Computer methods and programs in biomedicine · 2025
Integrating deep generative model with active learning for predicting immunotherapy responses in gastric cancer.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Lan Hao, Wang Jinzhou, Zhao Jinyi, Li Menglong, Pu Xuemei, Guo Yanzhi
Аннотация
BACKGROUND AND OBJECTIVE: Immune checkpoint therapy (ICT) has revolutionized the treatment of gastric cancer (GC), but only a small part of patients benefits from it, highlighting the urgency of developing models for precise immunotherapy response prediction of GC patients. However, the development of such models is hindered by the limited availability of ICT-related data. This study aimed to develop an accurate prediction model for immunotherapy response in GC patients using data augmentation and active learning techniques.
METHODS: We utilized a generative adversarial network (GAN) to augment current experimental data from 78 GC patients treated with anti-programmed death (anti-PD-1) therapy. We then introduced an active learning (AL) framework to select the most informative samples, minimizing redundant information and mitigating overfitting. Multiple machine learning algorithms were evaluated, with the Random Forest (RF) classifier identified as the optimal model. Its performance was validated on an independent cohort generated by the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm. Important genes and therapeutic agents were explored using Shapley Additive Explanations (SHAP) and drug-gene correlation analyses.
RESULTS: Trained on 317 curated samples derived from 46 real ones, our RF model yields an average accuracy of 0.911, Recall of 0.896, F1-score of 0.898 and area under the curve (AUC) of 0.902 on the real testing set. External validation on the TIDE-generated cohort yielded an AUC of 0.838. SHAP analysis revealed 15 key genes linked to critical immune processes, with over half displaying significant differential expression between responders and non-responders. Notably, two compounds, Docetaxel and Lapatinib, emerged as promising therapeutic agents based on their strong negative correlations with these key gene expression patterns.
CONCLUSION: The proposed framework effectively addresses data insufficiency in ICT response prediction and demonstrates high accuracy in identifying GC patients likely to benefit from immunotherapy. The identified key genes and candidate compounds provide mechanistic insights and potential strategies for combinatorial therapies. This approach could enhance the accuracy of ICT response predictions and reveal novel insights for potential therapeutic interventions.
METHODS: We utilized a generative adversarial network (GAN) to augment current experimental data from 78 GC patients treated with anti-programmed death (anti-PD-1) therapy. We then introduced an active learning (AL) framework to select the most informative samples, minimizing redundant information and mitigating overfitting. Multiple machine learning algorithms were evaluated, with the Random Forest (RF) classifier identified as the optimal model. Its performance was validated on an independent cohort generated by the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm. Important genes and therapeutic agents were explored using Shapley Additive Explanations (SHAP) and drug-gene correlation analyses.
RESULTS: Trained on 317 curated samples derived from 46 real ones, our RF model yields an average accuracy of 0.911, Recall of 0.896, F1-score of 0.898 and area under the curve (AUC) of 0.902 on the real testing set. External validation on the TIDE-generated cohort yielded an AUC of 0.838. SHAP analysis revealed 15 key genes linked to critical immune processes, with over half displaying significant differential expression between responders and non-responders. Notably, two compounds, Docetaxel and Lapatinib, emerged as promising therapeutic agents based on their strong negative correlations with these key gene expression patterns.
CONCLUSION: The proposed framework effectively addresses data insufficiency in ICT response prediction and demonstrates high accuracy in identifying GC patients likely to benefit from immunotherapy. The identified key genes and candidate compounds provide mechanistic insights and potential strategies for combinatorial therapies. This approach could enhance the accuracy of ICT response predictions and reveal novel insights for potential therapeutic interventions.