Identification of Pan-Cancer Prognostic Biomarkers Through Integration of Multi-Omics Data

Prognostic biomarkers dedicating to treat cancer are very difficult to identify. Although high-throughput sequencing technology allows us to mine prognostic biomarkers much deeper by analyzing omics data, there is lack of effective methods to comprehensively utilize multi-omics data. In this work, we integrated multi-omics data [DNA methylation (DM), gene expression (GE), somatic copy number alternation, and microRNA expression (ME)] and proposed a method to rank genes by desiring a "Score." Applying the method, cancer-specific prognostic biomarkers for 13 cancers were obtained. The prognostic powers of the biomarkers were further assessed by C-indexes (ranged from 0.76 to 0.96). Moreover, by comparing the 13 survival-related gene lists, seven genes (SLK, API5, BTBD2, PTAR1, VPS37A, EIF2B1, and ZRANB1) were found to be associated with prognosis in a variety of cancers. In particular, SLK was more likely to be cancer-related due to its high missense mutation rate and associated with cell adhesion. Furthermore, after network analysis, EPRS, HNRNPA2B1, BPTF, LRRK1, and PUM1 were demonstrated to have a broad correlation with cancers. In summary, our method has a better integration of multi-omics data that can be extended to the researches of other diseases. And the prognostic biomarkers had a better prognostic power than previous methods. Our results could provide a reference for translational medicine researchers and clinicians.