Abstract
Disease subtyping is essential for personalized medicine, enabling tailored treatment strategies based on disease heterogeneity. Advances in high-throughput technologies have led to the rapid accumulation of multi-omics data, driving the development of integration methods for comprehensive disease subtyping. However, existing approaches often lack explainability and fail to establish clear links between subtypes and clinical outcomes. To address these challenges, we developed EMitool, an explainable multi-omics integration tool that leverages a network-based fusion strategy to achieve biologically and clinically relevant disease subtyping without requiring prior clinical information. Using data from 31 cancer types in The Cancer Genome Atlas (TCGA), EMitool demonstrated superior subtyping accuracy compared to eight state-of-the-art methods. It also provides contribution scores for different omics data types, enhancing interpretability. EMitool-derived subtypes exhibited significant associations with the overall survival, pathological stage, tumor mutational burden, immune microenvironment characteristics, and therapeutic responses. Specifically, in kidney renal clear cell carcinoma (KIRC), EMitool identified three distinct subtypes with varying prognoses, immune cell compositions, and drug sensitivities. These findings highlight its potential for biomarker discovery and precision oncology.