Computational approaches in rheumatic diseases - Deciphering complex spatio-temporal cell interactions.

Introduction

Inflammatory arthritis (IA) comprises a group of diverse chronic autoimmune diseases that include Rheumatoid arthritis (RA), Psoriatic Arthritis (RA), Peripheral Spondylarthritis (SpA) and Systemic Lupus Erythematosus (SLE), amongst others. The complexity of IA pathogenesis stems from the interplay between environmental triggers and genetic predisposition [1] . The primary target of IA is the synovial tissue. Synovial tissue inflammation, if not adequately treated will lead to loss of function of the joint and is associated with reduced quality of life, extraarticular manifestations and significant comorbidities resulting in reduced life expectancy [2] . The molecular pathways and cellular populations involved in synovial immunopathology differ between IA and are varied even for the same disease due to intra-disease heterogeneity [3] . Temporal differences in the molecular pathways underpinning synovial inflammation highlight the dynamic nature of the immune mechanisms involved and may differ significantly between disease stages [4] . An unwanted by-product of the complexity of synovial pathophysiology is refractory response to treatment, with approximately 40% of patients with RA not responding to specific treatments and up to 20% not responding to any of the currently available therapies [5] , [6] . The rapid implementation of omic approaches has increased the resolution of the cellular landscape and our understanding of molecular pathways involved in synovial inflammation. The increased availability of transcriptomic, proteomic, epigenetic and flow cytometric datasets will require the use of machine learning (ML) models in order to concatenate data, build the regulatory pathways and interpret cell-cell interactions leading to a unifying model of synovial pathogenesis of IA.

Histological characterisation of synovial inflammation

Synovial inflammation, or synovitis, refers to inflammation of the synovial membrane, which lines the joint cavity and supports the normal function of the joint. The synovial membrane is vascularised and contributes nutrients and oxygen that maintain metabolic homeostasis and synovial fluid that lubricates the joint [7] . In IA, the synovial tissue morphology is dysregulated due to a characteristic hyperproliferation of synovial fibroblasts of the lining layer and increased neo-vascularisation with the formation of new immature, plastic blood vessels [8] . As a result of the extensive changes in vascularisation, delivery of oxygen and nutrients in the synovium is compromised leading to a highly hypoxic environment that correlates with increased infiltration of immune cells including, monocytes, macrophages, T cells, B cells, plasma cells and neutrophils [8] , [9] , [10] , [11] , [12] . If left, untreated synovial inflammation leads to bone erosion and loss of function of the joint.

Understanding of the pathogenesis of synovial inflammation has evolved and improved over time. For the most part, studies of synovial inflammation pathogenesis relied on histological evaluation of patient synovial tissue biopsies [2] . The major histological alterations that have been identified in the inflamed synovial tissue are synovial lining hyperplasia, immune cell infiltration and neo angiogenesis [13] . The use of histology in studying synovial inflammation involves the acquisition of a tissue biopsy, and histopathological scoring and characterisation of the biopsy is performed [14] . Over the years, advances in staining techniques and imaging technology have allowed for a more detailed characterisation of synovial histology. Haematoxylin and eosin (H&E) staining was the standard histological staining method used for synovitis assessment. Histological characterisation involves analysis of synovial biopsies and allows for the characterisation of structural factors such as the composition of cellular infiltrate and the existence of tumorous structures. This analysis has also been utilised in the differentiation of inflammatory arthropathies [15] , [16] .

One standardised evaluation method used in the differential diagnosis of arthropathies is the Krenn histopathological synovitis score [17] . This reproducible evaluation allows for the differentiation between inflammatory vs non-inflammatory arthritis [18] . This score is based on the microscopic histological alterations of synovitis, specifically lining layer hyperplasia, neoangiogenesis and immune cell infiltrate [17] .

Frequently, histological examination of synovial biopsies in RA reveals marked infiltration of immune cells, hyperplasia, and the involvement of inflammatory cascades [19] . Patient with RA synovial tissue has also been shown to express high levels of proinflammatory cytokines, such as tumour necrosis factor alpha (TNFα), interleukin 1 (IL-1) and interleukin 6 (IL-6) [20] . One hallmark of RA is the development of pannus, an abnormal tissue that arises due to the excessive inflammation caused by an increased concentration of immune cells and cytokines, resulting in increased receptor activator for nuclear factor kappa-B ligand (RANK-L) expression, osteoclast activation and bone erosion [21] .

Another common chronic inflammatory arthritis with synovial and systemic manifestations is PsA. RA and PsA share some common clinical characteristics; however, the underlying immune mechanisms involved are different as evident by differential response to treatment [22] , [23] . The potential availability of matched peripheral blood, skin and synovial tissue samples from patients with PsA a the availability of data on response to therapy have encouraged the utilisation of integrated omic data [23] . The high heterogeneity of the histological features of PsA synovial tissue makes classification a challenge, however, both oligo- and poly-articular PsA have more histopathological features in common with spondyloarthropathy (SpA) rather than RA and include increased neutrophil infiltration and absence of intracellular citrullinated proteins [15] . Probably the most striking histological feature that separates PsA from RA is vascular morphology, with PsA displaying complex tortuous, bushy vessels brought on by dysregulated angiogenesis versus RA which is characterised by straight, branching vessels [24] , [25] .

In addition to the histological differences observed between RA and PsA, there is significant heterogeneity that strongly supports the existence of disease specific pathotypes. Clinical heterogeneity as evident by variations in number and location of affected joints, autoantibody levels and types, serum cytokines and disease progression are reflected in variations of synovial tissue inflammation [18] .

Currently, for reasons that are poorly understood, approximately 40% of patients with RA do not respond to treatment [26] . Due to the lack of appropriate markers for diagnosis, patient stratification and response to treatment, precious time is lost at the early stages of synovial inflammation resulting in increased tissue damage and poor prognosis [27] . Awareness of disease pathotypes can improve patient stratification, expedite the identification of the right treatment for the right patient and improve disease remission rates [28] . By histological evaluation three distinct pathotypes of RA have been identified, the lympho-myeloid pathotype is the most common in patients with RA and is characterised by lymphoid cell infiltration (high CD20 and CD3 cell synovial staining) and ectopic lymphoid structure formation, the diffuse-myeloid pathotype primarily driven by macrophages (high CD68 synovial staining) and the pauci-immune pathotype with very low immune cell infiltration but substantial expansion of synovial fibroblasts [28] , [29] . Importantly, the three identified synovial pathotypes reflect differential underlying molecular pathway involvement and disease progression with increased TNF pathway and B cell related pathway scores significantly enriched in myeloid-diffuse and lymphoid pathotypes respectively [28] . Anti-TNF treatment showed increased efficacy for the myeloid-diffuse compared to the lympho-myeloid pathotype with the later, showing overall accelerated disease progression and joint damage [29] .

Limitations of histological evaluation of synovial pathology

Considering recent advances, a major limitation of histological classification of synovial tissue is that results cannot be obtained for low-tissue-quality samples [16] . As this is a two-dimensional representation of the tissue landscape, reconstructing the entire biopsy content can be time-consuming, and individual tissue slices will only provide information about a limited number of cells. Many studies have advised only analysing areas with the most significant alterations or multiple small biopsy areas. Previous research has suggested that 4–6 biopsies should be isolated per patient for optimal interpretation of synovial tissue inflammation [30] , [31] , [32] . Examination of only a small area of the joint can result in an underestimation of the extent and severity of inflammation and may not represent the degree of histopathology in the entire joint. Another common drawback is the standardisation of histological examination techniques that varies between centres. This is influenced by variations in protocols such as fixation, staining, antibodies used, patient cohort. Even under ideal conditions, histological analysis cannot be used to investigate specific molecular functions further. Histological scoring will also not allow for an effective, in-depth characterisation of the cellular infiltrate due to limitations on the number of targets that can be assessed simultaneously. Despite these limitations, histological evaluation of the synovial tissue biopsies when utilised in conjunction with omic approaches significantly contributes in deciphering the spatial organisation of the inflamed synovium as evident by the recent identification of protective, self-renewing CX 3 CR1 + lining macrophages that form an epithelial-like barrier that isolates the intra-articular space and protects against inflammation [33] .

Omic approaches improving the cellular resolution of the inflamed synovial tissue in IA

Recent utilisation of omic approaches has made a tremendous impact on our understanding of synovial tissue inflammation, the heterogeneity of the cellular landscape and the involvement of molecular pathways at different stages of disease. Although not considered a traditional omic, the evolution of flow cytometric analysis with a significant increase in dimensionality, versatility and data analysis have significant synergistic effects when paired with multi-omic approaches [34] . Particularly the introduction of spectral and mass flow cytometry allowing for single cell resolution of over fifty markers simultaneously can maximise the information gain from patient clinical samples [35] . Very early implementation of flow cytometric analysis of cultured synovial fibroblasts isolated from patient with RA synovial biopsies had limited capacity but showed differences in autofluorescence that reflect morphological changes and mitochondrial activity of RA derived compared to healthy control derived synovial fibroblasts [36] . Despite challenges in sample preparation, more recent implementation of mass cytometry has eluted to multiple and functionally distinct synovial fibroblast subsets complementing transcriptomic differences identified by scRNAseq [37] , [38] , [39] , [40] . Flow cytometric analysis has also been extensively utilised for the study of patient with IA synovial T cell responses resulting in the identification of pathogenic CD161 + ex-Th17 cells that show increased polyfunctionality due to the secretion of multiple pro-inflammatory cytokines simultaneously and resistance to autologous Treg cell suppression [41] , [42] . While evidence of polyfunctionality were present in early studies of synovial T cells, it was not until recently that the association between differential cytokine profiles in RA and PsA and disease progression was identified [43] , [44] , [45] , [46] , [47] . Distinct CD206 + MerTK + macrophage subset associated with clinical remission as well as protective CX 3 CR1 + macrophages have also been identified in the synovial tissue of RA patients on the basis of flow cytometric analysis paired with scRNAeq [33] , [48] .

The introduction of bulk RNAseq analysis to IA research has led to the identification of previously unappreciated molecular pathways involved in synovial inflammation raising possibilities for the development of novel therapeutics and predictive markers of disease progression. Deconvolution of bulk RNAseq data has shown distinct cellular involvement depending on RA pathotype with plasma cell gene enrichment representative of more aggressive synovial inflammation [3] . The CD40-CD40L signalling pathway showed a stepwise increase in enrichment from healthy to arthralgia to established RA patients raising the possibility for pharmacological blocking of the pathway [49] . Surprisingly, the expression of PD-1 and its ligands PD-L1 and PD-L2 showed a stepwise increase early in disease progression to RA, however, histological examination of synovial tissue biopsies showed very low levels of PD-L1 possibly due to epitope masking mediated by the increased levels of soluble PD-1 in the serum of patients with RA [50] . In addition to the identification of signalling pathway enrichment in RA synovial inflammation during different stages of disease, the potential identification of PsA specific transcriptomic signature in the blood could facilitate diagnostic biomarker discovery [51] .

Bulk RNAseq has been informative in elucidating the signalling pathways involved in different stages of synovial inflammation, however, it provides a “snapshot” of the overall transcriptome of the tissue and not the transcriptome of individual cells or the potential cell-cell interactions that may influence synovial pathology.

It has to be noted that despite limitations, bulk RNAseq on peripheral blood of patients with IA can inform on response to treatment and identification of novel biomarkers. RNA sequencing of CD4 T cell populations sorted from the peripheral blood of patients with RA before or after treatment with abatacept led to the identification of a 227 gene module that associated with disease progression (DAS28-CRP) with the transcriptomic profile of patients post treatment comparable to healthy controls [52] . Similarly, RNA sequencing of whole blood RNA of patients with RA prior to, or six months post rituximab mediated B cell depletion therapy, resulted in 198 differentially expressed genes between responders and non-responders [53] . Going beyond biomarker identification, peripheral blood transcriptomics in combination with in vitro studies could inform on novel therapeutic targets for subpopulations of patients with RA while deconvolution of transcriptomic data could provide insights regarding the relative frequency of immune cell populations at different stages of disease [54] , [55] . The added benefit compared to flow cytometric analysis is that as deconvolution methods and cell specific signatures become more detailed, the same data can be reanalysed to achieve higher resolution or to identify new immune cell populations and activation states [56] . Currently, however, bioinformatically deconvoluted data should be supported by flow cytometric analysis. Despite the aforementioned advances, biomarker identification continues to be difficult, this could be in part explained by a recent analysis of peripheral blood and synovial tissue RNAseq data, including scRNAseq from patients with RA and a murine model of arthritis revealing little overlap in molecular pathway involvement between peripheral blood and synovial tissue samples [57] . The application of machine learning (ML) models on peripheral blood transcriptomic data may expedite biomarker identification however, were possible, in parallel analysis of matched peripheral blood and synovial tissue omic data may provide a better degree of confidence [58] .

One major finding from recent studies utilising scRNAseq analysis of synovial tissue biopsies is the heterogeneity of immune and stromal cell populations as highlighted by the identification of a high number of functionally distinct synovial fibroblast clusters [37] . Macrophages are also highly heterogeneous at the site of inflammation in RA with CD206 + MerTK + associated with remission and increased expression of anti-inflammatory resolvin and IL-10 [48] . Although there is an accumulation of scRNAseq data for RA patient synovial tissue, until recently the transcriptomic cellular landscape in PsA was unknown. Recent scRNAseq characterisation of RA and PsA patient synovial tissue biopsies revealed T cell and synovial macrophage synergy mediated by TGFβ and IL1β driving the transcriptome of pro-inflammatory synovial fibroblasts in RA but not PsA [22] . Concatenation of scRNAseq and ATACseq for epigenomic regulation, revealed involvement of the NFkβ pathway in the expansion of proinflammatory synovial fibroblasts in a TNFα driven murine model of RA [59] . TNF mediated activation of synovial fibroblasts leads to distinct, joint specific signatures of chromatin rearrangement that may contribute to disease pathogenesis [60] .

The availability of scRNAseq data has also generated new outputs from retrospective examination of synovial tissue bulk RNAseq. Following integration of scRNAseq datasets, the transcriptome of potentially protective synovial HBEGF + fibroblasts was identified and utilised to perform de-convolution of bulk RNAseq [61] . The additional patient samples at different disease and treatment stages showed that the HBEGF fibroblast associated signature was elevated in patients with RA during remission [61] . A similar approach of scRNAseq data integration followed by bulk RNaseq deconvolution resulted in the assignment of distinct synovial fibroblast phenotypes to synovial pathotypes with correlations to clinical features further highlighting the heterogeneity of the cellular landscape and the contribution of immune-stromal cell interaction in synovial inflammation pathogenesis [62] . The need for multiomic approaches of peripheral blood and synovial tissue samples at different stages of disease is exemplified by the observed differential enrichment in signalling pathways in peripheral blood compared to synovial tissue, underlying the difficulty of identifying biomarkers in the serum of patients with IA [57] .

Single cell and whole tissue transcriptomic approaches in combination with established receptor-ligand interaction networks have allowed the bioinformatic interrogation of cell-cell interactions [63] , [64] . However, such studies require in vitro validation as the interactions are based on expression levels of receptors and their ligands. To assist with cell-cell interaction characterisation, spatial transcriptomics can provide spatial resolution of the target tissue, therefore the predicted cell-cell interactions can now be scored based on the physical distance between the potentially interacting cells. Structural organisation and local cell-cell interactions differ in autoantibody positive compared to autoantibody negative RA [65] . Currently, spatial transcriptomic assays are paired with H&E or low resolution histological analysis, with the increasing availability of spatial rnaseq and imaging mass cytometry or multiplexed ion beam imaging, interrogation of cell-cell interactions will become significantly more robust and complex [65] , [66] .

Early applications of in situ spatial transcriptomics in synovial tissue samples of three patients with seropositive and three patients with seronegative RA have indicated a potential dichotomy in receptor-ligand interaction involvement [65] . Evidence of cellular distribution differences were also noted with increased accumulation of macrophages in areas rich with CD55 + fibroblasts in seropositive compared to seronegative RA [65] . Similarly analysis of patient with RA and patient with spondylarthritis (SpA) revealed a relatively small number of differentially expressed genes globally in the tissue with increased expression of SP1 and CD37 in RA compared to SpA [67] . In early arthritis spatial transcriptomics showed B cell – T cell interactions preexisting clinical pathologies [68] . While spatial transcriptomic technologies are rapidly evolving, current commercial platforms of sequencing based spatial transcriptomics do not achieve single cell level resolution or similar sequencing depth as scRNAseq (reviewed here [69] ). Until emerging technologies (example: Stereo-Seq) with single cell resolution become commercially available spatial transcriptomic data require deconvolution based on the integration of scRNAseq [70] . Early spatial transcriptomic studies in RA have utilised non-paired scRNAseq data from previous studies, paired scRNAseq with spatial transcriptomics may improve cellular deconvolution. Alternatively, multiplexing with mass imaging or imaging based spatial transcriptomics may significantly improve resolution [69] , [71] .

Multiomic data integration

IA is the result of complex interplay between environmental, genetic and immunological factors. As a result, single omic approaches are not adequate to provide the necessary resolution required in order to identify the molecular mechanisms of synovial inflammation pathogenesis even under heavily optimised cell clustering and annotation conditions. In addition, combining multiple omic datasets may exhibit synergistic effects for a more accurate characterisation of cellular heterogeneity at the inflamed joint and support the identification of cell-cell interactions [72] . Combining multiple datasets, even in its simplest form of combining omic datasets of the same type, presents significant challenges related to batch effects of a technical nature or true reflections of biological heterogeneity and temporal changes during the course of the disease. For an example, recent studies show that synovial fibroblasts from metatarsophalangeal and larger joints maintain positional identity through distinct epigenetic control of HOX genes resulting in differential involvement of inflammatory pathways, including the TNF pathway [73] . Analysis of combined omic datasets is additionally important for the identification of predictors of response to treatment and the characterisation of dynamic cell-cell interaction networks that may be significantly susceptible to temporal changes [38] . A limitation of combining multi-omic datasets is the increase in number of features relevant to the small number of cells available. As a result, the concatenation of large omic datasets at an early stage is challenging [72] . ML approaches can combine omic datasets in layers and utilise all or a specific number/combination of layers in data analysis, depending on the task however, currently available ML packages have significantly variable success rates [72] . In a recent benchmarking study of 15 currently available ML packages with the capacity to combine multiple omic datasets, supervised ML package DIABLO showed superior accuracy in its ability to predict patient classification whereas, in predicting response to treatment all packages tested showed similar capacity with general purpose ML package moCluster outperforming the other approaches [72] , [74] , [75] . Despite an increasing availability in ML packages capable to integrate multiple omic datasets several key challenges remain. Better ways to control for technical variations ad batch effects are lacking and inconsistences in data storage and format may prevent wider integration of omic data. Accessibility of ML packages to biologists with no bioinformatic background limits the analysis and interpretation of omic data, therefore parallel to the increase in available tools an increase in user friendly interfaces that do not require command line input would significantly increase accessibility and accelerate the implementation of ML in omic data analysis.

Interrogation of cell metabolism on a single cell level

Autofluorescence imaging (AFI) techniques utilise the natural autofluorescence characteristics of proteins and cellular components in order to identify different protein or cell states [76] . One of the more wide-spread AFI is fluorescence lifetime imaging microscopy (FLIM). FLIM is a fluorescence microscopy technique capable of measuring the fluorescence intensity and the fluorescence lifetime of fluorophores.NAD(P)H emits fluorescence with a peak emission at 460 nm when excited with wavelengths of 340 ± 30 nm. Although the excitation and emission spectra of protein bound or free NAD(P)H remain almost identical, NAD(P)H has a defined fluorescence lifetime which allows for microenvironment and conformational state inquires such as folded (free) or extended (bound) state of NAD(P)H [77] , [78] , [79] . The biological significance of the two different states of NAD(P)H is related to the metabolic profile of the cell, with higher fraction of free NADH indicating a more glycolytic profile while a higher fraction of protein bound NAD(P)H being indicative of reliance on oxidative phosphorylation (OXPHOS) [80] .

Similarly, the autofluorescence characteristics of NADH and NADPH are identical and separating the two moieties has been challenging, however studies have identified observable differences in their respective lifetimes by using a mathematical model allowing for the quantification of NADH and NADPH and the observation of highly enriched in NADPH cells with implications for response to treatment [81] , [82] . The distinct advantages of FLIM for the interrogation of a cell’s metabolic profile include the absence of exogenous factors (label-free) such as fluorescently labelled antibodies and the single cell and subcellular level resolution that can be achieved [83] . Due to the aforementioned advantages, FLIM has been utilised to assess the metabolic profile of whole tissue biopsies in vivo and ex vivo since no single cell suspension is required, therefore the tissue can remain intact. The distinct benefit that this offer is the possibility of spatial interrogation of metabolism in the tissue and intraoperative utilisation of FLIM. A key difficulty in brain tumour resection is the accurate identification of the tumour from surrounding healthy tissue [84] . As surgical intervention and maximal tumour resection are the most significant indicators of good prognosis, the implementation of intraoperative microscopy including FLIM for the identification of areas of high glycolytic activity, indicative of the presence of cancer cells and areas of lower metabolic activity indicative of surrounding healthy tissue, could improve tumour resection capacity [84] . FLIM based interrogation of cellular metabolic profile can provide further biological information when combined with machine learning algorithms. The application of machine learning algorithms with cell-based FLIM data can inform the label-free classification of functionally distinct cell subsets [85] . In a recent study, Neto et al., identified metabolic signatures of monocyte derived macrophages polarised under M1 or M2 conditions. Importantly, with the acquired FLIM data, a training dataset was created established and used to train a random forest classification machine learning model with high accuracy and sensitivity. This random forest model is able to successfully identify macrophage subsets based on their metabolic FLIM variables without prior knowledge of the polarisation conditions that were utilised [85] .

In IA, FLIM has been utilised for the characterisation of pathogenic synovial B cells and highly proinflammatory T cell subsets in patients with RA and individuals at risk. Beyond their capacity to produce autoantibodies, B cells are key contributors to synovial inflammation by contributing chemokines, pro-inflammatory cytokines, perform antigen presentation and focusing the immune response [86] , [87] , [88] , [89] . In a recent study Floudas et al., identified significant accumulation of PD-1 expressing B cells in the inflamed synovial tissue of patients with RA. Importantly, PD-1 B cells were resistant to hypoxic conditions that simulated the inflamed joint, produced high amounts of TNF-α and IL-1β and showed increased T cell co-stimulatory capacity compared to their PD-1 negative counterparts [12] . FLIM based interrogation of the metabolic profile of RA patient PD-1 B cells showed a strongly glycolytic profile, due to the low frequency of these cells, it would have been very difficult to use other approaches in order to characterise their metabolic state [12] . Similarly, FLIM was utilised to identify the metabolic state of highly pathogenic, polyfunctional T cells that correlate with disease progression in RA and are present in the synovial tissue of individuals at risk [43] .

FLIM allows non-invasive imaging based on NAD(P)H fluorescence providing information regarding the cell’s metabolic state. This approach can be pursued in both in vitro and in vivo conditions to elucidate immunometabolic research questions in rheumatic diseases. An interesting approach is to combine FLIM intraoperatively with machine learning models for cellular classification. This offers exciting possibilities for label-free identification of inflammation “hot-spots” in the synovial tissue during minimally invasive arthroscopic surgery for targeted-localised therapeutic intervention.

While it can provide information on a single cell level in real time on live cells, one of the caveats of FLIM, is the reliance on a small number of auto-fluorescent molecules and the lack of information on the overall metabolome of the cell. Currently there is a very limited availability of techniques for single cell level metabolic analysis [90] . Recently, differential gene expression and pathway enrichment analysis from scRNAseq data has been utilised to assess the activity of metabolic pathways [91] . Ultimately however, this approach exacerbates the shortcomings of pathway enrichment analysis, primarily the fact that pathway level enrichment scores utilise expression of all genes in the pathway without weighed considerations for intermediate genes in the pathway that may bottleneck the activity of the entire pathway. One exciting prospect is the incorporation of flux balance analysis (FBA) in pathway enrichment analysis when interrogating cell metabolism [92] . FBA takes in to account the kinetic characteristics of chemical reactions along metabolic pathways in the form of stoichiometric coefficients that set limits on each step of a metabolic pathway [92] . Implementation of this approach with scRNAseq pathway enrichment analysis has not only confirmed the metabolic switch of pathogenic Th17 cells towards glycolysis but also identified previously unappreciated polyamine metabolism pathway activity as a characteristic of Th17 pathogenicity [93] . For quantitative analysis of metabolites on a single cell level, single cell mass spectrometry offers exciting prospects once key limitations which include sample preparation, ionizing technique and detection levels have been overcome [94] . The emergence of mass imaging and recent advances for improved resolution may lead to significant synergy with transcriptomics in single cell metabolic profiling [95] .

Potential for utilisation of Biophotonics in IA research

Single cell transcriptomics have revolutionised the study of gene expression and have led to new insights into cell heterogeneity and disease pathology [96] . In a similar manner, biophotonics is an emerging interdisciplinary field that combines the principles of physics, chemistry, biology, and engineering to study the interaction between light and biological systems [97] . The term "biophotonics" was coined in the late 1990 s to describe the use of light-based technologies in biological research and medical applications [98] . Since then, it has grown rapidly as a field, with applications ranging from basic research to clinical practice. Biophotonics encompasses a range of techniques, including spectroscopy, imaging and microscopy, that facilitates the interrogation of biological samples at the cellular and molecular level with high sensitivity and resolution and provides new insights into cellular heterogeneity, diversity and disease pathology [99] . Through the use of light-based technologies to explore and interrogate biological tissues, cells, and molecules, new diagnostic and therapeutic tools for medicine and biology have emerged [100] .

Several different classes of label-free sensor detectors have been developed including plasmonic, electrical, mechanical, and photonic (light) sensors [101] . Light as a diagnostic and prognostic tool has several potential advantages including high sensitivity, non-destructive measurement, non-invasive analysis and low limits of detection [101] . The innate optical response comprises scattering, absorbance, and auto-fluorescence signals that are closely associated with metabolism [102] and structural proteins [103] under normal, altered or pathological conditions, while fluorescence has been proposed as a non-invasive indicator of embryonic stem cell differentiation state [104] . In combination with microfluidics, photonics enables real time measurement of single cells in very small sample volumes [105] . In this context, label-free photonic technologies with highly efficient cell-to-light coupling have been developed and successfully deployed to measure the real-time response of individual cells in a population [106] .

In an inflamed joint with RA, hyperplasia of mesenchymal stromal cells (MSCs) and infiltration of inflammatory cells into the synovium causes localised pathophysiological changes consequently altering the photonic properties compared to healthy joints [107] . There is an ongoing debate about the relationship between MSCs and resident fibroblasts and their role in RA pathogenesis. The synovial microenvironment is critical to the progression of pre-RA to RA and harbours fibroblasts, macrophages and MSCs where synovial fibroblast-like cells (FLS) are known to promote pathogenic events leading to synovitis. FLS are thought to promote synovitis by interacting with MSCs to induce hyperplasia and these cells migrate from bone marrow under the influence of certain cytokines or in response to injury. While there is ambiguity around the relationship between the MSCs and FLS in synovial microenvironment, their contrasting in vitro properties suggest the possibility that FLS and synovial MSCs are one in the same but reflect different stages of differentiation. Similar characteristics between both cell types in terms of surface morphology and markers has led many researchers to speculate that FLS are mature MSCs whereby, with advancing age, the synovial microenvironment changes to harbour less plastic MSCs and more differentiated fibroblasts [108] . Miniaturisation of the biophotonic platforms for in vitro single cell interrogation of laser induced fluorescence may facilitate in vivo detection of RA disease phenotypes using supervised machine learning. Optical tweezers can be coupled with nanophotonic biosensor devices based on integrated fibre optics and microfluidics devices for the implementation of Lab-on-a-Chip platforms to reveal photonic differences that act as a powerful discriminator of cell phenotype within the synovial microenvironment, as has been implemented for other pathologies [99] , [106] . These platforms can also probe complex biophysical and biomechanical processes governing cell-cell interactions, cell – surface interactions, cell sorting and drug delivery/testing [109] . Previous studies have also used a multimodality imaging approach to distinguish RA patients from healthy controls [110] . The cellular heterogeneity within articular cartilage has also been revealed using novel optical techniques, including quantitative polarized light microscopy (qPLM), Fourier-transform infrared (FTIR) microscopy, Raman and optical hyperspectral reflectance and fluorescence imaging [111] .

Taken together, biophotonics and single cell transcriptomics are complimentary technology platforms that can be used to study the structure and function of individual cells and tissues and their origin in far great detail [112] . Biophotonics can be used to visualise the localization and dynamics of specific molecules within cells that informs cellular heterogeneity, diversity and disease pathology, while single cell transcriptomics can be used to identify the specific genes that are active in specific cell types or during specific biological processes that lead to disease progression. The combination of these two techniques is particularly powerful for understanding complex biological systems and for developing new therapeutic strategies [113] . This integration of technologies provides a deeper understanding of complex biological processes, with potential applications in fields such as cancer research, immunology, and regenerative medicine. While biophotonic platforms have not been widely utilised in IA research to date, the potential combination of label free biophotonics with FLIM and spatial transcriptomics can act synergistically to greatly enhance our understanding of cell-cell interactions in the pathogenesis of synovial inflammation and better model the cellular landscape at the site of injury.

Implementation of machine learning in the study of IA

Machine learning approaches are derived from the need to determine pattern recognition coupled with the development of computational learning theory in artificial intelligence. In particular, machine learning is rooted by algorithms that learn from a large dataset and makes prediction on the data [114] , [115] . Recent advances in computing, biochemical and imaging platforms in biomedical research has resulted in a rise in this new research dimension [114] . Large datasets can be obtained from a single experiment through which the degree of complexity increases with a higher number of dependent variables and conditions. Therefore, the increase of experimental data acquisition and its complexity calls for the need of precise machine learning-based data mining algorithms. The machine learning assignments are divided in two categories: unsupervised and supervised learning. Unsupervised learning algorithms are linked to clustering of data whilst supervised machine learning algorithms are committed to the prediction of a categorical outcome or classification [114] . Unsupervised machine learning algorithms have the capacity to detect patterns in experimental data without the need for user supervision or labelling. Due to their independence from labelled or descriptive information, unsupervised machine learning techniques offer a simpler implementation and are better suited for exploratory analysis purposes [116] . Supervised machine learning methods are utilized to predict or classify a targeted output of interest. These techniques are typically employed in prediction tasks where the goal is to anticipate or classify a particular outcome of interest [117] .

Machine learning approaches can be utilised in all facets of research in IA pathogenesis and treatment from improving bulk and scRNAseq data resolution, allowing for integrated analysis of multiple -omic datasets to informing therapeutic target identification and patient stratification based on available clinical and molecular data ( Fig. 1 ). Therefore, the implementation of ML approaches has the potential to transform clinical practice by accelerating drug discovery, improving patient stratification and predicting response to treatment. Importantly, the herein discussed implementation of ML are applicable in a wider range of diseases.

A limitation of RNAseq analysis is that it is a sub-sampling process based on certain assumptions that are not always true. For an example, during RNAseq reads are generated from DNA fragments of transcribed exons with the assumption that distribution of reads is even, therefore the number of reads is a direct reflection of the number of copies of the corresponding transcript. In reality however, read distribution is uneven leading to the introduction of technical bias on a local or global scale [118] . Pseudoalignment approaches between a reference dataset and reads, in order to identify gene transcripts, have increased speed and reduced the computational resources needed for alignment but have not solved the uneven distribution of reads [119] . ML based imputation of missing reads could alleviate this problem however, there is a very limited number of algorithms developed to perform read imputation [120] . Following the implementation of quality control measures, a key step in scRNAseq data analysis is the determination of cell clusters and their assignment to specific cell types [121] . Currently there is no unifying guidelines or process for the identification of the optimal number of cell clusters as a result the number of identified clusters is influenced by sequencing depth, number of cells in the dataset, the actual number of distinct cell types, the analysis platform used, the relative performance of clustering algorithms, prior knowledge of expected cell populations with further confoundment caused by the lack of benchmarking tools and asynchronous evolution of clustering algorithms [122] . Once clustering has been performed, differentially expressed genes (DEG) are identified, cluster specific DEG are then utilised for cluster annotation, the process of assigning specific cell types to the identified clusters. DEG identification can be significantly influenced by technical sources of variation due to the introduction of “dropouts”. Dropout is a term that refers to a feature that is artificially identified at a lower abundance than its actual expression. This can be a result of poor sampling during library preparation, suboptimal reverse transcription to DNA or poor PCR mediated amplification resulting in only a fraction of a cell’s transcriptome being captured for each cell [123] . DEG identification and restoring the expression of dropouts are two more areas where ML approaches can improve RNAseq analysis. Implementation of a ML model of logistic regression classification predicted a significant number of DEG that were not previously reported [124] . In addition to improving DEG identification, ML can be utilised for data imputation in order to minimise the effect of dropouts. While several “conventional” imputation methods have become available, they are computationally demanding and can introduce unexpected side-effects by altering the relative expression values of the entire transcriptome of the cell thus reducing the separation of distinct but closely related cell clusters [125] . ML based approaches could be computationally less demanding and simultaneously preserve transcriptomic differences between cells therefore, allowing for subpopulation identification and trajectory analysis, however further benchmarking studies are required before their introduction as part of routine pipelines of scRNAseq analysis [126] .

Following the identification of the optimal number of clusters under these considerations, the resulting clusters require annotation. Currently available automated annotation tools primarily use reference datasets from RNAseq analysis of known flow cell sorted populations or previously well annotated scRNAseq datasets. DEG markers of the different cell clusters are compared to the transcriptomic profile of the reference in order to perform the classification of the unknown clusters [127] . The success rate of this approach is dependent on the array of cells and tissue types available as part of the reference [127] , [128] . As a result, automated cluster annotation tools, depending on the sequencing depth, tissue of origin and cluster resolution of the data fail to capture cell populations. In those cases researchers revert to manual supervised annotation of clusters based on literature searches and prior knowledge of cell type specific markers introducing the risk of biasing annotations [129] . An added difficulty in applying cell annotations is the plasticity and transitory states that cells may exhibit especially under inflammatory conditions. Recently available ML pipeline devCellPy may be able to automate cell annotations and reduce bias. devCellPy utilises a hierarchically layered decision tree in order to predict cell labels, with this approach cells are first assigned in broader, “low” resolution subsets followed by layers of cell subtype specific annotations therefore, reducing the influence that transient cell states exhibit on cell annotations [130] .

ML and deep learning, a subset of machine learning based primarily on artificial neural network geometries, has rapidly grown as a predictive method to obtain real-time bio-photonic decision-making systems and analyse biophotonic and imaging data, in particular, spectroscopic data including spectral data pre-processing and spectral classification [131] . One specific application of supervised machine learning is in the field of bioimaging. Machine learning algorithms can be trained to recognise and analyse patterns in large volumes of imaging data, allowing for the automated identification of specific cell types, tissues, or disease states [132] . This can help to gain a better understanding of the underlying mechanisms of disease, and to develop new diagnostic and therapeutic approaches.

In a similar manner, supervised machine learning can be used to analyse large amounts of spectral data generated by light-based imaging techniques, such as fluorescence microscopy [133] , [134] , RAMAN spectroscopy [135] , confocal microscopy [136] , and optical coherence tomography [137] . By training a machine learning model on labelled data, it is possible to classify different types of cells, define cellular heterogeneity within tissues and organs based on their photonic properties [99] and to identify disease biomarkers and potential therapeutic targets.

In addition to the benefits of ML in omic and imaging data analysis and research there is tremendous potential in more practical applications that include diagnosis, patient stratification and identification of optimal treatment. In IA due to heterogeneity in clinical manifestations within a disease and potential similarities between related conditions, diagnosis can be challenging and may be subject to revision at follow up examinations [138] . ML approaches have been used for the stratification of patients in cancer and heart disease, however, there is a lack of studies in ML application for stratification of patients with IA [139] , [140] , [141] , [142] , [143] . Due to the heterogeneous nature of IA and similarities of clinical manifestations between related IAs diagnosis can be challenging resulting on significant impact on clinical practice and therapeutic approach [144] . Importantly, delays in IA treatment will negatively impact the probability of achieving lasting remission [145] . Therefore, implementation of ML for diagnostic and stratification purposes has the potential to reduce delays in treatment, associated costs and improve the patient’s quality of life by achieving early, lasting remission through refinement of the patient stratification process. Learning from the implementation of ML in other clinical settings, ML models were able to stratify, with high degree of accuracy, intensive care patients at risk of developing heart failure [139] . Such approaches can enable early monitoring of patients at risk and reduce mortality. Arthralgia subjects are individuals at risk (IAR) of developing arthritis. IAR experience joint pain but no canonical clinical symptoms of arthritis or synovitis [50] . Recent studies have identified dysregulated inflammatory pathways and increase in pathogenic polyfunctional T cells in the joint of IAR [43] . Similarly, stratification of individuals at risk of arthritis would facilitate early prophylactic disease-modifying anti-rheumatic drugs (DMARD) treatment of IAR leading to extended symptom free period prior to potential transition to RA and less severe disease progression and joint damage. In an RA patient cohort of 1204 patients Koo et al., utilised five ML models on several clinical features in order to predict disease remission following DMARD treatment [143] . Accuracy of the ML models varied between 50% and 75%, interestingly, the dominant clinical characteristics that determined response to treatment differ depending on DMARD used [143] .

Current limitations of ML model applications in omic data analysis and patient stratification

With the aggregation of readily available omic data accessible through public repositories (such as the National Center for Biotechnology Information (NCBI)), increased computing power and incremental re-focusing of healthcare towards personalised medicine, there is an increasing requirement for ML model applications [146] . As outlined in Table 1 , the efficacy of ML models is heavily impacted by the available training data ( Table 1 ). The majority of studies applying ML models in the omic space, utilise transcriptomic data which creates unique challenges [147] . Especially single-cell transcriptomics (scRNAseq, scATACseq) are examining datasets with a very high number of variables but relatively low number of samples [147] . As a result, dimensionality reduction is almost a mandatory first step for downstream analysis and ML applications [148] . Most commonly used method for dimensionality reduction of omic data is principal component analysis (PCA), however, PCA is not without shortcomings, most notably the generation of artifacts associated with missing values [149] . The impact of dimensionality reduction on ML models is further exacerbated when multiple omic datasets have been integrated. Transcriptomic, proteomic, metabolomic data even when matched, require different normalisation and scaling methods leading to differences in range, detection levels and outliers and may even require data imputation, hence adding another layer of variability [148] , [150] , [151] . Unsurprisingly, even after strict data preparation, the performance of different ML models can vary significantly in classification capacity as recently shown in predicting breast cancer metastasis based on transcriptomic data [152] . These considerations have limited ML implementation in a clinical setting and have been detrimental in the efforts to translate omic data to clinical practice. One significant aspect that is not frequently discussed when ML models are used to predict patient responses is whether the model can act independently of and/or outperform the physician. In a recent study comparing the predictive capacity of ML using either clinician initiated data from non-routine procedures or non-clinician initiated data from routine tests, on patients with myocardial infractions, ML performance was comparable, raising concerns regarding the utilisation of ML in the clinic [153] . With the increasing availability of omic data and ML models of patient stratification, there is an urgent, unmet need for unifying guidelines that will set minimum requirements for the reporting of ML models. These guidelines should be in addition to strict regulations regarding the availability of omic data in public repositories at the time of first publication and in formats that will provide ease of access for further analysis and are supplemented with all the available clinical data [154] . Due to the rapid progress in omics and ML models, data sharing standards should be updated more frequently and be accompanied by a universal “roadmap” for good practises when designing and reporting projects involving omic data and ML model applications.

Discussion

IA research is at an exciting precipice in the ML driven synergistic combination of omic approaches with multidimensional flow cytometric analysis and label free imaging in order to model the cellular landscape and molecular pathways involved in synovial inflammation. A short overview of benefits and limitations of transcriptomic and photonic techniques with applications in IA is included in Table 2 ( Table 2 ). In conjunction with precision medicine, personalised medicine utilising easily accessible clinical data and ML packages will make patient stratification relevant and improve disease progression. The efficacious and “safe” implementation of omic and clinical data driven ML models will require updated clinical practices and strong multidisciplinary collaborations between bioinformaticians, immunologists and clinical rheumatologists.

However, currently, with some notable exceptions, the majority of omic studies in IA appear to have a confirmatory effect regarding the involvement of specific molecular pathways that have already been suspected to play a role in disease pathogenesis rather than forward-leap our understating of the underlying immunological mechanisms [4] , [38] , [73] . As we are in the dawn of omic data analysis, we may have been “searching” for the familiar, unwillingly biasing hypothesis during omic data analysis. With increasing understanding of the limitations of omic data, however controversial, implementation of hypothesis-free approaches might lead to novel discoveries [167] . This is a highly evolving field and as new analytical tools become available and caveats of multi-omic data integration are overcome, secondary data analysis will become increasingly important and fuel future studies.

Another key consideration is that for ML models to be meaningful and not a distraction in clinical practice updated standards are needed across all facets of omic data analysis and ML model implementation and reporting [153] . Setting new standards now is timely as there is a significant increase in the number of studies performing new or re-analysing existing omic datasets [168] , [169] . With the increased availability of omic data new challenges have arisen that include the integration of multiple datasets for secondary analysis and the curation of large, open access data repositories. Importantly, data repositories such as NCBI Gene Expression Omnibus (GEO) are exceeding their original purpose as archives and are tasked with providing data quality assurances in order to enable secondary data analysis [168] . While some of these challenges could be facilitated with increased availability of curated data repositories such as Gemma, “equipped” with standardised tools for secondary data analysis there are needs for long term computational and funding support, therefore data storage and re-use needs to be included in all strategic planning in relation to omic data generation. Pre-processing of data and the identification of key features for ML model application has a significant influence on outcome. One of the areas for urgent establishment of universal standards is dimensionality reduction and data integration. Due to the nature of single cell transcriptomics especially, the statistical problems arising from the high number of variables compared to samples are going to remain and be further compiled with increasing demand for multi-omic data integration [147] , [151] .

Despite forward leaps in omic data availability, translation of omic studies to the clinic has been slow leading to a new bottleneck in the generation of actionable knowledge. One important parameter is the efficacy of ML models in clinical practice. If gains in predicting capacity are marginal or dependent on clinician initiated actions, they could create confusion and be detrimental for the patient, therefore minimum precision and gain limits should be set for reporting ML models in patient stratification [153] . Fundamentally, in order to bridge the gap between in silico analysis and the clinic, standards are needed for the in vitro interrogation of bioinformatically predicted molecular pathway involvement in disease. Legacy imaging techniques are not able to keep up with the increasing demand for multiplexing several targets during histological analysis. Recently available mass imaging platforms for spatial analysis of complex tissues in association with improvements in photonics and super resolution microscopy, offering molecular level resolution of protein-protein interactions have the potential to increase the translatability of omic data [170] , [171] .

Interdisciplinarity and collaboration of researchers, clinicians and bioinformaticians especially with ML experience, is increasingly in demand and a robust universal roadmap in analysis and updated standards in reporting are required.

Schematic representation of machine learning assisted omic data integration combining transcriptomic, proteomic and metabolomic approaches with clinical data in order to identify novel therapeutic targets, perform better patient stratification and predict patient’s response to treatment. In addition to scRNAseq and ATAseq, spatial transcriptomics and mass histology can be utilised for the identification of cell-cell interactions and molecular pathways involved in disease pathogenesis. This will enhance patient stratification based on disease pathotypes and enable targeted inhibition of cell-cell interactions depending on the dominant inflammatory pathway involved.

Funding body

This work was funded by the Health Research Board (HRB), under the HRB Emerging Investigator Award HRBEIA 2022–002.

CRediT authorship contribution statement

Ciara Hegarty : Writing – original draft. Nuno Neto : Conceptualisation, Writing – original draft. Paul Cahill : Conceptualisation, Writing – original draft. Achilleas Floudas Supervision, Writing – review & editing.

Declaration of Competing Interest

The authors of manuscript titled “Computational approaches in rheumatic diseases – deciphering complex spatio-temporal cell interactions. “ report no conflict of interest.

References

1. Rheumatoid arthritis | Nature Reviews Disease Primers [Internet]. [cited 2023 Mar 3]. Available from: https://www.nature.com/articles/nrdp20181.

2. Taylor P.C., Atzeni F., Balsa A., Gossec L., Müller-Ladner U., Pope J. The key comorbidities in patients with rheumatoid arthritis: a narrative review. J Clin Med. 2021;10(3):509. doi: 10.3390/jcm10030509.

3. Lewis M.J., Barnes M.R., Blighe K., Goldmann K., Rana S., Hackney J.A., et al. Molecular portraits of early rheumatoid arthritis identify clinical and treatment response phenotypes. Cell Rep. 2019;28(9):2455–2470.e5. doi: 10.1016/j.celrep.2019.07.091.

4. Armaka M., Konstantopoulos D., Tzaferis C., Lavigne M.D., Sakkou M., Liakos A., et al. Single-cell multimodal analysis identifies common regulatory programs in synovial fibroblasts of rheumatoid arthritis patients and modeled TNF-driven arthritis. Genome Med. 2022;14(1):78. doi: 10.1186/s13073-022-01081-3.

5. van der Heijde D., Klareskog L., Rodriguez-Valverde V., Codreanu C., Bolosiu H., Melo-Gomes J., et al. Comparison of etanercept and methotrexate, alone and combined, in the treatment of rheumatoid arthritis: two-year clinical and radiographic results from the TEMPO study, a double-blind, randomized trial. Arthritis Rheum. 2006;54(4):1063–1074. doi: 10.1002/art.21655.

6. Buch M.H. Defining refractory rheumatoid arthritis. Ann Rheum Dis. 2018;77(7):966–969. doi: 10.1136/annrheumdis-2017-212862.

7. Humby F., Mahto A., Ahmed M., Barr A., Kelly S., Buch M., et al. The relationship between synovial pathobiology and magnetic resonance imaging abnormalities in rheumatoid arthritis: a systematic review. J Rheuma. 2017;44(9):1311–1324. doi: 10.3899/jrheum.161314.

8. Paleolog E.M. The vasculature in rheumatoid arthritis: cause or consequence? Int J Exp Pathol. 2009;90(3):249–261. doi: 10.1111/j.1365-2613.2009.00640.x.

9. Malone D.G., Wahl S.M., Tsokos M., Cattell H., Decker J.L., Wilder R.L. Immune function in severe, active rheumatoid arthritis. A relationship between peripheral blood mononuclear cell proliferation to soluble antigens and synovial tissue immunohistologic characteristics. J Clin Invest. 1984;74(4):1173–1185. doi: 10.1172/JCI111526.

10. Mohr W., Beneke G., Mohing W. Proliferation of synovial lining cells and fibroblasts. Ann Rheum Dis. 1975;34(3):219–224. doi: 10.1136/ard.34.3.219.

11. Biniecka M., Connolly M., Gao W., Ng C.T., Balogh E., Gogarty M., et al. Redox-mediated angiogenesis in the hypoxic joint of inflammatory arthritis. Arthritis Rheuma Hoboken NJ. 2014;66(12):3300–3310. doi: 10.1002/art.38822.

12. Floudas A., Neto N., Marzaioli V., Murray K., Moran B., Monaghan M.G., et al. Pathogenic, glycolytic PD-1+ B cells accumulate in the hypoxic RA joint. JCI Insight. 2020 doi: 10.1172/jci.insight.139032. cited 2023 Mar 7];5(21). Available from: https://insight.jci.org/articles/view/139032.

13. van de Sande M.G., Baeten D.L. Immunopathology of synovitis: from histology to molecular pathways. Rheumatology. 2016 1;55(4):599–606. doi: 10.1093/rheumatology/kev330.

14. Ingegnoli F., Coletto L.A., Scotti I., Compagnoni R., Randelli P.S., Caporali R. The crucial questions on synovial biopsy: when, why, who, what, where, and how? Front Med. 2021 doi: 10.3389/fmed.2021.705382. cited 2023 Mar 16];8. Available from: https://www.frontiersin.org/articles/10.3389/fmed.2021.705382.

15. Kruithof E., Baeten D., De Rycke L., Vandooren B., Foell D., Roth J., et al. Synovial histopathology of psoriatic arthritis, both oligo- and polyarticular, resembles spondyloarthropathy more than it does rheumatoid arthritis. Arthritis Res Ther. 2005;7(3):R569–R580. doi: 10.1186/ar1698.

16. Baeten D., Demetter P., Cuvelier C., Van den Bosch F., Kruithof E., Van Damme N., et al. Comparative study of the synovial histology in rheumatoid arthritis, spondyloarthropathy, and osteoarthritis: influence of disease duration and activity. Ann Rheum Dis. 2000;59(12):945–953. doi: 10.1136/ard.59.12.945.

17. Krenn V., Morawietz L., Burmester G.R., Kinne R.W., Mueller-Ladner U., Muller B., et al. Synovitis score: discrimination between chronic low-grade and high-grade synovitis. Histopathology. 2006;49(4):358–364. doi: 10.1111/j.1365-2559.2006.02508.x.

18. Schmidt T., Najm A., Mussawy H., Burghardt R., Oehler N., Krenn V., et al. General synovitis score and immunologic synovitis score reflect clinical disease activity in patients with advanced stage rheumatoid arthritis. Sci Rep. 2019;9(1) doi: 10.1038/s41598-019-44895-9.

19. Scott D.L., Wolfe F., Huizinga T.W.J. Rheumatoid arthritis. Lancet Lond Engl. 2010;376(9746):1094–1108. doi: 10.1016/S0140-6736(10)60826-4.

20. Steiner G. Cytokine production by synovial T cells in rheumatoid arthritis. Rheumatology. 1999;38(3):202–213. doi: 10.1093/rheumatology/38.3.202.

21. Bresnihan B., Tak P.P. Synovial tissue analysis in rheumatoid arthritis. Baillieres Best Pr Res Clin Rheuma. 1999;13(4):645–659. doi: 10.1053/berh.1999.0051.

22. Floudas A., Smith C.M., Tynan O., Neto N., Krishna V., Wade S.M., et al. Distinct stromal and immune cell interactions shape the pathogenesis of rheumatoid and psoriatic arthritis. Ann Rheum Dis. 2022:2021–221761. doi: 10.1136/annrheumdis-2021-221761. (annrheumdis)

23. Grivas A., Fragoulis G., Garantziotis P., Banos A., Nikiphorou E., Boumpas D. Unraveling the complexities of psoriatic arthritis by the use of -Omics and their relevance for clinical care. Autoimmun Rev. 2021;20(11) doi: 10.1016/j.autrev.2021.102949.

24. Cantatore F.P., Maruotti N., Corrado A., Ribatti D. Angiogenesis dysregulation in psoriatic arthritis: molecular mechanisms. BioMed Res Int. 2017;2017:5312813. doi: 10.1155/2017/5312813.

25. Fromm S., Cunningham C.C., Dunne M.R., Veale D.J., Fearon U., Wade S.M. Enhanced angiogenic function in response to fibroblasts from psoriatic arthritis synovium compared to rheumatoid arthritis. Arthritis Res Ther. 2019;21(1):297. doi: 10.1186/s13075-019-2088-3. 21.

26. Wijbrandts C.A., Tak P.P. Prediction of response to targeted treatment in rheumatoid arthritis. Mayo Clin Proc. 2017;92(7):1129–1143. doi: 10.1016/j.mayocp.2017.05.009.

27. Puentes-Osorio Y., Amariles P., Calleja M.Á., Merino V., Díaz-Coronado J.C., Taborda D. Potential clinical biomarkers in rheumatoid arthritis with an omic approach. Autoimmun Highlights. 2021;12(1) doi: 10.1186/s13317-021-00152-6.

28. Dennis G., Holweg C.T.J., Kummerfeld S.K., Choy D.F., Setiadi A.F., Hackney J.A., et al. Synovial phenotypes in rheumatoid arthritis correlate with response to biologic therapeutics. Arthritis Res Ther. 2014;16(2):R90. doi: 10.1186/ar4555.

29. Humby F., Lewis M., Ramamoorthi N., Hackney J.A., Barnes M.R., Bombardieri M., et al. Synovial cellular and molecular signatures stratify clinical response to csDMARD therapy and predict radiographic progression in early rheumatoid arthritis patients. Ann Rheum Dis. 2019;78(6):761–772. doi: 10.1136/annrheumdis-2018-214539.

30. Mussawy H., Zustin J., Luebke A.M., Strahl A., Krenn V., Rüther W., et al. The histopathological synovitis score is influenced by biopsy location in patients with knee osteoarthritis. Arch Orthop Trauma Surg. 2022;142(11):2991–2997. doi: 10.1007/s00402-021-03889-x.

31. Hayer S., Vervoordeldonk M.J., Denis M.C., Armaka M., Hoffmann M., Bäcklund J., et al. SMASH’ recommendations for standardised microscopic arthritis scoring of histological sections from inflammatory arthritis animal models. Ann Rheum Dis. 2021;80(6):714–726. doi: 10.1136/annrheumdis-2020-219247.

32. Smith M.D., Baeten D., Ulfgren A.K., McInnes I.B., Fitzgerald O., Bresnihan B., et al. Standardisation of synovial tissue infiltrate analysis: how far have we come? how much further do we need to go? Ann Rheum Dis. 2006;65(1):93–100. doi: 10.1136/ard.2005.036905.

33. Culemann S., Grüneboom A., Nicolás-Ávila J.Á., Weidner D., Lämmle K.F., Rothe T., et al. Locally renewing resident synovial macrophages provide a protective barrier for the joint. Nature. 2019;572(7771):670–675. doi: 10.1038/s41586-019-1471-1.

34. Schraivogel D., Kuhn T.M., Rauscher B., Rodríguez-Martínez M., Paulsen M., Owsley K., et al. High-speed fluorescence image-enabled cell sorting. Science. 2022;375(6578):315–320. doi: 10.1126/science.abj3013.

35. Hartmann F.J., Bendall S.C. Immune monitoring using mass cytometry and related high-dimensional imaging approaches. Nat Rev Rheuma. 2020;16(2):87–99. doi: 10.1038/s41584-019-0338-z.

36. Eerola E., Pulkki K., Pelliniemi L.J., Granfors K., Vuorio E., Toivanen A. Arthritis-associated changes in flow cytometric characteristics of cultured synovial fibroblasts. Arthritis Rheum. 1988;31(3):339–347. doi: 10.1002/art.1780310305.

37. Zhang F., Wei K., Slowikowski K., Fonseka C.Y., Rao D.A., Kelly S., et al. Defining inflammatory cell states in rheumatoid arthritis joint synovial tissues by integrating single-cell transcriptomics and mass cytometry. Nat Immunol. 2019;20(7):928–942. doi: 10.1038/s41590-019-0378-1.

38. Mizoguchi F., Slowikowski K., Wei K., Marshall J.L., Rao D.A., Chang S.K., et al. Functionally distinct disease-associated fibroblast subsets in rheumatoid arthritis. Nat Commun. 2018;9(1):789. doi: 10.1038/s41467-018-02892-y.

39. Van Landuyt K.B., Jones E.A., McGonagle D., Luyten F.P., Lories R.J. Flow cytometric characterization of freshly isolated and culture expanded human synovial cell populations in patients with chronic arthritis. Arthritis Res Ther. 2010;12(1):R15. doi: 10.1186/ar2916.

40. Floudas A., Gorman A., Neto N., Monaghan M.G., Elliott Z., Fearon U., et al. Inside the joint of inflammatory arthritis patients: handling and processing of synovial tissue biopsies for high throughput analysis. Front Med. 2022 doi: 10.3389/fmed.2022.830998. [cited 2023 Mar 23];9. Available from: https://www.frontiersin.org/articles/10.3389/fmed.2022.830998.

41. Basdeo S.A., Cluxton D., Sulaimani J., Moran B., Canavan M., Orr C., et al. Ex-Th17 (Nonclassical Th1) cells are functionally distinct from classical Th1 and Th17 cells and are not constrained by regulatory T cells. J Immunol Balt Md 1950. 2017;198(6):2249–2259. doi: 10.4049/jimmunol.1600737.

42. Basdeo S.A., Moran B., Cluxton D., Canavan M., McCormick J., Connolly M., et al. Polyfunctional, pathogenic CD161+ Th17 lineage cells are resistant to regulatory T cell-mediated suppression in the context of autoimmunity. J Immunol Balt Md 1950. 2015;195(2):528–540. doi: 10.4049/jimmunol.1402990.

43. Floudas A., Neto N., Orr C., Canavan M., Gallagher P., Hurson C., et al. Loss of balance between protective and pro-inflammatory synovial tissue T-cell polyfunctionality predates clinical onset of rheumatoid arthritis. Ann Rheum Dis. 2022;81(2):193–205. doi: 10.1136/annrheumdis-2021-220458.

44. Wade S.M., Canavan M., McGarry T., Low C., Wade S.C., Mullan R.H., et al. Association of synovial tissue polyfunctional T-cells with DAPSA in psoriatic arthritis. Ann Rheum Dis. 2019;78(3):350–354. doi: 10.1136/annrheumdis-2018-214138.

45. Floudas A., Canavan M., McGarry T., Mullan R., Nagpal S., Veale D.J., et al. ACPA status correlates with differential immune profile in patients with rheumatoid arthritis. Cells. 2021;10(3):647. doi: 10.3390/cells10030647.

46. Morita Y., Yamamura M., Kawashima M., Harada S., Tsuji K., Shibuya K., et al. Flow cytometric single-cell analysis of cytokine production by CD4+ T cells in synovial tissue and peripheral blood from patients with rheumatoid arthritis. Arthritis Rheum. 1998;41(9):1669–1676. doi: 10.1002/1529-0131(199809)41:9<1669::AID-ART19>3.0.CO;2-G.

47. Menon B., Gullick N.J., Walter G.J., Rajasekhar M., Garrood T., Evans H.G., et al. Interleukin-17+CD8+ T cells are enriched in the joints of patients with psoriatic arthritis and correlate with disease activity and joint damage progression. Arthritis Rheuma Hoboken NJ. 2014;66(5):1272–1281. doi: 10.1002/art.38376.

48. Alivernini S., MacDonald L., Elmesmari A., Finlay S., Tolusso B., Gigante M.R., et al. Distinct synovial tissue macrophage subsets regulate inflammation and remission in rheumatoid arthritis. Nat Med. 2020;26(8):1295–1306. doi: 10.1038/s41591-020-0939-8.

49. Guo Y., Walsh A.M., Fearon U., Smith M.D., Wechalekar M.D., Yin X., et al. CD40L-dependent pathway is active at various stages of rheumatoid arthritis disease progression. J Immunol Balt Md 1950. 2017;198(11):4490–4501. doi: 10.4049/jimmunol.1601988.

50. Guo Y., Walsh A.M., Canavan M., Wechalekar M.D., Cole S., Yin X., et al. Immune checkpoint inhibitor PD-1 pathway is down-regulated in synovium at various stages of rheumatoid arthritis disease progression. PLoS One. 2018;13(2) doi: 10.1371/journal.pone.0192704.

51. Grivas A., Grigoriou M., Malissovas N., Sentis G., Filia A., Flouda S., et al. Combined – whole blood and skin fibroblasts- transcriptomic analysis in Psoriatic Arthritis reveals molecular signatures of activity, resistance and early response to treatment. Front Immunol. 2022;13 doi: 10.3389/fimmu.2022.964274.

52. Sumitomo S., Nagafuchi Y., Tsuchida Y., Tsuchiya H., Ota M., Ishigaki K., et al. A gene module associated with dysregulated TCR signaling pathways in CD4+ T cell subsets in rheumatoid arthritis. J Autoimmun. 2018;89:21–29. doi: 10.1016/j.jaut.2017.11.001.

53. Sellam J., Marion-Thore S., Dumont F., Jacques S., Garchon H.J., Rouanet S., et al. Use of whole-blood transcriptomic profiling to highlight several pathophysiologic pathways associated with response to rituximab in patients with rheumatoid arthritis: data from a randomized, controlled, open-label trial. Arthritis Rheuma. 2014;66(8):2015–2025. doi: 10.1002/art.38671.

54. Aran D., Hu Z., Butte A.J. xCell: digitally portraying the tissue cellular heterogeneity landscape. Genome Biol. 2017;18(1) doi: 10.1186/s13059-017-1349-1.

55. Saito Y., Kagami S.I., Sanayama Y., Ikeda K., Suto A., Kashiwakuma D., et al. AT-rich–interactive domain–containing protein 5A functions as a negative regulator of retinoic acid receptor–related orphan nuclear receptor γt–induced Th17 cell differentiation. Arthritis Rheuma. 2014;66(5):1185–1194. doi: 10.1002/art.38324.

56. Hedman Å.K., Winter E., Yoosuf N., Benita Y., Berg L., Brynedal B., et al. Peripheral blood cellular dynamics of rheumatoid arthritis treatment informs about efficacy of response to disease modifying drugs. Sci Rep. 2023;13(1):10058. doi: 10.1038/s41598-023-36999-0.

57. Lee E.J., Lilja S., Li X., Schäfer S., Zhang H., Benson M. Bulk and single cell transcriptomic data indicate that a dichotomy between inflammatory pathways in peripheral blood and arthritic joints complicates biomarker discovery. Cytokine. 2020;127 doi: 10.1016/j.cyto.2019.154960.

58. Poppenberg K.E., Jiang K., Li L., Sun Y., Meng H., Wallace C.A., et al. The feasibility of developing biomarkers from peripheral blood mononuclear cell RNAseq data in children with juvenile idiopathic arthritis using machine learning approaches. Arthritis Res Ther. 2019;21(1) doi: 10.1186/s13075-019-2010-z.

59. Armaka M., Konstantopoulos D., Tzaferis C., Lavigne M.D., Sakkou M., Liakos A., et al. Single-cell multimodal analysis identifies common regulatory programs in synovial fibroblasts of rheumatoid arthritis patients and modeled TNF-driven arthritis. Genome Med. 2022;14(1) doi: 10.1186/s13073-022-01081-3.

60. Ge X., Frank-Bertoncelj M., Klein K., McGovern A., Kuret T., Houtman M., et al. Functional genomics atlas of synovial fibroblasts defining rheumatoid arthritis heritability. Genome Biol. 2021;22(1):247. doi: 10.1186/s13059-021-02460-6.

61. Chen N., Fan B., He Z., Yu X., Wang J. Identification of HBEGF+ fibroblasts in the remission of rheumatoid arthritis by integrating single-cell RNA sequencing datasets and bulk RNA sequencing datasets. Arthritis Res Ther. 2022;24(1):215. doi: 10.1186/s13075-022-02902-x.

62. Micheroli R., Elhai M., Edalat S., Frank-Bertoncelj M., Bürki K., Ciurea A., et al. Role of synovial fibroblast subsets across synovial pathotypes in rheumatoid arthritis: a deconvolution analysis. RMD Open. 2022;8(1) doi: 10.1136/rmdopen-2021-001949.

63. Armingol E., Officer A., Harismendy O., Lewis N.E. Deciphering cell–cell interactions and communication from gene expression. Nat Rev Genet. 2021;22(2):71–88. doi: 10.1038/s41576-020-00292-x.

64. Browaeys R., Saelens W., Saeys Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods. 2020&;;;17(2):159–162. doi: 10.1038/s41592-019-0667-5.

65. Vickovic S., Schapiro D., Carlberg K., Lötstedt B., Larsson L., Hildebrandt F., et al. Three-dimensional spatial transcriptomics uncovers cell type localizations in the human rheumatoid arthritis synovium. Commun Biol. 2022;5(1):1–11. doi: 10.1038/s42003-022-03050-3.

66. Ptacek J., Locke D., Finck R., Cvijic M.E., Li Z., Tarolli J.G., et al. Multiplexed ion beam imaging (MIBI) for characterization of the tumor microenvironment across tumor types. Lab Invest. 2020;100(8):1111–1123. doi: 10.1038/s41374-020-0417-4.

67. Carlberg K., Korotkova M., Larsson L., Catrina A.I., Ståhl P.L., Malmström V. Exploring inflammatory signatures in arthritic joint biopsies with Spatial Transcriptomics. Sci Rep. 2019;9(1):18975. doi: 10.1038/s41598-019-55441-y.

68. Hardt U., Carlberg K., af Klint E., Sahlström P., Larsson L., van Vollenhoven A., et al. Integrated single cell and spatial transcriptomics reveal autoreactive differentiated B cells in joints of early rheumatoid arthritis. Sci Rep. 2022;12(1):11876. doi: 10.1038/s41598-022-15293-5.

69. Williams C.G., Lee H.J., Asatsuma T., Vento-Tormo R., Haque A. An introduction to spatial transcriptomics for biomedical research. Genome Med. 2022;14(1):68. doi: 10.1186/s13073-022-01075-1.

70. Chen A., Liao S., Cheng M., Ma K., Wu L., Lai Y., et al. Spatiotemporal transcriptomic atlas of mouse organogenesis using DNA nanoball-patterned arrays. Cell. 2022;185(10):1777–1792.e21. doi: 10.1016/j.cell.2022.04.003.

71. Rae Buchberger A., DeLaney K., Johnson J., Li L. Mass spectrometry imaging: a review of emerging advancements and future insights. Anal Chem. 2018;90(1):240–265. doi: 10.1021/acs.analchem.7b04733.

72. Cai Z., Poulos R.C., Liu J., Zhong Q. Machine learning for multi-omics data integration in cancer. iScience. 2022;25(2) doi: 10.1016/j.isci.2022.103798.

73. Frank-Bertoncelj M., Trenkmann M., Klein K., Karouzakis E., Rehrauer H., Bratus A., et al. Epigenetically-driven anatomical diversity of synovial fibroblasts guides joint-specific fibroblast functions. Nat Commun. 2017;8(1):14852. doi: 10.1038/ncomms14852.

74. Singh A., Shannon C.P., Gautier B., Rohart F., Vacher M., Tebbutt S.J., et al. DIABLO: an integrative approach for identifying key molecular drivers from multi-omics assays. Bioinforma Oxf Engl. 2019 1;35(17):3055–3062. doi: 10.1093/bioinformatics/bty1054.

75. Meng C., Helm D., Frejno M., Kuster B. moCluster: identifying joint patterns across multiple omics data sets. J Proteome Res. 2016;15(3):755–765. doi: 10.1021/acs.jproteome.5b00824.

76. Chacko J.V., Eliceiri K.W. Autofluorescence Lifetime Imaging of cellular metabolism: sensitivity towards cell density, pH, intracellular and intercellular heterogeneity. Cytom Part J Int Soc Anal Cytol. 2019;95(1):56–69. doi: 10.1002/cyto.a.23603.

77. Lakowicz J. Principles of Fluorescence Spectroscopy. Vol. 1, Principles of Fluorescent Spectroscopy, 3rd Edn. 2006.

78. Drozdowicz-Tomsia K., Anwer A.G., Cahill M.A., Madlum K.N., Maki A.M., Baker M.S., et al. Multiphoton fluorescence lifetime imaging microscopy reveals free-to-bound NADH ratio changes associated with metabolic inhibition. J Biomed Opt. 2014;19(8) doi: 10.1117/1.JBO.19.8.086016.

79. Skala M.C., Riching K.M., Gendron-Fitzpatrick A., Eickhoff J., Eliceiri K.W., White J.G., et al. In vivo multiphoton microscopy of NADH and FAD redox states, fluorescence lifetimes, and cellular morphology in precancerous epithelia. Proc Natl Acad Sci. 2007;104(49):19494–19499. doi: 10.1073/pnas.0708425104.

80. Sameni S., Syed A., Marsh J.L., Digman M.A. The phasor-FLIM fingerprints reveal shifts from OXPHOS to enhanced glycolysis in Huntington disease. Sci Rep. 2016;6:34755. doi: 10.1038/srep34755.

81. Blacker T.S., Mann Z.F., Gale J.E., Ziegler M., Bain A.J., Szabadkai G., et al. Separating NADH and NADPH fluorescence in live cells and tissues using FLIM. Nat Commun. 2014;5(1):3936. doi: 10.1038/ncomms4936.

82. Chen C., Lai X., Zhang Y., Xie L., Yu Z., Dan S., et al. NADPH metabolism determines the leukemogenic capacity and drug resistance of AML cells. Cell Rep. 2022;39(1) doi: 10.1016/j.celrep.2022.110607.

83. Niederschweiberer M.A., Schaefer P.M., Singh L.N., Lausser L., Bhosale D., Hesse R., et al. NADH fluorescence lifetime imaging microscopy reveals selective mitochondrial dysfunction in neurons overexpressing Alzheimer’s disease–related proteins. Front Mol Biosci. 2021 doi: 10.3389/fmolb.2021.671274. cited 2023 Mar 6];8. Available from: https://www.frontiersin.org/articles/10.3389/fmolb.2021.671274.

84. Erkkilä M.T., Reichert D., Gesperger J., Kiesel B., Roetzer T., Mercea P.A., et al. Macroscopic fluorescence-lifetime imaging of NADH and protoporphyrin IX improves the detection and grading of 5-aminolevulinic acid-stained brain tumors. Sci Rep. 2020;10(1):20492. doi: 10.1038/s41598-020-77268-8.

85. Neto N.G., O’Rourke S.A., Zhang M., Fitzgerald H.K., Dunne A., Monaghan M.G. Non-invasive classification of macrophage polarisation by 2P-FLIM and machine learning. Dustin ML, Walczak AM, Dustin ML, Padilla-Parra S, editors. eLife. 2022 Oct 18;11:e77373.

86. Guo X., Xu T., Zheng J., Cui X., Li M., Wang K., et al. Accumulation of synovial fluid CD19+CD24hiCD27+ B cells was associated with bone destruction in rheumatoid arthritis. Sci Rep. 2020;10(1):14386. doi: 10.1038/s41598-020-71362-7.

87. Wu F., Gao J., Kang J., Wang X., Niu Q., Liu J., et al. B cells in rheumatoid arthritis：pathogenic mechanisms and treatment prospects. Front Immunol. 2021;12 doi: 10.3389/fimmu.2021.750753.

88. Yeo L., Toellner K.M., Salmon M., Filer A., Buckley C.D., Raza K., et al. Cytokine mRNA profiling identifies B cells as a major source of RANKL in rheumatoid arthritis. Ann Rheum Dis. 2011;70(11):2022–2028. doi: 10.1136/ard.2011.153312.

89. Scherer H.U., van der Woude D., Toes R.E.M. From risk to chronicity: evolution of autoreactive B cell and antibody responses in rheumatoid arthritis. Nat Rev Rheuma. 2022;18(7):371–383. doi: 10.1038/s41584-022-00786-4.

90. Hrovatin K., Fischer D.S., Theis F.J. Toward modeling metabolic state from single-cell transcriptomics. Mol Metab. 2021;57 doi: 10.1016/j.molmet.2021.101396.

91. Xiao Z., Dai Z., Locasale J.W. Metabolic landscape of the tumor microenvironment at single cell resolution. Nat Commun. 2019;10(1):3763. doi: 10.1038/s41467-019-11738-0.

92. Orth J.D., Thiele I., Palsson B.Ø. What is flux balance analysis? Nat Biotechnol. 2010;28(3):245–248. doi: 10.1038/nbt.1614.

93. Wagner A., Wang C., Fessler J., DeTomaso D., Avila-Pacheco J., Kaminski J., et al. Metabolic modeling of single Th17 cells reveals regulators of autoimmunity. Cell. 2021;184(16):4168–4185.e21. doi: 10.1016/j.cell.2021.05.045.

94. Lanekoff I., Sharma V.V., Marques C. Single-cell metabolomics: where are we and where are we going? Curr Opin Biotechnol. 2022;75 doi: 10.1016/j.copbio.2022.102693.

95. Bowman A.P., Bogie J.F.J., Hendriks J.J.A., Haidar M., Belov M., Heeren R.M.A., et al. Evaluation of lipid coverage and high spatial resolution MALDI-imaging capabilities of oversampling combined with laser post-ionisation. Anal Bioanal Chem. 2020;412(10):2277–2289. doi: 10.1007/s00216-019-02290-3.

96. Gupta R.K., Kuznicki J. Biological and medical importance of cellular heterogeneity deciphered by single-cell RNA Sequencing. Cells. 2020;9(8):1751. doi: 10.3390/cells9081751.

97. Sarbadhikary P., George B.P., Abrahamse H. Paradigm shift in future biophotonics for imaging and therapy: Miniature living lasers to cellular scale optoelectronics. Theranostics. 2022;12(17):7335–7350. doi: 10.7150/thno.75905.

98. Krasnodębski M. Throwing light on photonics: the genealogy of a technological paradigm. Centaurus. 2018;60(1–2):3–24.

99. Molony C., King D., Di Luca M., Kitching M., Olayinka A., Hakimjavadi R., et al. Disease-relevant single cell photonic signatures identify S100β stem cells and their myogenic progeny in vascular lesions. Stem Cell Rev Rep. 2021;17(5):1713–1740. doi: 10.1007/s12015-021-10125-x.

100. Yun S.H., Kwok S.J.J. Light in diagnosis, therapy and surgery. Nat Biomed Eng. 2017;1:0008. doi: 10.1038/s41551-016-0008.

101. Qavi A.J., Washburn A.L., Byeon J.Y., Bailey R.C. Label-free technologies for quantitative multiparameter biological analysis. Anal Bioanal Chem. 2009;394(1):121–135. doi: 10.1007/s00216-009-2637-8.

102. Habibalahi A., Moghari M.D., Campbell J.M., Anwer A.G., Mahbub S.B., Gosnell M., et al. Non-invasive real-time imaging of reactive oxygen species (ROS) using auto-fluorescence multispectral imaging technique: a novel tool for redox biology. Redox Biol. 2020;34 doi: 10.1016/j.redox.2020.101561.

103. Ranjit S., Dvornikov A., Stakic M., Hong S.H., Levi M., Evans R.M., et al. Imaging fibrosis and separating collagens using second harmonic generation and phasor approach to fluorescence lifetime imaging. Sci Rep. 2015;5 doi: 10.1038/srep13378.

104. Squirrell J.M., Fong J.J., Ariza C.A., Mael A., Meyer K., Shevde N.K., et al. Endogenous fluorescence signatures in living pluripotent stem cells change with loss of potency. PLoS One. 2012;7(8):43708. doi: 10.1371/journal.pone.0043708.

105. Burger R., Reith P., Kijanka G., Akujobi V., Abgrall P., Ducrée J. Array-based capture, distribution, counting and multiplexed assaying of beads on a centrifugal microfluidic platform. Lab Chip. 2012;12(7):1289–1295. doi: 10.1039/c2lc21170j.

106. King D., Glynn M.D., Cindric S., Kernan D., O’Connell T., Hakimjavadi R., et al. Label-free multi parameter optical interrogation of endothelial activation in single cells using a lab on a disc platform. Sci Rep. 2019;9(1):4157. doi: 10.1038/s41598-019-40612-8.

107. Sardana Y., Bhatti G.K., Singh C., Sharma P.K., Reddy P.H., Bhatti J.S. Progression of pre-rheumatoid arthritis to clinical disease of joints: potential role of mesenchymal stem cells. Life Sci. 2023 doi: 10.1016/j.lfs.2023.121641.

108. Soundararajan M., Kannan S. Fibroblasts and mesenchymal stem cells: two sides of the same coin? J Cell Physiol. 2018;233(12):9099–9109. doi: 10.1002/jcp.26860.

109. Spyratou E. Advanced Biophotonics techniques for cell and molecules manipulation associated with cancer and autoimmune diseases: the role of optical tweezers [Internet]. arXiv; 2021 [cited 2023 Mar 31]. Available from: http://arxiv.org/abs/2106.05562.

110. Yuan Z., Ieong F.H., Jiang H., Sobel E.S. Osteoarthritis and psoriatic arthritis: findings in three-dimensional biophotonics imaging. Biomed Mater Eng. 2014;24(6):3063–3071. doi: 10.3233/BME-141128.

111. Matcher S.J. What can biophotonics tell us about the 3D microstructure of articular cartilage? Quant Imaging Med Surg. 2015;5(1) doi: 10.3978/j.issn.2223-4292.2014.12.03. 14358–14158.

112. Tang F., Lao K., Surani M.A. Development and applications of single cell transcriptome analysis. Nat Methods. 2011;8(4 Suppl):S6–S11. doi: 10.1038/nmeth.1557.

113. Vu T., Vallmitjana A., Gu J., La K., Xu Q., Flores J., et al. Spatial transcriptomics using combinatorial fluorescence spectral and lifetime encoding, imaging and analysis. Nat Commun. 2022;13(1):169. doi: 10.1038/s41467-021-27798-0.

114. Park C., Took C.C., Seong J.K. Machine learning in biomedical engineering. Biomed Eng Lett. 2018;1(1):1–3. doi: 10.1007/s13534-018-0058-3. 8.

115. Samuel A.L. Some Studies in Machine Learning Using the Game of Checkers.

116. Nadif M., Role F. Unsupervised and self-supervised deep learning approaches for biomedical text mining. Brief Bioinf. 2021;22(2):1592–1603. doi: 10.1093/bib/bbab016.

117. Jiang T., Gradus J.L., Rosellini A.J. Supervised machine learning: a brief primer. Behav Ther. 2020;51(5):675–687. doi: 10.1016/j.beth.2020.05.002.

118. Davies P., Jones M., Liu J., Hebenstreit D. Anti-bias training for (sc)RNA-seq: experimental and computational approaches to improve precision. Brief Bioinform. 2021;22(6):bbab148. doi: 10.1093/bib/bbab148.

119. Bray N.L., Pimentel H., Melsted P., Pachter L. Near-optimal probabilistic RNA-seq quantification. Nat Biotechnol. 2016;34(5):525–527. doi: 10.1038/nbt.3519.

120. Liu J., Fan Z., Zhao W., Zhou X. Machine Intelligence in Single-Cell Data Analysis: Advances and New Challenges. Front Genet [Internet]. 2021 [cited 2023 Mar 7];12. Available from: https://www.frontiersin.org/articles/10.3389/fgene.2021.655536.

121. Trapnell C. Defining cell types and states with single-cell genomics. Genome Res. 2015;25(10):1491–1498. doi: 10.1101/gr.190595.115.

122. Yu L., Cao Y., Yang J.Y.H., Yang P. Benchmarking clustering algorithms on estimating the number of cell types from single-cell RNA-sequencing data. Genome Biol. 2022 8;23(1):49. doi: 10.1186/s13059-022-02622-0.

123. Qiu P. Embracing the dropouts in single-cell RNA-seq analysis. Nat Commun. 2020;11(1):1169. doi: 10.1038/s41467-020-14976-9.

124. Wang L., Xi Y., Sung S., Qiao H. RNA-seq assistant: machine learning based methods to identify more transcriptional regulated genes. BMC Genom. 2018;19(1):546. doi: 10.1186/s12864-018-4932-2.

125. Boratyn G.M., Thierry-Mieg J., Thierry-Mieg D., Busby B., Madden T.L. Magic-BLAST, an accurate RNA-seq aligner for long and short reads. BMC Bioinforma. 2019;20(1):405. doi: 10.1186/s12859-019-2996-x.

126. Xu J., Cai L., Liao B., Zhu W., Yang J. CMF-Impute: an accurate imputation tool for single-cell RNA-seq data. Bioinformatics. 2020;36(10):3139–3147. doi: 10.1093/bioinformatics/btaa109.

127. Ianevski A., Giri A.K., Aittokallio T. Fully-automated and ultra-fast cell-type identification using specific marker combinations from single-cell transcriptomic data. Nat Commun. 2022;13(1):1246. doi: 10.1038/s41467-022-28803-w.

128. Zhang Y., Zhang F., Wang Z., Wu S., Tian W. scMAGIC: accurately annotating single cells using two rounds of reference-based classification. Nucleic Acids Res. 2022;50(8) doi: 10.1093/nar/gkab1275.

129. Iida K., Kondo J., Wibisana J.N., Inoue M., Okada M. ASURAT: functional annotation-driven unsupervised clustering of single-cell transcriptomes. Bioinformatics. 2022;38(18):4330–4336. doi: 10.1093/bioinformatics/btac541.

130. Galdos F.X., Xu S., Goodyer W.R., Duan L., Huang Y.V., Lee S., et al. devCellPy is a machine learning-enabled pipeline for automated annotation of complex multilayered single-cell transcriptomic data. Nat Commun. 2022;13(1):5271. doi: 10.1038/s41467-022-33045-x.

131. Pradhan P., Guo S., Ryabchykov O., Popp J., Bocklitz T.W. Deep learning a boon for biophotonics? Vol. 13. J Biophotonics Wiley-VCH Verl. 2020 doi: 10.1002/jbio.201960186.

132. Melanthota S.K., Gopal D., Chakrabarti S., Kashyap A.A., Radhakrishnan R., Mazumder N. Deep learning-based image processing in optical microscopy. Biophys Rev. 2022;14(2):463–481. doi: 10.1007/s12551-022-00949-3.

133. von Chamier L., Laine R.F., Henriques R. Artificial intelligence for microscopy: what you should know. Biochem Soc Trans. 2019;47(4):1029–1040. doi: 10.1042/BST20180391.

134. Duh Q.Y., Davis S.N. Autofluorescence and artificial intelligence: the future of parathyroid surgery? Ann Surg Oncol. 2022;29(7):4026–4027. doi: 10.1245/s10434-022-11732-9.

135. Kothari R., Jones V., Mena D., Bermúdez Reyes V., Shon Y., Smith J.P., et al. Raman spectroscopy and artificial intelligence to predict the Bayesian probability of breast cancer. Sci Rep. 2021;11(1):6482. doi: 10.1038/s41598-021-85758-6.

136. Alam U., Anson M., Meng Y., Preston F., Kirthi V., Jackson T.L., et al. Artificial intelligence and corneal confocal microscopy: the start of a beautiful relationship. J Clin Med. 2022;11(20):6199. doi: 10.3390/jcm11206199.

137. Kapoor R., Whigham B.T., Al-Aswad L.A. Artificial intelligence and optical coherence tomography imaging. Asia-Pac J Ophthalmol. 2019;8(2):187. doi: 10.22608/APO.201904.

138. Roodenrijs N.M.T., Kedves M., Hamar A., Nagy G., Laar J.M. van, Heijde D. van der, et al. Diagnostic issues in difficult-to-treat rheumatoid arthritis: a systematic literature review informing the EULAR recommendations for the management of difficult-to-treat rheumatoid arthritis. RMD Open. 2021;7(1) doi: 10.1136/rmdopen-2020-001511.

139. Luo C., Zhu Y., Zhu Z., Li R., Chen G., Wang Z. A machine learning-based risk stratification tool for in-hospital mortality of intensive care unit patients with heart failure. J Transl Med. 2022;20(1):136. doi: 10.1186/s12967-022-03340-8.

140. Prasad B., Bjourson A.J., Shukla P. Data-driven patient stratification of UK Biobank cohort suggests five endotypes of multimorbidity. Brief Bioinform. 2022;23(6):bbac410. doi: 10.1093/bib/bbac410.

141. Momtazmanesh S., Nowroozi A., Rezaei N. Artificial intelligence in rheumatoid arthritis: current status and future perspectives: a state-of-the-art review. Rheuma Ther. 2022;9(5):1249–1304. doi: 10.1007/s40744-022-00475-4.

142. Gola D., König I.R. Empowering individual trait prediction using interactions for precision medicine. BMC Bioinforma. 2021;22(1):74. doi: 10.1186/s12859-021-04011-z.

143. Koo B.S., Eun S., Shin K., Yoon H., Hong C., Kim D.H., et al. Machine learning model for identifying important clinical features for predicting remission in patients with rheumatoid arthritis treated with biologics. Arthritis Res Ther. 2021;23(1):178. doi: 10.1186/s13075-021-02567-y.

144. Merola J.F., Espinoza L.R., Fleischmann R. Distinguishing rheumatoid arthritis from psoriatic arthritis. RMD Open. 2018;4(2) doi: 10.1136/rmdopen-2018-000656.

145. Raza K., Buckley C.E., Salmon M., Buckley C.D. Treating very early rheumatoid arthritis. Best Pr Res Clin Rheuma. 2006;20(5):849–863. doi: 10.1016/j.berh.2006.05.005.

146. Martorell-Marugán J., Tabik S., Benhammou Y., del Val C., Zwir I., Herrera F., et al. Deep Learning in Omics Data Analysis and Precision Medicine. In: Husi H, editor. Computational Biology [Internet]. Brisbane (AU): Codon Publications; 2019 [cited 2023 Jul 31]. Available from: http://www.ncbi.nlm.nih.gov/books/NBK550335/.

147. Feldner-Busztin D., Firbas Nisantzis P., Edmunds S.J., Boza G., Racimo F., Gopalakrishnan S., et al. Dealing with dimensionality: the application of machine learning to multi-omics data. Bioinformatics. 2023:btad021. doi: 10.1093/bioinformatics/btad021.

148. Reel P.S., Reel S., Pearson E., Trucco E., Jefferson E. Using machine learning approaches for multi-omics data analysis: a review. Biotechnol Adv. 2021;49 doi: 10.1016/j.biotechadv.2021.107739.

149. Nyamundanda G., Brennan L., Gormley I.C. Probabilistic principal component analysis for metabolomic data. BMC Bioinforma. 2010;11(1):571. doi: 10.1186/1471-2105-11-571.

150. van Dijk D., Sharma R., Nainys J., Yim K., Kathail P., Carr A.J., et al. Recovering gene interactions from single-cell data using data diffusion. Cell. 2018;174(3):716–729.e27. doi: 10.1016/j.cell.2018.05.061.

151. de Tayrac M., Lê S., Aubry M., Mosser J., Husson F. Simultaneous analysis of distinct Omics data sets with integration of biological knowledge: multiple Factor Analysis approach. BMC Genom. 2009;10(1):32. doi: 10.1186/1471-2164-10-32.

152. Burton M., Thomassen M., Tan Q., Kruse T.A. Gene expression profiles for predicting metastasis in breast cancer: a cross-study comparison of classification methods. Sci World J. 2012;2012 doi: 10.1100/2012/380495.

153. Beaulieu-Jones B.K., Yuan W., Brat G.A., Beam A.L., Weber G., Ruffin M., et al. Machine learning for patient risk stratification: standing on, or looking over, the shoulders of clinicians? Npj Digit Med. 2021;4(1):1–6. doi: 10.1038/s41746-021-00426-3.

154. Chervitz S.A., Deutsch E.W., Field D., Parkinson H., Quackenbush J., Rocca-Serra P., et al. Data standards for omics data: the basis of data sharing and reuse. Methods Mol Biol Clifton NJ. 2011;719:31–69. doi: 10.1007/978-1-61779-027-0_2.

155. Melanthota S.K., Gopal D., Chakrabarti S., Kashyap A.A., Radhakrishnan R., Mazumder N. Deep learning-based image processing in optical microscopy. Biophys Rev. 2022;14(2):463–481. doi: 10.1007/s12551-022-00949-3.

156. Belthangady C., Royer L.A. Applications, promises, and pitfalls of deep learning for fluorescence image reconstruction. Nat Methods. 2019;16(12):1215–1225. doi: 10.1038/s41592-019-0458-z.

157. Bornhorst J., Nustede E.J., Fudickar S. Mass surveilance of C. elegans—smartphone-based DIY microscope and machine-learning-based approach for worm detection. Sensors. 2019;19(6):1468. doi: 10.3390/s19061468.

158. Molony C., King D., Di Luca M., Kitching M., Olayinka A., Hakimjavadi R., et al. Disease-relevant single cell photonic signatures identify S100β stem cells and their myogenic progeny in vascular lesions. Stem Cell Rev Rep. 2021;17(5):1713–1740. doi: 10.1007/s12015-021-10125-x.

159. Kothari R., Jones V., Mena D., Bermúdez Reyes V., Shon Y., Smith J.P., et al. Raman spectroscopy and artificial intelligence to predict the Bayesian probability of breast cancer. Sci Rep. 2021;11(1):6482. doi: 10.1038/s41598-021-85758-6.

160. Ranzoni A.M., Tangherloni A., Berest I., Riva S.G., Myers B., Strzelecka P.M., et al. Integrative single-Cell RNA-seq and ATAC-Seq analysis of human developmental hematopoiesis. Cell Stem Cell. 2021;28(3):472–487.e7. doi: 10.1016/j.stem.2020.11.015.

161. Cai Z., Poulos R.C., Liu J., Zhong Q. Machine learning for multi-omics data integration in cancer. iScience. 2022;25(2) doi: 10.1016/j.isci.2022.103798.

162. Bostanci E., Kocak E., Unal M., Guzel M.S., Acici K., Asuroglu T. Machine learning analysis of RNA-seq data for diagnostic and prognostic prediction of colon cancer. Sensors. 2023;23(6):3080. doi: 10.3390/s23063080.

163. Borisov N., et al. Sergeeva A., Suntsova M., Raevskiy M., Gaifullin N., Mendeleeva L. Machine Learning Applicability for Classification of PAD/VCD Chemotherapy Response Using 53 Multiple Myeloma RNA Sequencing Profiles. Frontiers in Oncology. 2021 doi: 10.3389/fonc.2021.652063.

164. Parker J.S., Mullins M., Cheang M.C.U., Leung S., Voduc D., Vickery T., et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. J Clin Oncol. 2009;27(8):1160–1167. doi: 10.1200/JCO.2008.18.1370.

165. Cascianelli S., Molineris I., Isella C., Masseroli M., Medico E. Machine learning for RNA sequencing-based intrinsic subtyping of breast cancer. Sci Rep. 2020;10(1) doi: 10.1038/s41598-020-70832-2.

166. Moriel N., Senel E., Friedman N., Rajewsky N., Karaiskos N., Nitzan M. NovoSpaRc: flexible spatial reconstruction of single-cell gene expression with optimal transport. Nat Protoc. 2021;16(9):4177–4200. doi: 10.1038/s41596-021-00573-7.

167. Clark C., Dayon L., Masoodi M., Popp J. An integrative, hypothesis-free, multi-omics approach uncovers biological pathway alterations in Alzheimer’s disease. Alzheimers Dement. 2020;16(S2) doi: 10.1186/s13195-021-00814-7.

168. Lim N., Tesar S., Belmadani M., Poirier-Morency G., Mancarci B.O., Sicherman J., et al. Curation of over 10 000 transcriptomic studies to enable data reuse. Database J Biol Databases Curation. 2021;2021:baab006. doi: 10.1093/database/baab006.

169. Su C., Tong J., Wang F. Mining genetic and transcriptomic data using machine learning approaches in Parkinson’s disease. Npj Park Dis. 2020;6(1):1–10. doi: 10.1038/s41531-020-00127-w.

170. Phan J.H., Quo C.F., Cheng C., Wang M.D. Multiscale integration of -omic, imaging, and clinical data in biomedical informatics. IEEE Rev Biomed Eng. 2012;5:74–87. doi: 10.1109/RBME.2012.2212427.

171. Schmidt R., Weihs T., Wurm C.A., Jansen I., Rehman J., Sahl S.J., et al. MINFLUX nanometer-scale 3D imaging and microsecond-range tracking on a common fluorescence microscope. Nat Commun. 2021;12(1):1478. doi: 10.1038/s41467-021-21652-z.