Predicting the Prognostic Value of Expression in Different Cancers via a Machine Learning Approach.

Translesion synthesis (TLS) is a cell signaling pathway that facilitates the tolerance of replication stress. Increased TLS activity, the particularly elevated expression of TLS polymerases, has been linked to resistance to cancer chemotherapeutics and significantly altered patient outcomes. Building upon current knowledge, we found that the expression of one of these TLS polymerases () is associated with significant differences in cervical and pancreatic cancer survival. These data led us to hypothesize that expression is associated with cancer survival more broadly. However, when cancers were grouped cancer type, expression did not have a significant prognostic value. We presented a binary cancer random forest classifier using 396 genes that influence the prognostic characteristics of in cervical and pancreatic cancer selected via graphical least absolute shrinkage and selection operator. The classifier was then used to cluster patients with bladder, breast, colorectal, head and neck, liver, lung, ovary, melanoma, stomach, and uterus cancer when high expression was associated with worsened survival (Group I) or with improved survival (Group II). This approach allowed us to identify cancers where expression is a significant prognostic factor for survival ( = 0.028 in Group I and = 0.0059 in Group II). Multiple independent validation approaches, including the gene ontology enrichment analysis and visualization tool and network visualization support the classification scheme. The functions of the selected genes involving mitochondrial translational elongation, Wnt signaling pathway, and tumor necrosis factor-mediated signaling pathway support their association with TLS and replication stress. Our multidisciplinary approach provides a novel way of identifying tumors where increased TLS polymerase expression is associated with significant differences in cancer survival.