Integrating gene mutation spectra from tumors and the general population with gene expression topological networks to identify novel cancer driver genes.

Discovering cancer driver genes is critical for improving survival rates. Current methods often overlook the varying functional impacts of mutations. It is necessary to develop a method integrating mutation pathogenicity and gene expression data, enhancing the identification of novel cancer drivers. To predict cancer drivers, we have developed a framework (DGAT-cancer) that integrates the pathogenicity of somatic mutation in tumors and germline variants in the healthy population, with topological networks of gene expression in tumors, and the gene expressions in tumor and paracancerous tissues. This integration overcomes the limitations of current methods that assume a uniform impact of all mutations by leveraging a comprehensive view of mutation function within its biological context. These features were filtered by an unsupervised approach, Laplacian selection, and combined by Hotelling and Box-Cox transformations to score genes. By using gene scores as weights, Gibbs sampling was performed to identify cancer drivers. DGAT-cancer was applied to seven types of cancer cohorts, and achieved the best area under the precision-recall curve (AUPRC ranging from 0.646 to 0.862) compared to five commonly used methods (AUPRC ranging from 0.357 to 0.629). DGAT-cancer has identified 505 cancer drivers. Knockdown of the top ranked gene, EEF1A1 indicated a ~ 41-50% decrease in glioma size and improved the temozolomide sensitivity of glioma cells. By combining heterogeneous genomics and transcriptomics data, DGAT-cancer has significantly improved our ability to detect novel cancer drivers, and is an innovative approach revealing cancer therapeutic targets, thereby advancing the development of more precise and effective cancer treatments.