UNITe: Exploiting Conserved links between Oncogenesis and Homeostasis to Identify Novel Cancer Drivers

Piyush Agrawal¹Annan Timon²Vishaka Gopalan³Arashdeep Singh³Sridhar Hannenhalli^3*

• ¹SRM Institutes for Medical Science, SRM University, Chennai, India
• ²University of Pennsylvania, Philadelphia, United States
• ³National Institutes of Health National Cancer Institute, Bethesda, United States

Background: Similarities between oncogenesis and several homeostatic processes (HP) -- wound healing, regeneration, and cellular stress response, have long been recognized. However, the molecular underpinning of these similarities is not fully understood. While several molecular aspects of HP are evolutionarily conserved, different species exhibit substantial variation in the genes involved in these HP as well as in their predisposition to cancer. Methods: We leveraged 75 published (with 321 experiments across 14 species) experimental datasets of genes implicated in HP across multiple species from Gene Expression Omnibus (GEO), pan-cancer (32 cancer types) multi-omics datasets from TCGA, and several benchmarking datasets from public repositories such as GTEx, MSigDb, COSMIC, DepMap, as well as literature to comprehensively investigate links between conserved aspects of HP and human cancers. We performed several analyses to understand broad mechanistic links between cancer and the homeostatic processes. Results: We compiled a high-confidence conserved consensus gene sets for SR, WH, and RG - jointly as HP for 'homeostatic process'. We found that broadly across cancers the HP genes exhibit elevated mutations, including copy number aberrations, differential gene expression in the tumor compared to healthy tissue, and are associated with patient survival. In the human protein interaction network, HP genes cluster by the process type as well as with the known cancer driver genes. Leveraging this observation, here we present a tool - UNITe (Uncovering Network-based Interactions between Homeostatic processes and Tumorigenesis), which predicts cancer drivers based solely on network-proximity to HP genes, with an AUROC of 0.81, far better than several current approaches, and across multiple benchmark datasets. Applying UNITe genome-wide, we report several novel potential cancer drivers and validate them using multiple lines of evidence. Conclusions: Overall, we present a first comparative analysis of cancer drivers with conserved homeostatic processes, suggesting a complementary approach to prioritize cancer drivers. Model and the codes are freely available for public usage at https://github.com/hannenhalli-lab/conserved_links_homeostasis_oncogenesis.