Fig. 2

Examples of platforms for the prediction of biological significance for genetic abnormalities and its application to diagnosis. A Schematic diagram showing an overview of BoostDM modified from Ref. [136]. A specific model (gradient boosted tree) was constructed for each of the 282 gene-tissue combinations based on 18 features that characterize the mechanism of tumorigenesis of oncogenes. Specifically, 50 basic classifiers were trained on random subsets with equal numbers of positive and negative mutations to adequately represent the diversity of passenger mutations and prevent overfitting. B Schematic diagram of the identification of seven molecular subtypes of RCC tumors utilizing machine learning, modified from Ref. [138]. An integrative, multi-omics analysis of 823 tumors from RCC patients identified molecular subsets associated with differences in clinical outcomes with angiogenesis inhibitors alone or in combination with checkpoint inhibitors. Unsupervised transcriptome analysis using NMF revealed seven molecular subsets with different angiogenic, immune, cell cycle, metabolic, and stromal programs