CLCluster: A redundancy-reduction contrastive learning-based clustering method of cancer subtype based on multi-omics data

Alternative splicing (AS) allows one gene to produce several protein variants, offering valuable predictive insights into cancer and facilitating targeted therapies. Although multi-omics data are used to identify cancer subtypes, AS is rarely utilized for this purpose. Here, we propose a redundancy-reduction contrastive learning-based method (CLCluster) based on expression, and AS for cancer subtype clustering of 33 cancer types. Ablation experiments emphasize the benefits of using AS data to subtype cancer. We identified 2,921 cancer subtype-related AS events associated with patient survival and conducted multiple analyses including open reading frame annotation, RNA binding protein (RBP)-associated AS regulation, and splicing-related anticancer peptides (ACPs) prediction for therapeutic biomarkers. The CLCluster model is more effective in identifying prognostic-relevant cancer subtypes than other models. The effective annotation of cancer subtype related AS events facilitates the identification of therapeutically targetable biomarkers in patients.