Machine Learning Predictor of Immune Checkpoint Blockade Response in Gastric Cancer

SIMPLE SUMMARY: This study deals with the identification of signature genes through a model using four machine learning algorithms for two cohorts of bulk and single cell RNA seq to predict immune checkpoint blockade (ICB) response in gastric cancer. Through LASSO feature selection, we identified VCAN as a marker gene signature that distinguishes responders from non-responders. ABSTRACT: Predicting responses to immune checkpoint blockade (ICB) lacks official standards despite the discovery of several markers. Expensive drugs and different reactivities for each patient are the main disadvantages of immunotherapy. Gastric cancer is refractory and stem-like in nature and does not respond to immunotherapy. In this study, we aimed to identify a characteristic gene that predicts ICB response in gastric cancer and discover a drug target for non-responders. We built and evaluated a model using four machine learning algorithms for two cohorts of bulk and single-cell RNA seq to predict ICB response in gastric cancer patients. Through the LASSO feature selection, we discovered a marker gene signature that distinguishes responders from non-responders. VCAN, a candidate characteristic gene selected by all four machine learning algorithms, had a significantly high prevalence in non-responders (p = 0.0019) and showed a poor prognosis (p = 0.0014) at high expression values. This is the first study to discover a signature gene for predicting ICB response in gastric cancer by molecular subtype and provides broad insights into the treatment of stem-like immuno-oncology through precision medicine.