Multi-omics and AI-driven immune subtyping to optimize neoantigen-based vaccines for colorectal cancer

Abstract

Colorectal cancer (CRC) presents significant challenges due to limited targeted therapeutic options. This study integrates multi-omics analysis and AI to identify tumor antigens and immune gene targets for personalized immunotherapy. Using TCGA, differential expression and mutation analysis, we identified overexpressed and mutated genes in CRC. Among these, 62 neoantigens were shortlisted as potential tumor antigens. Survival analysis highlighted prognostic antigens, while their correlation with immune gene expression suggested these antigens could trigger immune activation. Three key neoantigens (TTK, EZH2, and KIF4A) emerged as promising candidates for immunotherapy. Based on immune gene activity, patients were categorized into three Immune Subtypes (IS). IS groups 1 and 2, characterized by high immune gene expression and immune activation markers, exhibited better survival outcomes, while IS 3, with low immune gene expression, showed poor survival and immune unresponsiveness. Neoantigen-based vaccines could potentially boost tumor recognition and improve survival for patients in immune-cold subtypes. Machine learning models like LightGBM, XGBoost, and XGBRF predicted optimal immune targets for vaccine design, validated through SHAP analysis. This study provides a machine learning- driven framework to identify tumor antigens and immune targets, offering a promising strategy for CRC immunotherapy tailored to immune subtype-specific responses.