Multimodal integration of bioinformatic data and digital pathology as a tool for precision oncology. Part 1. Development of prognostic models based on multi-omics data and machine learning

The article reviews current methodologies for developing prognostic models in solid tumors using bioinformatics data. Primary focus is placed on the transition from discrete genetic alteration analysis toward the synthesis of functionally related gene panels and prognostic risk scores leveraging TCGA, GEO, and specialized databases (e.g., FerrDb, ImmPort). The role of various regulated cell death programs - including apoptosis, pyroptosis, ferroptosis, and cuproptosis - is analyzed in detail as critical predictors of patient survival and therapeutic response. The study systematizes machine learning algorithms (LASSO, Random Forest, matrix factorization techniques) and cellular deconvolution methods (CIBERSORT, TIMER) employed to characterize the immune landscape of the tumor microenvironment. The work discusses the potential of mathematical synthesis of molecular data to enhance the objective assessment of tumor invasive potential and explores future perspectives for personalized patient stratification within precision oncology frameworks.