Sparse Modeling of Genomic Landscape Identifies Pathogenic Processes and Therapeutic Targets in Metastatic Breast Cancer

Abstract

Breast cancer is a heterogeneous disease and ranks as one of the most lethal and frequently detected disease in the world. It poses significant challenges for precision therapy. To better decipher the patterns of heterogeneous nature in human genome and converge them into common functionals, mutational signatures are introduced to define the types of DNA damage, repair and replicative mechanisms that shape the genomic landscape of each cancer patient. In this study, we developed a deep learning (DL) model, MetaWise 2.0, based on pruning technology that improved model generalization with deep sparsity. We applied it to patient samples from multiple sequencing studies, and identified statistically significant mutational signatures associated with metastatic progression using Shapley additive explanations (SHAP). We also employed gene cumulative contribution abundance analysis to link the mutational signatures with relevant genes, which could unearth the shared molecular mechanisms behind tumorigenesis and metastasis of each patient and lead to novel therapeutic target identification. Our study illustrates that MetaWise 2.0 is an effective DL tool for discovering clinically meaningful mutational signatures in metastatic breast cancer (MBC) and relating them directly to relevant biological functions and gene targets. These findings could facilitate the development of novel therapeutic strategies and improve the clinical outcomes for individual patients.