Background <p>Despite advances in whole-genome sequencing and identifying cancer-associated genetic alterations, understanding the influence of multiple genetic alterations collectively on cancer phenotypes remains challenging, owing to mutation pattern complexity and variability. Here, we present the NETwork-based Genotype-to-Phenotype Transformation (NetG2P), which utilizes network propagation to translate genomic information into pathway interaction networks. Using the Cancer Genome Atlas dataset across 10 cancer types, we conducted a pan-cancer analysis using NetG2P to uncover critical oncogenic features associated with cancer prognosis using machine learning and explainable artificial intelligence models.</p> Results <p>Our results suggest that these features, which primarily represent signaling crosstalk, can serve as functional units for determining cancer prognosis. Network analysis of these critical oncogenic features reveals distinct patterns among cancer types, categorizing them into “distributed” and “modular” networks based on pathway interactions. Applying this technique to cancer cell line data has helped predict novel drug targets for high-risk groups and proposed candidates for drug repurposing.</p> Conclusions <p>NetG2P generates patient-specific networks of critical oncogenic features and suggests personalized treatments, hence advancing precision medicine in oncology.</p>