A Single-Cell Guided Machine Learning Model Predicts Response to Immune Checkpoint Inhibitors in Gastric Cancer

Resistance to immune checkpoint inhibitors is a major clinical obstacle in the treatment of gastric cancer. Identifying drug-resistant cell populations and markers remains an urgent problem to be solved. This study by constructing a single-cell transcriptomic atlas of gastric cancer, we identified a subset of T/NK cells associated with ICI resistance. These cells exhibited impaired MHC-I-mediated immune recognition with tumor cells, were positioned at an early stage of T cell differentiation, and displayed elevated histidine metabolism. Mechanistically, we identified the transcription factor IRF1 as a potential suppressor of immune resistance in gastric cancer. Building on these findings, we developed a machine learning model that effectively predicts patient responses to immunotherapy. Notably, the model predicted responses reasonably well across two independent cohorts (AUCs 0.75 and 0.73). In vitro experiments further demonstrated that IRF1 inhibits cancer cell invasion and promotes apoptosis. In summary, this study identifies potential cellular and molecular determinants of immune resistance in gastric cancer and suggests that targeting this T/NK cell subset or restoring IRF1 function represents a promising strategy worth further exploration to overcome ICI resistance.