Metastatic tracking of cancer subclones for collateral lethality prediction

University Of Pittsburgh At Pittsburgh, Pittsburgh PA

Investigators

Abstract

PROJECT SUMMARY/ABSTRACT Late-stage metastatic breast cancer remains a leading cause of cancer-related deaths, largely due to genomic instability driving tumor evolution and therapeutic resistance. While DNA structural variations (SVs) are known to be crucial in this process, existing tools for tracking these changes across tumor subclones and linking them to functional effects are inadequate. Our group recently showed that these genomic alterations, especially deletions, may be ideal landmarks for tracking tumor phylogeny and can be leveraged for treatment target identification. By leveraging the unique combination of genomic data from the AURORA US Metastasis Project and the University of Pittsburgh's Hope for OTHERS - Rapid Autopsy Program (HfO), this study proposes an innovative integration of machine learning techniques with a comprehensive autopsy program to identify diverse and recurrent patterns of gene deletions in metastatic disease. Central to this project is the hypothesis that genomic deletions accumulate over the process of unstable genomic evolution and that these identifiable patterns of genomic variation can be targeted through advanced computational analysis. Our approach utilizes Transformer models and cutting-edge phylogeny tools to analyze multi-omic data, creating a detailed landscape of the genomic evolution in metastatic breast cancer. This proposal addresses the challenges of limited dataset size and complexity in machine learning applications by directly collating said data from clinic to hospice and collecting the tissue required to capture the complete clinical and biological context of metastatic breast cancer. It additionally leverages the unique possibility of growing organoids from those same tissue samples for orthogonal wet lab validation of computational findings. The project seeks to answer critical questions: 1) Are genomic deletions truncal clonal features in cancer genomes that are persistent over time? 2) What potential therapeutic targets are hidden within the genomic patterns of gene deletions as cancer genomes lose fundamental redundancy pathways? This project advances the translational implementation of machine learning in elucidating complex genetic patterns in cancer evolution to improve the precision treatment of late-stage metastatic breast cancer. This interdisciplinary study lies at the crossroads of machine learning, cancer biology, and clinical research and represents a unique collaboration between the University of Pittsburgh and Carnegie Mellon University. It promises to make significant contributions to the fields of oncology, genomics, and computational biology, ultimately improving clinical outcomes for patients with metastatic breast cancer. A fundamental issue in this area of research is defining the best methods to process vast biological data forms into machine-readable structures and interpretable findings; this proposal will provide essential training in developing and utilizing those methods to support the advancement of such machine tools in the clinic and the laboratory.