Knowledge-guided multi-level network modeling with experimental characterization identifies PRKCA as a novel biomarker and tumor suppressor triggering ferroptosis in prostate cancer.

Prostate cancer (PCa) is observed with high incidence in men worldwide. Ferroptosis, occurred from disorders in a series of gene and pathway regulation, is an emerging target against cancer. However, most of the computational approaches solely treated ferroptosis-related genes (FRGs) as independent variables in model training, and the interactions among FRGs and other candidates were not fully deciphered in a disease-specific content. In this study, a novel network-based and knowledge-guided bioinformatics model was proposed by integrating ferroptosis-related prior knowledge with topological and functional characterization on a protein-protein interaction network for biomarker discovery in PCa development and ferroptosis. The model started at a random walk with restart algorithm for weighting genes close to known FRGs in the PCa-specific network to extract a core subnetwork for robustness and vulnerability analysis. Then key regulatory modules and a candidate gene, i.e. PRKCA, were respectively identified using a multi-level prioritization strategy with hub-bottleneck node filtering, edge-based gene co-expression measuring, community module detecting and a newly defined Ferr.neighbor functional score. The experimental validation using human clinical samples, cell lines, and nude mice convinced the role of PRKCA as a latent biomarker and a tumor suppressor in PCa carcinogenesis with a potential mechanism on triggering GPX4-mediated ferroptosis of PCa cells. This study provides a general-purpose systems biology framework for significant FRG screening, and future translational perspectives of PRKCA as a novel diagnostic and therapeutic signature for PCa management should be explored.