-2020-0131

12. Vyas A Kumar K Sharma A Verma D Bhatia D Wahi N Advancing the frontier of artificial intelligence on emerging technologies to redefine cancer diagnosis and care Comput Biol Med 2025 191 110178 10.1016/j.compbiomed.2025.110178 40228444 Vyas A, Kumar K, Sharma A, Verma D, Bhatia D, Wahi N, et al. Advancing the frontier of artificial intelligence on emerging technologies to redefine cancer diagnosis and care. Comput Biol Med. 2025;191:110178. 40228444 10.1016/j.compbiomed.2025.110178

13. Gupta R, Srivastava D, Sahu M, Tiwari S, Ambasta RK, Kumar P. Artificial intelligence to deep learning: machine intelligence approach for drug discovery. Mol Divers. 2021. 10.1007/s11030-021-10217-3 PMC8040371 33844136

14. Choi RY Coyner AS Kalpathy-Cramer J Chiang MF Peter Campbell J Introduction to machine learning, neural networks, and deep learning Transl Vis Sci Technol 2020 9 1 12 10.1167/tvst.9.2.14 PMC7347027 32704420 Choi RY, Coyner AS, Kalpathy-Cramer J, Chiang MF, Peter Campbell J. Introduction to machine learning, neural networks, and deep learning. Transl Vis Sci Technol. 2020;9:1–12. 10.1167/tvst.9.2.14 PMC7347027 32704420

15. Aung YYM Wong DCS Ting DSW The promise of artificial intelligence: a review of the opportunities and challenges of artificial intelligence in healthcare Br Med Bull 2021 139 4 15 10.1093/bmb/ldab016 34405854 Aung YYM, Wong DCS, Ting DSW. The promise of artificial intelligence: a review of the opportunities and challenges of artificial intelligence in healthcare. Br Med Bull. 2021;139:4–15. 34405854 10.1093/bmb/ldab016

16. Kufel J, Bargieł-Łączek K, Kocot S, Koźlik M, Bartnikowska W, Janik M, et al. What is machine learning, artificial neural networks and deep learning?-examples of practical applications in medicine. Diagnostics. 2023;13. 10.3390/diagnostics13152582 PMC10417718 37568945

17. Bhalla S, Laganà A. Artificial intelligence for precision oncology. 2022;249–68. 10.1007/978-3-030-91836-1_14 35230693

18. Fiskus W Mill CP Birdwell C Davis JA Das K Boettcher S Targeting of epigenetic co-dependencies enhances anti-AML efficacy of Menin inhibitor in AML with MLL1-r or mutant NPM1 Blood Cancer J 2023 13 53 10.1038/s41408-023-00826-6 37055414 PMC10102188 Fiskus W, Mill CP, Birdwell C, Davis JA, Das K, Boettcher S, et al. Targeting of epigenetic co-dependencies enhances anti-AML efficacy of Menin inhibitor in AML with MLL1-r or mutant NPM1. Blood Cancer J. 2023;13:53. 37055414 10.1038/s41408-023-00826-6 PMC10102188

19. Abdel-Aziz AK Advances in acute myeloid leukemia differentiation therapy: a critical review Biochem Pharmacol 2023 215 115709 10.1016/j.bcp.2023.115709 37506924 Abdel-Aziz AK. Advances in acute myeloid leukemia differentiation therapy: a critical review. Biochem Pharmacol. 2023;215:115709. 37506924 10.1016/j.bcp.2023.115709

20. Broxterman HJ Gotink KJ Verheul HMW Understanding the causes of multidrug resistance in cancer: a comparison of doxorubicin and sunitinib Drug Resist Updat 2009 12 114 26 10.1016/j.drup.2009.07.001 19648052 Broxterman HJ, Gotink KJ, Verheul HMW. Understanding the causes of multidrug resistance in cancer: a comparison of doxorubicin and sunitinib. Drug Resist Updat. 2009;12:114–26. 19648052 10.1016/j.drup.2009.07.001

21. Wu S-G Liu Y-N Tsai M-F Chang Y-L Yu C-J Yang P-C The mechanism of acquired resistance to irreversible EGFR tyrosine kinase inhibitor-afatinib in lung adenocarcinoma patients Oncotarget 2016 7 12404 13 10.18632/oncotarget.7189 26862733 PMC4914294 Wu S-G, Liu Y-N, Tsai M-F, Chang Y-L, Yu C-J, Yang P-C, et al. The mechanism of acquired resistance to irreversible EGFR tyrosine kinase inhibitor-afatinib in lung adenocarcinoma patients. Oncotarget. 2016;7:12404–13. 26862733 10.18632/oncotarget.7189 PMC4914294

22. Tcgar N Genomic and epigenomic landscapes of adult De novo acute myeloid leukemia N Engl J Med 2013 368 2059 74 10.1056/NEJMoa1301689 23634996 PMC3767041 Tcgar N. Genomic and epigenomic landscapes of adult De novo acute myeloid leukemia. N Engl J Med. 2013;368:2059–74. 23634996 10.1056/NEJMoa1301689 PMC3767041

23. Moretti S Abdel-Aziz AK Ceccacci E Pallavicini I Santoro F de Thé H Co-targeting leukemia-initiating cells and leukemia bulk leads to disease eradication Leukemia 2022 36 1306 12 10.1038/s41375-022-01530-3 35246604 Moretti S, Abdel-Aziz AK, Ceccacci E, Pallavicini I, Santoro F, de Thé H, et al. Co-targeting leukemia-initiating cells and leukemia bulk leads to disease eradication. Leukemia. 2022;36:1306–12. 35246604 10.1038/s41375-022-01530-3

24. Abdel-Aziz AK Dokla EME Saadeldin MK FLT3 inhibitors and novel therapeutic strategies to reverse AML resistance: an updated comprehensive review Crit Rev Oncol Hematol 2023 191 104139 10.1016/j.critrevonc.2023.104139 37717880 Abdel-Aziz AK, Dokla EME, Saadeldin MK. FLT3 inhibitors and novel therapeutic strategies to reverse AML resistance: an updated comprehensive review. Crit Rev Oncol Hematol. 2023;191:104139. 37717880 10.1016/j.critrevonc.2023.104139

25. Tecik M Adan A Therapeutic targeting of FLT3 in acute myeloid leukemia: current status and novel approaches Onco Targets Ther 2022 15 1449 78 10.2147/OTT.S384293 36474506 PMC9719701 Tecik M, Adan A. Therapeutic targeting of FLT3 in acute myeloid leukemia: current status and novel approaches. Onco Targets Ther. 2022;15:1449–78. 36474506 10.2147/OTT.S384293 PMC9719701

26. Kumai T Shinomiya H Shibata H Takahashi H Kishikawa T Okada R Translational research in head and neck cancer: molecular and immunological updates Auris Nasus Larynx 2024 51 391 400 10.1016/j.anl.2023.08.006 37640594 Kumai T, Shinomiya H, Shibata H, Takahashi H, Kishikawa T, Okada R, et al. Translational research in head and neck cancer: molecular and immunological updates. Auris Nasus Larynx. 2024;51:391–400. 37640594 10.1016/j.anl.2023.08.006

27. Kamath S Buolamwini JK Targeting EGFR and HER-2 receptor tyrosine kinases for cancer drug discovery and development Med Res Rev 2006 26 569 94 10.1002/med.20070 16788977 Kamath S, Buolamwini JK. Targeting EGFR and HER-2 receptor tyrosine kinases for cancer drug discovery and development. Med Res Rev. 2006;26:569–94. 16788977 10.1002/med.20070

28. Chien JH Tang JL Chen RL Li CC Lee CP Detection of BCR-ABL gene mutations in Philadelphia chromosome positive leukemia patients resistant to STI-571 cancer therapy Leuk Res 2008 32 1724 34 10.1016/j.leukres.2008.04.023 18603297 Chien JH, Tang JL, Chen RL, Li CC, Lee CP. Detection of BCR-ABL gene mutations in Philadelphia chromosome positive leukemia patients resistant to STI-571 cancer therapy. Leuk Res. 2008;32:1724–34. 18603297 10.1016/j.leukres.2008.04.023

29. Abdel-Aziz AK, Azab SSE, Youssef SS, El-Sayed AM, El-Demerdash E, Shouman S. Modulation of imatinib cytotoxicity by selenite in HCT116 colorectal cancer cells. Basic Clin Pharmacol Toxicol. 2015;116. 10.1111/bcpt.12281 24930392

30. Kitagawa D Yokota K Gouda M Narumi Y Ohmoto H Nishiwaki E Activity-based kinase profiling of approved tyrosine kinase inhibitors Genes Cells 2013 18 110 22 10.1111/gtc.12022 23279183 Kitagawa D, Yokota K, Gouda M, Narumi Y, Ohmoto H, Nishiwaki E, et al. Activity-based kinase profiling of approved tyrosine kinase inhibitors. Genes Cells. 2013;18:110–22. 23279183 10.1111/gtc.12022

31. Yang J Ikezoe T Nishioka C Udaka K Yokoyama A Bcr-Abl activates AURKA and AURKB in chronic myeloid leukemia cells via AKT signaling Int. J. Cancer 2014 134 1183 94 10.1002/ijc.28434 23934627 Yang J, Ikezoe T, Nishioka C, Udaka K, Yokoyama A. Bcr-Abl activates AURKA and AURKB in chronic myeloid leukemia cells via AKT signaling. Int J Cance. 2014;134:1183–94. 23934627 10.1002/ijc.28434

32. Burchert A Wang Y Cai D von Bubnoff N Paschka P Müller-Brüsselbach S Compensatory PI3-kinase/akt/mtor activation regulates imatinib resistance development Leukemia 2005 19 1774 82 10.1038/sj.leu.2403898 16136169 Burchert A, Wang Y, Cai D, von Bubnoff N, Paschka P, Müller-Brüsselbach S, et al. Compensatory PI3-kinase/akt/mtor activation regulates imatinib resistance development. Leukemia. 2005;19:1774–82. 16136169 10.1038/sj.leu.2403898

33. Dokla EME Abdel-Aziz AK Milik SN McPhillie MJ Minucci S Abouzid KAM Discovery of a benzimidazole-based dual FLT3/TrKA inhibitor targeting acute myeloid leukemia Bioorg Med Chem 2022 56 116596 10.1016/j.bmc.2021.116596 35033885 Dokla EME, Abdel-Aziz AK, Milik SN, McPhillie MJ, Minucci S, Abouzid KAM. Discovery of a benzimidazole-based dual FLT3/TrKA inhibitor targeting acute myeloid leukemia. Bioorg Med Chem. 2022;56:116596. 35033885 10.1016/j.bmc.2021.116596

34. Milik SN Abdel-Aziz AK El-Hendawy MM El-Gogary RI Saadeldin MK Minucci S Insights into the design of inhibitors of the EGFR family with anticancer activity overcoming resistance: a case of optimizing thieno[2, 3-d]pyrimidine-based EGFR inhibitors J Mol Struct 2022 1259 132724 10.1016/j.molstruc.2022.132724 Milik SN, Abdel-Aziz AK, El-Hendawy MM, El-Gogary RI, Saadeldin MK, Minucci S, et al. Insights into the design of inhibitors of the EGFR family with anticancer activity overcoming resistance: a case of optimizing thieno[2, 3-d]pyrimidine-based EGFR inhibitors. J Mol Struct. 2022;1259:132724.

35. Milik SN Abdel-Aziz AK Lasheen DS Serya RAT Minucci S Abouzid KAM Surmounting the resistance against EGFR inhibitors through the development of thieno[2, 3-d]pyrimidine-based dual EGFR/HER2 inhibitors Eur J Med Chem 2018 155 316 36 10.1016/j.ejmech.2018.06.011 29902719 Milik SN, Abdel-Aziz AK, Lasheen DS, Serya RAT, Minucci S, Abouzid KAM. Surmounting the resistance against EGFR inhibitors through the development of thieno[2, 3-d]pyrimidine-based dual EGFR/HER2 inhibitors. Eur J Med Chem. 2018;155:316–36. 29902719 10.1016/j.ejmech.2018.06.011

36. Mariam Raju R Joy A J Nulgumnalli Manjunathaiah R Justin A Prashantha Kumar BR EGFR as therapeutic target to develop new generation tyrosine kinase inhibitors against breast cancer: a critical review Results Chem 2024 7 101490 10.1016/j.rechem.2024.101490 Mariam Raju R, Joy A J, Nulgumnalli Manjunathaiah R, Justin A, Prashantha Kumar BR. EGFR as therapeutic target to develop new generation tyrosine kinase inhibitors against breast cancer: a critical review. Results Chem. 2024;7:101490.

37. Chu TF a RM Kerkela R Dallabrida SM Zurakowski D Nguyen L Cardiotoxicity associated with tyrosine kinase inhibitor sunitinib Lancet 2007 370 2011 19 10.1016/S0140-6736(07)61865-0 18083403 PMC2643085 Chu TF, a RM, Kerkela R, Dallabrida SM, Zurakowski D, Nguyen L, et al. Cardiotoxicity associated with tyrosine kinase inhibitor sunitinib. Lancet. 2007;370:2011–19. 18083403 10.1016/S0140-6736(07)61865-0 PMC2643085

38. Abdel-Aziz AK Mantawy EM Said RS Helwa R The tyrosine kinase inhibitor, sunitinib malate, induces cognitive impairment in vivo via dysregulating VEGFR signaling, apoptotic and autophagic machineries Exp Neurol 2016 283 129 41 10.1016/j.expneurol.2016.06.004 27288242 Abdel-Aziz AK, Mantawy EM, Said RS, Helwa R. The tyrosine kinase inhibitor, sunitinib malate, induces cognitive impairment in vivo via dysregulating VEGFR signaling, apoptotic and autophagic machineries. Exp Neurol. 2016;283:129–41. 27288242 10.1016/j.expneurol.2016.06.004

39. Joshi SK Nechiporuk T Bottomly D Piehowski PD Reisz JA Pittsenbarger J The AML microenvironment catalyzes a stepwise evolution to gilteritinib resistance Cancer Cell 2021 39 999 1014.e8 10.1016/j.ccell.2021.06.003 34171263 PMC8686208 Joshi SK, Nechiporuk T, Bottomly D, Piehowski PD, Reisz JA, Pittsenbarger J, et al. The AML microenvironment catalyzes a stepwise evolution to gilteritinib resistance. Cancer Cell. 2021;39:999–1014.e8. 34171263 10.1016/j.ccell.2021.06.003 PMC8686208

40. Khadela A Shah Y Mistry P Mansuri M Sureja D Bodiwala K A review of efficacy and safety of cetuximab and bevacizumab-based monoclonal antibodies in head and neck cancer Med Oncol 2023 40 1 13 10.1007/s12032-022-01939-x 36583766 Khadela A, Shah Y, Mistry P, Mansuri M, Sureja D, Bodiwala K. A review of efficacy and safety of cetuximab and bevacizumab-based monoclonal antibodies in head and neck cancer. Med Oncol. 2023;40:1–13. 10.1007/s12032-022-01939-x 36583766

41. Elgendy M Abdel-Aziz AK Renne SL Bornaghi V Procopio G Colecchia M Dual modulation of MCL-1 and mTOR determines the response to sunitinib J. Clin. Invest 2017 127 153 68 10.1172/JCI84386 27893461 PMC5199697 Elgendy M, Abdel-Aziz AK, Renne SL, Bornaghi V, Procopio G, Colecchia M, et al. Dual modulation of MCL-1 and mTOR determines the response to sunitinib. J. Clin. Invest. 2017;127:153–68. 27893461 10.1172/JCI84386 PMC5199697

42. Abdel-Aziz AK Abdel-Naim AB Shouman S Minucci S From resistance to sensitivity: insights and implications of biphasic modulation of autophagy by sunitinib Front Pharmacol 2017 8 1 10 10.3389/fphar.2017.00718 29066973 PMC5641351 Abdel-Aziz AK, Abdel-Naim AB, Shouman S, Minucci S. From resistance to sensitivity: insights and implications of biphasic modulation of autophagy by sunitinib. Front Pharmacol. 2017;8:1–10. 29066973 10.3389/fphar.2017.00718 PMC5641351

43. Yarden Y Sliwkowski MX Untangling the ErbB signalling network Nat Rev Mol Cell Biol 2001 2 127 37 10.1038/35052073 11252954 Yarden Y, Sliwkowski MX. Untangling the ErbB signalling network. Nat Rev Mol Cell Biol. 2001;2:127–37. 11252954 10.1038/35052073

44. Abdel-Aziz AK Saadeldin MK D’Amico P Orecchioni S Bertolini F Curigliano G Preclinical models of breast cancer: two-way shuttles for immune checkpoint inhibitors from and to patient bedside Eur J Cancer 2019 122 22 41 10.1016/j.ejca.2019.08.013 31606656 Abdel-Aziz AK, Saadeldin MK, D’Amico P, Orecchioni S, Bertolini F, Curigliano G, et al. Preclinical models of breast cancer: two-way shuttles for immune checkpoint inhibitors from and to patient bedside. Eur J Cancer. 2019;122:22–41. 31606656 10.1016/j.ejca.2019.08.013

45. Nanda R Chow LQM Dees EC Berger R Gupta S Geva R Pembrolizumab in patients with advanced triple-negative breast cancer: phase ib KEYNOTE-012 study J Clin Oncol 2019 34 2460 67 10.1200/JCO.2015.64.8931 PMC6816000 27138582 Nanda R, Chow LQM, Dees EC, Berger R, Gupta S, Geva R, et al. Pembrolizumab in patients with advanced triple-negative breast cancer: phase ib KEYNOTE-012 study. J Clin Oncol. 2019;34:2460–67. 10.1200/JCO.2015.64.8931 PMC6816000 27138582

46. Shi H Zhang W Zhi Q Jiang M Lapatinib resistance in HER2+ cancers: latest findings and new concepts on molecular mechanisms Tumor Biol 2016 37 15411 31 10.1007/s13277-016-5467-2 27726101 Shi H, Zhang W, Zhi Q, Jiang M. Lapatinib resistance in HER2+ cancers: latest findings and new concepts on molecular mechanisms. Tumor Biol. 2016;37:15411–31. 10.1007/s13277-016-5467-2 27726101

47. Hollander N. Immunotherapy for B-cell lymphoma: current status and prospective advances. 2012;3:1–15. 10.3389/fimmu.2012.00003 PMC3342070 22566889

48. Rehmani HS Issaeva N EGFR in head and neck squamous cell carcinoma: exploring possibilities of novel drug combinations Ann Transl Med 2020 8 813 813 10.21037/atm.2020.04.07 32793658 PMC7396252 Rehmani HS, Issaeva N. EGFR in head and neck squamous cell carcinoma: exploring possibilities of novel drug combinations. Ann Transl Med. 2020;8:813–813. 32793658 10.21037/atm.2020.04.07 PMC7396252

49. Zhu K Yang X Tai H Zhong X Luo T Zheng H HER2-targeted therapies in cancer: a systematic review Biomark Res 2024 12 16 10.1186/s40364-024-00565-1 38308374 PMC10835834 Zhu K, Yang X, Tai H, Zhong X, Luo T, Zheng H. HER2-targeted therapies in cancer: a systematic review. Biomark Res. 2024;12:16. 38308374 10.1186/s40364-024-00565-1 PMC10835834

50. Hu Y-L DeLay M Jahangiri A Molinaro AM Rose SD Carbonell WS Hypoxia-induced autophagy promotes tumor cell survival and adaptation to antiangiogenic treatment in glioblastoma Cancer Res 2012 72 1773 83 10.1158/0008-5472.CAN-11-3831 22447568 PMC3319869 Hu Y-L, DeLay M, Jahangiri A, Molinaro AM, Rose SD, Carbonell WS, et al. Hypoxia-induced autophagy promotes tumor cell survival and adaptation to antiangiogenic treatment in glioblastoma. Cancer Res. 2012;72:1773–83. 22447568 10.1158/0008-5472.CAN-11-3831 PMC3319869

51. Wang GX Justin F Sullivan RJ Flaherty KT Lee SI Immune checkpoint inhibitor cancer therapy: spectrum of imaging findings Radiographics 2017 37 2132 44 10.1148/rg.2017170085 29131763 Wang GX, Justin F, Sullivan RJ, Flaherty KT, Lee SI. Immune checkpoint inhibitor cancer therapy: spectrum of imaging findings. Radiographics. 2017;37:2132–44. 29131763 10.1148/rg.2017170085

52. Oiseth SJ, Aziz MS. Cancer immunotherapy: a brief review of the history, possibilities, and challenges ahead. 2017.

53. Nixon MJ, Balko JM. Biomarkers for assessing the effectiveness of immunotherapy in breast cancer. 2018;12:97–100. 10.2217/bmm-2017-0408 29338306

54. Hu-Lieskovan S Malouf GG Jacobs I Chou J Liu L Melissa L Addressing resistance to immune checkpoint inhibitor therapy: an urgent unmet need Future Oncol 2021 17 1401 39 10.2217/fon-2020-0967 33475012 Hu-Lieskovan S, Malouf GG, Jacobs I, Chou J, Liu L, Melissa L. Addressing resistance to immune checkpoint inhibitor therapy: an urgent unmet need. Future Oncol. 2021;17:1401–39. 33475012 10.2217/fon-2020-0967

55. Jenkins RW Barbie DA Flaherty KT Mechanisms of resistance to immune checkpoint inhibitors Br J Cancer 2018 118 9 16 10.1038/bjc.2017.434 29319049 PMC5765236 Jenkins RW, Barbie DA, Flaherty KT. Mechanisms of resistance to immune checkpoint inhibitors. Br J Cancer. 2018;118:9–16. 29319049 10.1038/bjc.2017.434 PMC5765236

56. Lewis GD, Li G, Guo J, Yu SF, Fields CT, Lee G, et al. The HER2-directed antibody-drug conjugate DHES0815A in advanced and/or metastatic breast cancer: preclinical characterization and phase 1 trial results. Nat Commun. 2024;15. 10.1038/s41467-023-44533-z PMC10784567 38212321

57. Zagozdzon R Gallagher WM Crown J Truncated HER2: implications for HER2-targeted therapeutics Drug Discov Today 2011 16 810 16 10.1016/j.drudis.2011.06.003 21704182 Zagozdzon R, Gallagher WM, Crown J. Truncated HER2: implications for HER2-targeted therapeutics. Drug Discov Today. 2011;16:810–16. 21704182 10.1016/j.drudis.2011.06.003

58. Cortés J Kim S-B Chung W-P Im S-A Park YH Hegg R Trastuzumab Deruxtecan versus Trastuzumab emtansine for breast cancer N Engl J Med 2022 386 1143 54 10.1056/NEJMoa2115022 35320644 Cortés J, Kim S-B, Chung W-P, Im S-A, Park YH, Hegg R, et al. Trastuzumab Deruxtecan versus Trastuzumab emtansine for breast cancer. N Engl J Med. 2022;386:1143–54. 35320644 10.1056/NEJMoa2115022

59. Carr DF Turner RM Pirmohamed M Pharmacogenomics of anticancer drugs: personalising the choice and dose to manage drug response Br J Clin Pharmacol 2021 87 237 55 10.1111/bcp.14407 32501544 Carr DF, Turner RM, Pirmohamed M. Pharmacogenomics of anticancer drugs: personalising the choice and dose to manage drug response. Br J Clin Pharmacol. 2021;87:237–55. 32501544 10.1111/bcp.14407

60. Chiranth CR, Kumkum S, Yellapur SG, Desai VP, Reddy AHM, Manokaran S. A review on Pharmacogenomics of anticancer drugs in targeted therapy. Proc Natl Acad Sci, India, Sect B Biol Sci. 2025.

61. He Z, Fu X, Xu Z, Alshargi O, Rodríguez-Labrada R, Zhong X-B. Editorial: pharmacogenomics for improving drug safety and efficacy in cancer. Front Pharmacol. 2025;16. 10.3389/fphar.2025.1649258 PMC12259549 40667512

62. Narendra G Choudhary S Raju B Verma H Silakari O Role of genetic polymorphisms in drug-metabolizing enzyme-mediated toxicity and pharmacokinetic resistance to anti-cancer agents: a review on the pharmacogenomics aspect Clin Pharmacokinet 2022 61 1495 517 10.1007/s40262-022-01174-7 36180817 Narendra G, Choudhary S, Raju B, Verma H, Silakari O. Role of genetic polymorphisms in drug-metabolizing enzyme-mediated toxicity and pharmacokinetic resistance to anti-cancer agents: a review on the pharmacogenomics aspect. Clin Pharmacokinet. 2022;61:1495–517. 36180817 10.1007/s40262-022-01174-7

63. Hammami MB Qasim A Thakur R Vegivinti CTR Patton CD Vikash S Rasburicase-induced hemolytic anemia and methemoglobinemia: a systematic review of current reports Ann Hematol 2024 103 3399 411 10.1007/s00277-023-05364-6 37468669 Hammami MB, Qasim A, Thakur R, Vegivinti CTR, Patton CD, Vikash S, et al. Rasburicase-induced hemolytic anemia and methemoglobinemia: a systematic review of current reports. Ann Hematol. 2024;103:3399–411. 37468669 10.1007/s00277-023-05364-6

64. Boussaid I Mondesir J Chapuis N Birsen R Methaemoglobinaemia and haemolysis after rasburicase administration in a patient with G6PD deficiency BMJ Case Rep 2022 15 2 3 10.1136/bcr-2022-252180 PMC9535139 36198433 Boussaid I, Mondesir J, Chapuis N, Birsen R. Methaemoglobinaemia and haemolysis after rasburicase administration in a patient with G6PD deficiency. BMJ Case Rep. 2022;15:2–3. 10.1136/bcr-2022-252180 PMC9535139 36198433

65. Relling M V Schwab M Whirl-Carrillo M Suarez-Kurtz G Pui CH Stein CM Clinical pharmacogenetics implementation consortium guideline for thiopurine dosing based on TPMT and NUDT15 genotypes: 2018 update Clin Pharmacol Ther 2019 105 1095 105 10.1002/cpt.1304 30447069 PMC6576267 Relling M V, Schwab M, Whirl-Carrillo M, Suarez-Kurtz G, Pui CH, Stein CM, et al. Clinical pharmacogenetics implementation consortium guideline for thiopurine dosing based on TPMT and NUDT15 genotypes: 2018 update. Clin Pharmacol Ther. 2019;105:1095–105. 30447069 10.1002/cpt.1304 PMC6576267

66. Ramudo-Cela L Busto-Fernández F Outeda-Macías M Antolín S Calvo-Martínez L Martín-Herranz I Results of the implementation of a pharmacogenomics platform based on NGS technologies Combining Clin And Res Approaches Farm Hosp 2021 45 11 37 35379108 Ramudo-Cela L, Busto-Fernández F, Outeda-Macías M, Antolín S, Calvo-Martínez L, Martín-Herranz I. Results of the implementation of a pharmacogenomics platform based on NGS technologies. Combining Clin And Res Approaches Farm Hosp. 2021;45:11–37. 35379108

67. Bianchi V Ceol A Ogier AGE De Pretis S Galeota E Kishore K Integrated systems for NGS data management and analysis: open issues and available solutions Front Genet 2016 7 1 8 10.3389/fgene.2016.00075 27200084 PMC4858535 Bianchi V, Ceol A, Ogier AGE, De Pretis S, Galeota E, Kishore K, et al. Integrated systems for NGS data management and analysis: open issues and available solutions. Front Genet. 2016;7:1–8. 27200084 10.3389/fgene.2016.00075 PMC4858535

68. Haferlach T Schmidts I The power and potential of integrated diagnostics in acute myeloid leukaemia Br J Haematol 2020 188 36 48 10.1111/bjh.16360 31808952 PMC6973013 Haferlach T, Schmidts I. The power and potential of integrated diagnostics in acute myeloid leukaemia. Br J Haematol. 2020;188:36–48. 31808952 10.1111/bjh.16360 PMC6973013

69. Aldea M, Ghigna MR, Lacroix-Triki M, Andre F. Unlocking the potential of AI-assisted pathology for molecular alteration screening. Eur J Cancer Engl. 2024;113467. 10.1016/j.ejca.2023.113467 38103329

70. Wiest IC, Bhat M, Clusmann J, Schneider C V, Jiang X, Kather JN. Large language models for clinical decision support in gastroenterology and hepatology. Nat Rev Gastroenterol Hepatol. 2025. 10.1038/s41575-025-01108-1 40846793

71. Alifiansyah MRT Herdiansyah MA Pratiwi RC Pramesti RP Hafsyah NW Rania AP Qsar of acyl alizarin red Biocompound derivatives of Rubia Tinctorum Roots and its admet properties as anti-breast cancer candidates against mmp-9 protein receptor: in Silico study Food Syst 2024 7 312 20 10.21323/2618-9771-2024-7-2-312-320 Alifiansyah MRT, Herdiansyah MA, Pratiwi RC, Pramesti RP, Hafsyah NW, Rania AP, et al. Qsar of acyl alizarin red Biocompound derivatives of Rubia Tinctorum Roots and its admet properties as anti-breast cancer candidates against mmp-9 protein receptor: in Silico study. Food Syst. 2024;7:312–20.

72. Parikesit AA Ansori ANM Kharisma VD Computational design of siRNA targeting homo sapiens HER2 splice variant mRNA: a potential strategy for breast cancer intervention Biosaintifika 2024 16 384 93 10.15294/biosaintifika.v16i3.3685 Parikesit AA, Ansori ANM, Kharisma VD. Computational design of siRNA targeting homo sapiens HER2 splice variant mRNA: a potential strategy for breast cancer intervention. Biosaintifika. 2024;16:384–93.

73. Parikesit AA Ansori ANM Kharisma VD A computational design of siRNA in SARS-CoV-2 Spike Glycoprotein Gene and its binding capability toward mRNA Indones J Chem 2022 22 1163 76 10.22146/ijc.68415 Parikesit AA, Ansori ANM, Kharisma VD. A computational design of siRNA in SARS-CoV-2 Spike Glycoprotein Gene and its binding capability toward mRNA. Indones J Chem. 2022;22:1163–76.

74. Widyananda MH Pratama SK Samoedra RS Sari FN Kharisma VD Ansori ANM Molecular docking study of sea urchin (Arbacia lixula) peptides as multi-target inhibitor for non-small cell lung cancer (NSCLC) associated proteins J Pharm Pharmacogn Res 2021 9 484 96 10.56499/jppres21.1047_9.4.484 Widyananda MH, Pratama SK, Samoedra RS, Sari FN, Kharisma VD, Ansori ANM, et al. Molecular docking study of sea urchin (Arbacia lixula) peptides as multi-target inhibitor for non-small cell lung cancer (NSCLC) associated proteins. J Pharm Pharmacogn Res. 2021;9:484–96.

75. Johnson SJS. AI revolutionizing cell and genetic engineering: innovations and applications. 2025;219–32. 10.1007/978-1-0716-4690-8_12 40553335

76. Kim J McFee M Fang Q Abdin O Kim PM Computational and artificial intelligence-based methods for antibody development Trends Pharmacol Sci 2023 44 175 89 10.1016/j.tips.2022.12.005 36669976 Kim J, McFee M, Fang Q, Abdin O, Kim PM. Computational and artificial intelligence-based methods for antibody development. Trends Pharmacol Sci. 2023;44:175–89. 36669976 10.1016/j.tips.2022.12.005

77. Jumper J Evans R Pritzel A Green T Figurnov M Ronneberger O Highly accurate protein structure prediction with AlphaFold Nature 2021 596 583 89 10.1038/s41586-021-03819-2 34265844 PMC8371605 Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596:583–89. 34265844 10.1038/s41586-021-03819-2 PMC8371605

78. Bailey MH Tokheim C Porta-Pardo E Sengupta S Bertrand D Weerasinghe A Comprehensive characterization of cancer driver genes and mutations Cell 2018 173 371 85.e18 10.1016/j.cell.2018.02.060 29625053 PMC6029450 Bailey MH, Tokheim C, Porta-Pardo E, Sengupta S, Bertrand D, Weerasinghe A, et al. Comprehensive characterization of cancer driver genes and mutations. Cell. 2018;173:371–85.e18. 29625053 10.1016/j.cell.2018.02.060 PMC6029450

79. Kamya P Ozerov I V Pun FW Tretina K Fokina T Chen S PandaOmics: an AI-Driven platform for therapeutic target and biomarker discovery J Chem Inf Model 2024 64 3961 69 10.1021/acs.jcim.3c01619 38404138 PMC11134400 Kamya P, Ozerov I V, Pun FW, Tretina K, Fokina T, Chen S, et al. PandaOmics: an AI-Driven platform for therapeutic target and biomarker discovery. J Chem Inf Model. 2024;64:3961–69. 38404138 10.1021/acs.jcim.3c01619 PMC11134400

80. Chung D-S Leung GHD Liu S Lok SWY Xin Y Xia Y AI-Driven drug target screening platform identified oncogene CACNA2D1 activated by enhancer infestation in epstein-barr virus-associated nasopharyngeal carcinoma Int J Mol Sci 2025 26 4697 10.3390/ijms26104697 40429844 PMC12111453 Chung D-S, Leung GHD, Liu S, Lok SWY, Xin Y, Xia Y, et al. AI-Driven drug target screening platform identified oncogene CACNA2D1 activated by enhancer infestation in epstein-barr virus-associated nasopharyngeal carcinoma. Int J Mol Sci. 2025;26:4697. 40429844 10.3390/ijms26104697 PMC12111453

81. Di Filippo JI Cavasotto CN Guided structure-based ligand identification and design via artificial intelligence modeling Expert Opin Drug Discov 2022 17 71 78 10.1080/17460441.2021.1979514 34544293 Di Filippo JI, Cavasotto CN. Guided structure-based ligand identification and design via artificial intelligence modeling. Expert Opin Drug Discov. 2022;17:71–78. 34544293 10.1080/17460441.2021.1979514

82. Yang K Xie Z Li Z Qian X Sun N He T MolProphet: a one-stop, general purpose, and AI-based platform for the early stages of drug discovery J Chem Inf Model 2024 64 2941 47 10.1021/acs.jcim.3c01979 38563534 PMC11040716 Yang K, Xie Z, Li Z, Qian X, Sun N, He T, et al. MolProphet: a one-stop, general purpose, and AI-based platform for the early stages of drug discovery. J Chem Inf Model. 2024;64:2941–47. 38563534 10.1021/acs.jcim.3c01979 PMC11040716

83. Pandey AK AI-assisted generation and in-depth in-silico evaluation of potential inhibitor targeting aurora kinase a (AURKA): an anticancer discovery exploiting synthetic lethality approach Arch Biochem Biophys 2024 762 110209 10.1016/j.abb.2024.110209 39491662 Pandey AK. AI-assisted generation and in-depth in-silico evaluation of potential inhibitor targeting aurora kinase a (AURKA): an anticancer discovery exploiting synthetic lethality approach. Arch Biochem Biophys. 2024;762:110209. 39491662 10.1016/j.abb.2024.110209

84. Saadeldin MK Curigliano G Abdel-Aziz AK Deciphering the complex interplay of long noncoding RNAs and Aurora Kinases: novel insights into breast cancer development and therapeutic strategies Futur Pharmacol 2024 4 466 78 10.3390/futurepharmacol4030026 Saadeldin MK, Curigliano G, Abdel-Aziz AK. Deciphering the complex interplay of long noncoding RNAs and Aurora Kinases: novel insights into breast cancer development and therapeutic strategies. Futur Pharmacol. 2024;4:466–78.

85. Dokla EME Abdel-Aziz AK Milik SN Mahmoud AH Kamal M Mcphillie MJ Indolin-2-one derivatives as selective aurora B kinase inhibitors targeting breast cancer Bioorg Chem 2021 117 105451 10.1016/j.bioorg.2021.105451 34736137 Dokla EME, Abdel-Aziz AK, Milik SN, Mahmoud AH, Kamal M, Mcphillie MJ, et al. Indolin-2-one derivatives as selective aurora B kinase inhibitors targeting breast cancer. Bioorg Chem. 2021;117:105451. 34736137 10.1016/j.bioorg.2021.105451

86. Manen-Freixa L Antolin AA Polypharmacology prediction: the long road toward comprehensively anticipating small-molecule selectivity to de-risk drug discovery Expert Opin Drug Discov 2024 19 1043 69 10.1080/17460441.2024.2376643 39004919 Manen-Freixa L, Antolin AA. Polypharmacology prediction: the long road toward comprehensively anticipating small-molecule selectivity to de-risk drug discovery. Expert Opin Drug Discov. 2024;19:1043–69. 39004919 10.1080/17460441.2024.2376643

87. Luciani F Safavi A Guruprasad P Chen L Ruella M Advancing CAR T-cell therapies with artificial intelligence: opportunities and challenges Blood Cancer Discov 2025 6 159 62 10.1158/2643-3230.BCD-23-0240 40152695 PMC12050963 Luciani F, Safavi A, Guruprasad P, Chen L, Ruella M. Advancing CAR T-cell therapies with artificial intelligence: opportunities and challenges. Blood Cancer Discov. 2025;6:159–62. 40152695 10.1158/2643-3230.BCD-23-0240 PMC12050963

88. Li P, Jiang Z, Liu T, Liu X, Qiao H, Yao X. Improving drug response prediction via integrating gene relationships with deep learning. Brief Bioinform. 2024;25. 10.1093/bib/bbae153 PMC11006795 38600666

89. Park A Joo M Kim K Son W-J Lim G Lee J A comprehensive evaluation of regression-based drug responsiveness prediction models, using cell viability inhibitory concentrations (IC50 values) Bioinformatics 2022 38 2810 17 10.1093/bioinformatics/btac177 35561188 Park A, Joo M, Kim K, Son W-J, Lim G, Lee J, et al. A comprehensive evaluation of regression-based drug responsiveness prediction models, using cell viability inhibitory concentrations (IC50 values). Bioinformatics. 2022;38:2810–17. 35561188 10.1093/bioinformatics/btac177

90. Wang Y Yu X Gu Y Li W Zhu K Chen L XGraphCDS: an explainable deep learning model for predicting drug sensitivity from gene pathways and chemical structures Comput Biol Med 2024 168 107746 10.1016/j.compbiomed.2023.107746 38039896 Wang Y, Yu X, Gu Y, Li W, Zhu K, Chen L, et al. XGraphCDS: an explainable deep learning model for predicting drug sensitivity from gene pathways and chemical structures. Comput Biol Med. 2024;168:107746. 38039896 10.1016/j.compbiomed.2023.107746

91. Pang W Chen M Qin Y Prediction of anticancer drug sensitivity using an interpretable model guided by deep learning BMC Bioinf 2024 25 182 10.1186/s12859-024-05669-x PMC11080240 38724920 Pang W, Chen M, Qin Y. Prediction of anticancer drug sensitivity using an interpretable model guided by deep learning. BMC Bioinf. 2024;25:182. 10.1186/s12859-024-05669-x PMC11080240 38724920

92. Yang W Soares J Greninger P Edelman EJ Lightfoot H Forbes S Genomics of drug sensitivity in cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells Nucleic Acids Res 2013 41 955 61 10.1093/nar/gks1111 PMC3531057 23180760 Yang W, Soares J, Greninger P, Edelman EJ, Lightfoot H, Forbes S, et al. Genomics of drug sensitivity in cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res. 2013;41:955–61. 10.1093/nar/gks1111 PMC3531057 23180760

93. Tom Hu Z, Yu Y, Chen R, S-J Y, Chen B, Huang H. Large-scale information retrieval and correction of noisy pharmacogenomic datasets through residual thresholded deep matrix factorization. Brief Bioinform. 2025;26. 10.1093/bib/bbaf226 PMC12106859 40420482

94. Chang Y Park H Yang HJ Lee S Lee KY Kim TS Cancer drug response profile scan (CDRscan): a deep learning model that predicts drug effectiveness from cancer genomic signature Sci Rep 2018 8 1 11 29891981 10.1038/s41598-018-27214-6 PMC5996063 Chang Y, Park H, Yang HJ, Lee S, Lee KY, Kim TS, et al. Cancer drug response profile scan (CDRscan): a deep learning model that predicts drug effectiveness from cancer genomic signature. Sci Rep. 2018;8:1–11. 29891981 10.1038/s41598-018-27214-6 PMC5996063

95. Chiu Y-C Chen H-I Zhang T Zhang S Gorthi A Wang L-J Correction to: predicting drug response of tumors from integrated genomic profiles by deep neural networks BMC Med Genomics 2019 12 119 10.1186/s12920-019-0569-5 31405368 PMC6689858 Chiu Y-C, Chen H-I, Zhang T, Zhang S, Gorthi A, Wang L-J, et al. Correction to: predicting drug response of tumors from integrated genomic profiles by deep neural networks. BMC Med Genomics. 2019;12:119. 31405368 10.1186/s12920-019-0569-5 PMC6689858

96. Park A Lee Y Nam S A performance evaluation of drug response prediction models for individual drugs Sci Rep 2023 13 11911 10.1038/s41598-023-39179-2 37488424 PMC10366128 Park A, Lee Y, Nam S. A performance evaluation of drug response prediction models for individual drugs. Sci Rep. 2023;13:11911. 37488424 10.1038/s41598-023-39179-2 PMC10366128

97. Kuenzi B Park J Fong S Sanchez K Lee J Kreisberg J Predicting drug response and synergy using a deep learning model of human cancer cells Cancer Cell 2020 38 672 84 10.1016/j.ccell.2020.09.014 33096023 PMC7737474 Kuenzi B, Park J, Fong S, Sanchez K, Lee J, Kreisberg J, et al. Predicting drug response and synergy using a deep learning model of human cancer cells. Cancer Cell. 2020;38:672–84. 33096023 10.1016/j.ccell.2020.09.014 PMC7737474

98. Adès L Duployez N Guerci-Bresler A Laribi K Peterlin P Vey N A randomised phase II study of azacitidine (AZA) alone or with lenalidomide (LEN), valproic acid (VPA) or idarubicin (IDA) in higher-risk MDS or low blast AML: GFM’s “pick a winner” trial, with the impact of somatic mutations Br J Haematol 2022 198 535 44 10.1111/bjh.18193 35438802 Adès L, Duployez N, Guerci-Bresler A, Laribi K, Peterlin P, Vey N, et al. A randomised phase II study of azacitidine (AZA) alone or with lenalidomide (LEN), valproic acid (VPA) or idarubicin (IDA) in higher-risk MDS or low blast AML: GFM’s “pick a winner” trial, with the impact of somatic mutations. Br J Haematol. 2022;198:535–44. 35438802 10.1111/bjh.18193

99. Kunacheewa C Thongthang P Ungprasert P Utchariyaprasit E Owattanapanich W A systematic review and meta-analysis of the efficacy and adverse events of azacitidine-plus-lenalidomide treatment for patients with acute myeloid leukemia, myelodysplastic syndromes and chronic myelomonocytic leukemia Hematology 2019 24 498 506 10.1080/16078454.2019.1631425 31221030 Kunacheewa C, Thongthang P, Ungprasert P, Utchariyaprasit E, Owattanapanich W. A systematic review and meta-analysis of the efficacy and adverse events of azacitidine-plus-lenalidomide treatment for patients with acute myeloid leukemia, myelodysplastic syndromes and chronic myelomonocytic leukemia. Hematology. 2019;24:498–506. 31221030 10.1080/16078454.2019.1631425

100. Abdel-Aziz AK, Minucci S. Comparing apples with oranges: studying LSD1 inhibitors in cellular assays. Pharmacol Res. 2019;104345. 10.1016/j.phrs.2019.104345 31295569

101. Chia S Wen Seow JJ Peres da Silva R Suphavilai C Shirgaonkar N Murata-Hori M CAN-Scan: a multi-omic phenotype-driven precision oncology platform identifies prognostic biomarkers of therapy response for colorectal cancer Cell Rep Med 2025 6 102053 10.1016/j.xcrm.2025.102053 40187357 PMC12047494 Chia S, Wen Seow JJ, Peres da Silva R, Suphavilai C, Shirgaonkar N, Murata-Hori M, et al. CAN-Scan: a multi-omic phenotype-driven precision oncology platform identifies prognostic biomarkers of therapy response for colorectal cancer. Cell Rep Med. 2025;6:102053. 40187357 10.1016/j.xcrm.2025.102053 PMC12047494

102. Califf RM Biomarker definitions and their applications Exp Biol Med 2018 243 213 21 10.1177/1535370217750088 PMC5813875 29405771 Califf RM. Biomarker definitions and their applications. Exp Biol Med. 2018;243:213–21. 10.1177/1535370217750088 PMC5813875 29405771

103. Abdel-Aziz AK. OXPHOS mediators in acute myeloid leukemia patients: prognostic biomarkers and therapeutic targets for personalized medicine. World J Surg Oncol. 2024;22. 10.1186/s12957-024-03581-5 PMC11559054 39533394

104. Helmy SWA Abdel-Aziz AK Dokla EME Ahmed TE Hatem Y Abdel Rahman EA Novel sulfonamide-indolinone hybrids targeting mitochondrial respiration of breast cancer cells Eur J Med Chem 2024 268 1 18 10.1016/j.ejmech.2024.116255 38401190 Helmy SWA, Abdel-Aziz AK, Dokla EME, Ahmed TE, Hatem Y, Abdel Rahman EA, et al. Novel sulfonamide-indolinone hybrids targeting mitochondrial respiration of breast cancer cells. Eur J Med Chem. 2024;268:1–18. 10.1016/j.ejmech.2024.116255 38401190

105. Haneberg AG Pierre K Winter-Reinhold E Hochhegger B Peters KR Grajo J Introduction to radiomics and artificial intelligence: a primer for radiologists Semin Roentgenol 2023 58 152 57 10.1053/j.ro.2023.02.002 37087135 Haneberg AG, Pierre K, Winter-Reinhold E, Hochhegger B, Peters KR, Grajo J, et al. Introduction to radiomics and artificial intelligence: a primer for radiologists. Semin Roentgenol. 2023;58:152–57. 37087135 10.1053/j.ro.2023.02.002

106. Zhou H Zhao Q Xie Q Peng Y Chen M Huang Z Preoperative prediction model of lymph node metastasis in the inguinal and femoral region based on radiomics and artificial intelligence Int J Gynecol Cancer 2024 34 1437 44 10.1136/ijgc-2024-005580 39089728 Zhou H, Zhao Q, Xie Q, Peng Y, Chen M, Huang Z, et al. Preoperative prediction model of lymph node metastasis in the inguinal and femoral region based on radiomics and artificial intelligence. Int J Gynecol Cancer. 2024;34:1437–44. 39089728 10.1136/ijgc-2024-005580

107. Zhou L, Ouyang L, Guo B, Huang X, Yang S, Pan J, et al. Enhanced prediction of preoperative mesenteric lymph node metastasis in colorectal cancer using machine learning with CT-based data. MedComm - Futur Med. 2024;3.

108. Monthatip K Boonnag C Muangmool T Charoenkwan K A machine learning-based prediction model of pelvic lymph node metastasis in women with early-stage cervical cancer J Gynecol Oncol 2024 35 1 12 10.3802/jgo.2024.35.e17 PMC10948976 37921601 Monthatip K, Boonnag C, Muangmool T, Charoenkwan K. A machine learning-based prediction model of pelvic lymph node metastasis in women with early-stage cervical cancer. J Gynecol Oncol. 2024;35:1–12. 10.3802/jgo.2024.35.e17 PMC10948976 37921601

109. Cepeda S, Luppino LT, Pérez-Núñez A, Solheim O, García-García S, Velasco-Casares M, et al. Predicting regions of local recurrence in glioblastomas using voxel-based radiomic features of multiparametric postoperative MRI. Cancers (basel). 2023;15. 10.3390/cancers15061894 PMC10047582 36980783

110. Koyama J Morise M Furukawa T Oyama S Matsuzawa R Tanaka I Artificial intelligence-based personalized survival prediction using clinical and radiomics features in patients with advanced non-small cell lung cancer BMC Cancer 2024 24 1417 10.1186/s12885-024-13190-w 39558311 PMC11572056 Koyama J, Morise M, Furukawa T, Oyama S, Matsuzawa R, Tanaka I, et al. Artificial intelligence-based personalized survival prediction using clinical and radiomics features in patients with advanced non-small cell lung cancer. BMC Cancer. 2024;24:1417. 39558311 10.1186/s12885-024-13190-w PMC11572056

111. Al-Obeidat F Rashid A Hafez W Gibbaoui H Ayoub G Al Ameer S The accuracy of artificial intelligence in the diagnosis of soft tissue sarcoma: a systematic review and meta-analysis Curr Probl Surg 2025 66 101743 10.1016/j.cpsurg.2025.101743 40306879 Al-Obeidat F, Rashid A, Hafez W, Gibbaoui H, Ayoub G, Al Ameer S, et al. The accuracy of artificial intelligence in the diagnosis of soft tissue sarcoma: a systematic review and meta-analysis. Curr Probl Surg. 2025;66:101743. 40306879 10.1016/j.cpsurg.2025.101743

112. Kang H Witanto JN Pratama K Lee D Choi KS Choi SH Fully automated MRI segmentation and volumetric measurement of intracranial meningioma using deep learning J Magn Reson Imag 2023 57 871 81 10.1002/jmri.28332 35775971 Kang H, Witanto JN, Pratama K, Lee D, Choi KS, Choi SH, et al. Fully automated MRI segmentation and volumetric measurement of intracranial meningioma using deep learning. J Magn Reson Imag. 2023;57:871–81. 10.1002/jmri.28332 35775971

113. Xin W Rixin S Linrui L Zhihui Q Long L Yu Z Machine learning-based radiomics for predicting outcomes in cervical cancer patients undergoing concurrent chemoradiotherapy Comput Biol Med 2024 177 108593 10.1016/j.compbiomed.2024.108593 38801795 Xin W, Rixin S, Linrui L, Zhihui Q, Long L, Yu Z. Machine learning-based radiomics for predicting outcomes in cervical cancer patients undergoing concurrent chemoradiotherapy. Comput Biol Med. 2024;177:108593. 38801795 10.1016/j.compbiomed.2024.108593

114. Ligero M Gielen B Navarro V Cresta Morgado P Prior O Dienstmann R A whirl of radiomics-based biomarkers in cancer immunotherapy, why is large scale validation still lacking? Npj Precis Oncol 2024 8 42 10.1038/s41698-024-00534-9 38383736 PMC10881558 Ligero M, Gielen B, Navarro V, Cresta Morgado P, Prior O, Dienstmann R, et al. A whirl of radiomics-based biomarkers in cancer immunotherapy, why is large scale validation still lacking? Npj Precis Oncol. 2024;8:42. 38383736 10.1038/s41698-024-00534-9 PMC10881558

115. Tian P Chen Z Fang B Wang X Yu X Lu M Investigating the relationship between tongue diagnosis features and gastric cancer: a machine learning-based prediction model Eur J Surg Oncol 2025 51 1 8 10.1016/j.ejso.2025.110352 40730130 Tian P, Chen Z, Fang B, Wang X, Yu X, Lu M. Investigating the relationship between tongue diagnosis features and gastric cancer: a machine learning-based prediction model. Eur J Surg Oncol. 2025;51:1–8. 10.1016/j.ejso.2025.110352 40730130

116. Li S-W Zhang L-H Cai Y Zhou X-B Fu X-Y Song Y-Q Deep learning assists detection of esophageal cancer and precursor lesions in a prospective, randomized controlled study Sci Transl Med 2024 16 eadk 5395 10.1126/scitranslmed.adk5395 38630847 Li S-W, Zhang L-H, Cai Y, Zhou X-B, Fu X-Y, Song Y-Q, et al. Deep learning assists detection of esophageal cancer and precursor lesions in a prospective, randomized controlled study. Sci Transl Med. 2024;16:eadk 5395. 10.1126/scitranslmed.adk5395 38630847

117. Chen H Gou L Fang Z Dou Q Chen H Chen C Artificial intelligence assisted real-time recognition of intra-abdominal metastasis during laparoscopic gastric cancer surgery NPJ Digit Med 2025 8 1 12 10.1038/s41746-024-01372-6 39757250 PMC11701130 Chen H, Gou L, Fang Z, Dou Q, Chen H, Chen C, et al. Artificial intelligence assisted real-time recognition of intra-abdominal metastasis during laparoscopic gastric cancer surgery. NPJ Digit Med. 2025;8:1–12. 39757250 10.1038/s41746-024-01372-6 PMC11701130

118. Barua I Wieszczy P Kudo S Misawa M Holme Ø Gulati S Real-time artificial intelligence-based optical diagnosis of neoplastic polyps during colonoscopy NEJM Evid 2022 1 1 10 10.1056/EVIDoa2200003 38319238 Barua I, Wieszczy P, Kudo S, Misawa M, Holme Ø, Gulati S, et al. Real-time artificial intelligence-based optical diagnosis of neoplastic polyps during colonoscopy. NEJM Evid. 2022;1:1–10. 10.1056/EVIDoa2200003 38319238

119. Lebret T Pignot G Colombel M Guy L Rebillard X Savareux L Artificial intelligence to improve cytology performances in bladder carcinoma detection: results of the VisioCyt test BJU Int 2022 129 356 63 10.1111/bju.15382 33751774 Lebret T, Pignot G, Colombel M, Guy L, Rebillard X, Savareux L, et al. Artificial intelligence to improve cytology performances in bladder carcinoma detection: results of the VisioCyt test. BJU Int. 2022;129:356–63. 33751774 10.1111/bju.15382

120. Lebret T Paoletti X Pignot G Roumiguié M Colombel M Savareux L Artificial intelligence to improve cytology performance in urothelial carcinoma diagnosis: results from validation phase of the French, multicenter, prospective VISIOCYT1 trial World J Urol 2023 41 2381 88 10.1007/s00345-023-04519-4 37480491 PMC10465399 Lebret T, Paoletti X, Pignot G, Roumiguié M, Colombel M, Savareux L, et al. Artificial intelligence to improve cytology performance in urothelial carcinoma diagnosis: results from validation phase of the French, multicenter, prospective VISIOCYT1 trial. World J Urol. 2023;41:2381–88. 37480491 10.1007/s00345-023-04519-4 PMC10465399

121. Aljabban J Syed S Syed S Rohr M Mukhtar M Aljabban H Characterization of monoclonal gammopathy of undetermined significance progression to multiple myeloma through meta-analysis of GEO data Heliyon 2023 9 e17298 10.1016/j.heliyon.2023.e17298 37539132 PMC10394915 Aljabban J, Syed S, Syed S, Rohr M, Mukhtar M, Aljabban H, et al. Characterization of monoclonal gammopathy of undetermined significance progression to multiple myeloma through meta-analysis of GEO data. Heliyon. 2023;9:e17298. 37539132 10.1016/j.heliyon.2023.e17298 PMC10394915

122. Ruhela V Gupta R Oberoi R Gupta A A comprehensive targeted panel of 295 genes: unveiling key disease initiating and transformative biomarkers in multiple myeloma Comput Biol Med 2025 196 110619 10.1016/j.compbiomed.2025.110619 40617081 Ruhela V, Gupta R, Oberoi R, Gupta A. A comprehensive targeted panel of 295 genes: unveiling key disease initiating and transformative biomarkers in multiple myeloma. Comput Biol Med. 2025;196:110619. 40617081 10.1016/j.compbiomed.2025.110619

123. Hou Y Zhang F Zong J Li T Gan W Lv S Integrated analysis reveals a novel 5-fluorouracil resistance-based prognostic signature with promising implications for predicting the efficacy of chemotherapy and immunotherapy in patients with colorectal cancer Apoptosis 2024 29 1126 44 10.1007/s10495-024-01981-2 38824480 Hou Y, Zhang F, Zong J, Li T, Gan W, Lv S, et al. Integrated analysis reveals a novel 5-fluorouracil resistance-based prognostic signature with promising implications for predicting the efficacy of chemotherapy and immunotherapy in patients with colorectal cancer. Apoptosis. 2024;29:1126–44. 38824480 10.1007/s10495-024-01981-2

124. Kather JN Pearson AT Halama N Jäger D Krause J Loosen SH Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer Nat Med 2019 25 1054 56 10.1038/s41591-019-0462-y 31160815 PMC7423299 Kather JN, Pearson AT, Halama N, Jäger D, Krause J, Loosen SH, et al. Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer. Nat Med. 2019;25:1054–56. 31160815 10.1038/s41591-019-0462-y PMC7423299

125. Kather JN Krisam J Charoentong P Luedde T Herpel E Weis CA Predicting survival from colorectal cancer histology slides using deep learning: a retrospective multicenter study PLoS Med 2019 16 1 22 10.1371/journal.pmed.1002730 PMC6345440 30677016 Kather JN, Krisam J, Charoentong P, Luedde T, Herpel E, Weis CA, et al. Predicting survival from colorectal cancer histology slides using deep learning: a retrospective multicenter study. PLoS Med. 2019;16:1–22. 10.1371/journal.pmed.1002730 PMC6345440 30677016

126. Loeffler CML, Bando H, Sainath S, Muti HS, Jiang X, van Treeck M, et al. HIBRID: histology-based risk-stratification with deep learning and ctDNA in colorectal cancer. Nat Commun. 2025;16. 10.1038/s41467-025-62910-8 PMC12354865 40813777

127. Roddam AW Duffy MJ Hamdy FC Ward AM Patnick J Price CP Use of prostate-specific antigen (PSA) isoforms for the detection of prostate cancer in men with a PSA level of 2–10 ng/ml: systematic review and meta-analysis Eur Urol 2005 48 386 89 10.1016/j.eururo.2005.04.015 15982797 Roddam AW, Duffy MJ, Hamdy FC, Ward AM, Patnick J, Price CP, et al. Use of prostate-specific antigen (PSA) isoforms for the detection of prostate cancer in men with a PSA level of 2–10 ng/ml: systematic review and meta-analysis. Eur Urol. 2005;48:386–89. 15982797 10.1016/j.eururo.2005.04.015

128. Shu CP, Sop TGT, Tanyi TJ, Lambi SJ, Signang NA, Juma PI. The sensitivity and specificity of total prostate specific antigen in the diagnosis of prostate cancer in a resource-limited African setting: a 5-year review in a Kenyan hospital. Clin Genitourin Cancer. 2025;23. 10.1016/j.clgc.2025.102363 40373482

129. Smelik M Diaz-Roncero Gonzalez D An X Heer R Henningsohn L Li X Combining spatial transcriptomics, pseudotime, and machine learning enables discovery of biomarkers for prostate cancer Cancer Res 2025 85 2514 26 10.1158/0008-5472.CAN-25-0269 40293712 PMC12214874 Smelik M, Diaz-Roncero Gonzalez D, An X, Heer R, Henningsohn L, Li X, et al. Combining spatial transcriptomics, pseudotime, and machine learning enables discovery of biomarkers for prostate cancer. Cancer Res. 2025;85:2514–26. 40293712 10.1158/0008-5472.CAN-25-0269 PMC12214874

130. Zhou N Zhang C-T Lv H-Y Hao C-X Li T-J Zhu J-J Concordance study between IBM Watson for oncology and clinical practice for patients with cancer in China Oncologist 2019 24 812 19 10.1634/theoncologist.2018-0255 30181315 PMC6656482 Zhou N, Zhang C-T, Lv H-Y, Hao C-X, Li T-J, Zhu J-J, et al. Concordance study between IBM Watson for oncology and clinical practice for patients with cancer in China. Oncologist. 2019;24:812–19. 30181315 10.1634/theoncologist.2018-0255 PMC6656482

131. Jamrat S Sukasem C Sratthaphut L Hongkaew Y Samanchuen T A precision medicine approach to personalized prescribing using genetic and nongenetic factors for clinical decision-making Comput Biol Med 2023 165 107329 10.1016/j.compbiomed.2023.107329 37611418 Jamrat S, Sukasem C, Sratthaphut L, Hongkaew Y, Samanchuen T. A precision medicine approach to personalized prescribing using genetic and nongenetic factors for clinical decision-making. Comput Biol Med. 2023;165:107329. 37611418 10.1016/j.compbiomed.2023.107329

132. Paganelli A Spadafora M Navarrete-Dechent C Guida S Pellacani G Longo C Natural language processing in dermatology: a systematic literature review and state of the art J Eur Acad Dermatol Venereol 2024 38 2225 34 10.1111/jdv.20286 39150311 PMC11587683 Paganelli A, Spadafora M, Navarrete-Dechent C, Guida S, Pellacani G, Longo C. Natural language processing in dermatology: a systematic literature review and state of the art. J Eur Acad Dermatol Venereol. 2024;38:2225–34. 39150311 10.1111/jdv.20286 PMC11587683

133. Linna N, Kahn CE. Applications of natural language processing in radiology: a systematic review. Int J Med Inf [Internet]. 2022;163:104779. Available from: 10.1016/j.ijmedinf.2022.104779 10.1016/j.ijmedinf.2022.104779 35533413

134. Li X Dowling EK Yan G Dereli Z Bozorgui B Imanirad P Precision combination therapies based on recurrent oncogenic Coalterations Cancer Discov 2022 12 1542 59 10.1158/2159-8290.CD-21-0832 35412613 PMC9524464 Li X, Dowling EK, Yan G, Dereli Z, Bozorgui B, Imanirad P, et al. Precision combination therapies based on recurrent oncogenic Coalterations. Cancer Discov. 2022;12:1542–59. 35412613 10.1158/2159-8290.CD-21-0832 PMC9524464

135. Ammo T Guillaume VGJ Hofmann UK Ulmer NM Buenting N Laenger F Evaluating ChatGPT-4o as a decision support tool in multidisciplinary sarcoma tumor boards: heterogeneous performance across various specialties Front Oncol 2024 14 1 10 10.3389/fonc.2024.1526288 PMC11782276 39896191 Ammo T, Guillaume VGJ, Hofmann UK, Ulmer NM, Buenting N, Laenger F, et al. Evaluating ChatGPT-4o as a decision support tool in multidisciplinary sarcoma tumor boards: heterogeneous performance across various specialties. Front Oncol. 2024;14:1–10. 10.3389/fonc.2024.1526288 PMC11782276 39896191

136. Akdogan O Uyar GC Yesilbas E Baskurt K Malkoc NA Ozdemir N Effect of a ChatGPT-based digital counseling intervention on anxiety and depression in patients with cancer: a prospective, randomized trial Eur J Cancer 2025 221 115408 10.1016/j.ejca.2025.115408 40215593 Akdogan O, Uyar GC, Yesilbas E, Baskurt K, Malkoc NA, Ozdemir N, et al. Effect of a ChatGPT-based digital counseling intervention on anxiety and depression in patients with cancer: a prospective, randomized trial. Eur J Cancer. 2025;221:115408. 40215593 10.1016/j.ejca.2025.115408

137. Al-Hilli Z Noss R Dickard J Wei W Chichura A Wu V A randomized trial comparing the effectiveness of pre-test genetic counseling using an artificial intelligence automated Chatbot and traditional In-person genetic counseling in women newly diagnosed with breast cancer Ann Surg Oncol 2023 30 5990 96 10.1245/s10434-023-13888-4 37567976 Al-Hilli Z, Noss R, Dickard J, Wei W, Chichura A, Wu V, et al. A randomized trial comparing the effectiveness of pre-test genetic counseling using an artificial intelligence automated Chatbot and traditional In-person genetic counseling in women newly diagnosed with breast cancer. Ann Surg Oncol. 2023;30:5990–96. 37567976 10.1245/s10434-023-13888-4

138. Yang E-W Velazquez-Villarreal E Nikolski M AI-HOPE: an AI-driven conversational agent for enhanced clinical and genomic data integration in precision medicine research Bioinformatics 2025 41 10.1093/bioinformatics/btaf359 PMC12212640 40577785 Yang E-W, Velazquez-Villarreal E. AI-HOPE: an AI-driven conversational agent for enhanced clinical and genomic data integration in precision medicine research. In: Nikolski M, editor. Bioinformatics. 2025. p. 41. 10.1093/bioinformatics/btaf359 PMC12212640 40577785

139. Abdelhalim H, Berber A, Lodi M, Jain R, Nair A, Pappu A, et al. Artificial intelligence, healthcare, clinical genomics, and pharmacogenomics approaches in precision medicine. Front Genet. 2022;13. 10.3389/fgene.2022.929736 PMC9299079 35873469

140. Sebastian AM, Peter D. Artificial intelligence in cancer research: trends, challenges and future directions. Life. 2022;12. 10.3390/life12121991 PMC9786074 36556356

141. Mehraeen E, Siami H, Montazeryan S, Molavi R, Feyzabadi A, Parvizy I, et al. Artificial intelligence challenges in the healthcare industry: a systematic review of recent evidence. Healthc Technol Lett. 2025;12. 10.1049/htl2.70017 PMC12402815 40904736

142. Kore A, Abbasi Bavil E, Subasri V, Abdalla M, Fine B, Dolatabadi E, et al. Empirical data drift detection experiments on real-world medical imaging data. Nat Commun. 2024;15. 10.1038/s41467-024-46142-w PMC10904813 38424096

143. Dankwa-Mullan I Ndoh K Akogo D Rocha HAL Juaçaba SF Artificial intelligence and cancer health equity: bridging the divide or widening the gap Curr Oncol Rep 2025 27 95 111 10.1007/s11912-024-01627-1 39753817 Dankwa-Mullan I, Ndoh K, Akogo D, Rocha HAL, Juaçaba SF. Artificial intelligence and cancer health equity: bridging the divide or widening the gap. Curr Oncol Rep. 2025;27:95–111. 39753817 10.1007/s11912-024-01627-1

144. Khan MS Umer H Faruqe F Artificial intelligence for low income countries Humanit Soc Sci Commun 2024 11 1 13 10.1057/s41599-024-03947-w Khan MS, Umer H, Faruqe F. Artificial intelligence for low income countries. Humanit Soc Sci Commun. 2024;11:1–13.

145. Abgrall G Holder AL Chelly Dagdia Z Zeitouni K Monnet X Should AI models be explainable to clinicians? Crit Care 2024 28 1 8 10.1186/s13054-024-05005-y 39267172 PMC11391805 Abgrall G, Holder AL, Chelly Dagdia Z, Zeitouni K, Monnet X. Should AI models be explainable to clinicians? Crit Care. 2024;28:1–8. 39267172 10.1186/s13054-024-05005-y PMC11391805

146. Xu C, Coen-Pirani P, Jiang X. Empirical study of overfitting in deep learning for predicting breast cancer metastasis. Cancers (Basel). 2023;15. 10.3390/cancers15071969 PMC10093528 37046630

147. World Health Organization. Ethics and Governance of Artificial Intelligence for Health. Guidance. 2024.

148. Warraich HJ Tazbaz T Califf RM FDA perspective on the regulation of artificial intelligence in health care and biomedicine JAMA 2025 333 241 47 10.1001/jama.2024.21451 39405330 Warraich HJ, Tazbaz T, Califf RM. FDA perspective on the regulation of artificial intelligence in health care and biomedicine. JAMA. 2025;333:241–47. 39405330 10.1001/jama.2024.21451

149. Food and Drug Administration. Considerations for the use of artificial intelligence to support regulatory decision-making for drug and biological products. 2025. https://www.fda.gov/media/184830/download, 1–23. Available from.

150. UK Department of Health and Social Care. A guide to good practice for digital and data-driven health technologiesgovernment/publications/code-of-conduct-for-data-driven-health-and-care-technology/initial-code-of-conduct-for-data-driven-health-and-care-technology [Internet]. 2021. https://www.gov.uk/government/publications/code-of-conduct-for-data- . cited 2025 Sep 21]. p. 1–26. Available from.

151. Sharaky M Dokla EME Abdel-Aziz AK Anticancer activity of EMD37 against human head and neck cancer: impact on apoptotic and inflammatory machineries Toxicol Vitr 2025 102 105967 10.1016/j.tiv.2024.105967 39510359 Sharaky M, Dokla EME, Abdel-Aziz AK. Anticancer activity of EMD37 against human head and neck cancer: impact on apoptotic and inflammatory machineries. Toxicol Vitr. 2025;102:105967. 10.1016/j.tiv.2024.105967 39510359

152. Abdel-Aziz AK, Dokla EME, Abouzid KAM, Minucci S. Discovery of EMD37, a 1, 2, 4-oxadiazole derivative, as a novel endoplasmic reticulum stress inducer with potent anticancer activity. Biochem Pharmacol. 2022;115316. 10.1016/j.bcp.2022.115316 36283444

153. Id NAP, Xia Y, Estabrooks SK, Ye Z, Id RMB, Assimon VA, et al. Predicting protein targets for drug-like compounds using transcriptomics. 2018;1–24. 10.1371/journal.pcbi.1006651 PMC6300300 30532261