Accurate cancer subtype classification is critical for personalized treatment, yet integrating multi-omics data remains challenging. We present a novel method that constructs multi-level cross-omics graphs and uses GraphSAGE with hierarchical contrastive learning to extract discriminative features for SVM-based classification. Tests on TCGA BRCA and GBM datasets show superior accuracy and reduced computational cost over state-of-the-art methods, enhancing clinical applicability.
Yang et al. (Fri,) studied this question.