Abstract 889: Machine learning-enhanced targeted versus whole-exome sequencing as a guide to cancer care

Abstract

Abstract Targeted DNA sequencing of cancer-associated genes has been widely adopted clinically. In addition, the increasing availability of targeted and whole-exome/genome sequencing (WES/WGS)-derived big data suggests avenues of computationally enhancing power of targeted assays via machine learning. We directly compared targeted (MSK-IMPACT) and WES in a cohort of 1,483 cancer patients using the same DNA libraries. We found high sensitivity and specificity of mutation detection by WES for overlapping genes. Comparatively higher estimates of tumor mutation burden by MSK-IMPACT arose from enrichment of oncogenic mutations in the targeted gene set. WES had minimal value in identifying actionable alterations beyond those from MSK-IMPACT, whereas RNA-sequencing identified additional targetable fusions that led to durable responses. We developed a deep learning-based algorithm (DeepSig) that enhances power of mutational signature detection, enabling detection of a select group of signatures with mutation counts of ~5 or less. Hypermutation-associated signatures, including POLE, temozolomide, and mismatch repair deficiency could be detected robustly using targeted sequencing data. In summary, WES had limited clinical value over targeted sequencing panels and future clinical diagnostic development should focus on transcriptome and WGS that can detect additional signatures and gene fusions. Citation Format: Hyung Jun Woo, Shweta S. Chavan, Aphrothiti J. Hanrahan, Allison L. Richards, Anne Marie Noronha, Ahmet Zehir, Alexander Drilon, Michael F. Berger, David B. Solit, Mark T. Donoghue. Machine learning-enhanced targeted versus whole-exome sequencing as a guide to cancer care [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 889.