Harnessing Multi-Omics and Deep Learning to Predict Immunotherapy Response in Lung Cancer: A Step Toward Precision Oncology

Background and Aims: Lung cancer patients have accessed better therapy options through immunotherapy treatments although patient responses continue to vary. The current biomarkers which include PD-L۱ expression prove to have poor predictive capabilities. This study creates a deep learning prediction model that combines genomic and transcriptomic and proteomic data approaches to determine lung cancer patient response to immunotherapy. Methods: Our research initiated by obtaining ۲۰۰ lung cancer patients receiving anti-PD-۱/PD-L۱ therapy. The researchers performed WES together with RNA-seq and mass spectrometry-based proteomics testing on their ۲۰۰ lung cancer patients undergoing anti-PD-۱/PD-L۱ therapy. The patient cohort was divided into responders and non-responders following the evaluation under RECIST v۱.۱ criteria. A deep learning model utilized multiple networks which processed individual omics data to combine their outputs through a fusion layer that performed response predictions. A model training process used both cross-validation with hyperparameter adjustment. Results: Using this model resulted in ۸۵% accuracy alongside a measure of ۰.۹۰ AUC value. The SHAP values analysis revealed important features that comprised specific gene mutations together with immune-related gene expressions and protein abundance measurements. Conclusion: The deep learning model illustrates the effectiveness of multi-omic data combination for lung cancer immunotherapy prediction which establishes procedures for individualized therapy approaches. Nonetheless, this model necessitates further research and clinical trials to accurately assess its efficacy. Additionally, it is imperative to evaluate this proposed model across various types of cancer.

کلیدواژه ها:

Deep Learning ، Immunotherapy ، Lung Cancer ، Multi-Omics