Integrating weighted correlation network analysis and machine learning identifies common trajectories of prostate cancer

The identification of molecular determinants of prostate cancer development and progression represents a major clinical and biological undertaking. We integrated transcriptomic data of primary localized and advanced prostate cancer from two cancer databases. Transcriptomic analysis showed upregulation of genes encoding for cell surface proteins in metastatic tumors, which were associated with cell-matrix components and chemokine signaling. Integration of machine learning and weighted correlation network modules identified the EZH2-TROAP axis as the main trajectory from initial tumor development to lethal metastatic disease. In addition to known biomarkers, we identified several targets including HROB and GEN1 that are implicated in DNA damage repair pathways or prognostic markers including SBK1, DLX2 and E2F2, whose biological role has not yet been elucidated for prostate cancer. Our results demonstrate the usability of complex bioinformatics approaches to identify biological drivers of prostate cancer progression and their suitability for clinical applications as prognostic biomarkers.