Unpacking Genomic Biomarkers for Programmed Cell Death Receptor-1 Immunotherapy Success in Non-Small Cell Lung Cancer Using Deep Neural Networks: Quantitative Study.

Background: Non-small cell lung cancer (NSCLC) is one of the leading causes of cancer-related mortality. Programmed cell death receptor-1 (PD-1) immunotherapy has shown results in the treatment of NSCLC; however, not all patients respond effectively to it. Identifying predictive biomarkers for PD-1 therapy response is critical to improving patient outcomes and treatment strategies. Traditional methods of biomarker discovery often fall short in terms of accuracy and comprehensiveness. Recent advancements in deep learning provide a powerful approach to analyze complex genomic data to resolve this issue. Objective: This study aims to leverage deep neural networks (DNNs) to identify genomic biomarkers predictive of patient responses to PD-1 immunotherapy in NSCLC. DeepImmunoGene is a model designed using a reduced feature set to identify the most critical biomarkers. We use feature selection to reduce the space and apply deep learning to identify the highly predictive gene subset. Methods: Differentially expressed genes were identified in RNA-seq data from 355 patients with NSCLC using the LIMMA package in R, followed by preprocessing with log2 transformation, removing outliers, and detecting easily identified genes. Machine learning models, including support vector machines, extreme gradient boosting (XGBoost), and DNNs, were applied to gene expression data to predict patient responses to immunotherapy. Key predictive genes were identified through model interpretation techniques, and differences in model performance were assessed for statistical significance. Primarily, the metric used identifies which genes serve as key biomarkers in regard to immunotherapy detection. Results: Initially, we identified 1093 differentially expressed genes from RNA-seq data of 355 patients. We then trained models using SVM, XGBoost, and DNN to predict immunotherapy response. The DNN model outperformed both SVM and XGBoost with an accuracy of 82%, an area under the curve of 90%, and recall of 85%. To identify key biomarkers, we performed a permutation importance analysis, narrowing down the gene set to 98 genes. DeepImmunoGene, trained on these 98 genes, showed superior results, with an accuracy of 87% and an area under the curve of 95%. The top 36 upregulated genes in responders and 62 upregulated genes in nonresponders were identified, which could serve as potential biomarkers for predicting response to PD-1 inhibitors. These findings suggest that DeepImmunoGene can reliably forecast immunotherapy outcomes and aid in biomarker discovery, supporting the development of more personalized treatment strategies in NSCLC. Conclusions: The DeepImmunoGene predictive model identified 36 upregulated genes that may represent candidate genomic biomarkers associated with response to PD-1 immunotherapy in patients with NSCLC. Notably, the 10 most significant genes offer valuable insights into the underlying mechanisms of treatment responses. These biomarkers may not only aid in predicting which patients are more likely to respond to PD-1 immunotherapy but also offer insights into the molecular differences associated with nonresponse.