Comprehensive characterization of DNA polymerases driver variants using large genomic dataset

POLE or POLD1 deficiency (PPD) leads to an exceptionally high mutation burden and distinct mutational signatures that can inform immunotherapy responses. While some PPD-associated variants have been identified, many remain classified as variants of unknown significance. We analyzed 235,161 tumors across multiple datasets (GENIE, TCGA, and CPC) to develop a comprehensive catalog of PPD driver variants, using an iterative Support Vector Machine algorithm that incorporates both the number of SNVs and PPD-associated mutational signatures. Our iterative approach identified 581 PPD tumors and 65 driver variants, validating many previously reported variants and expanding this list. Thus, we curated the most comprehensive list of PPD drivers up to date. Notably, while the majority of POLE driver variants are located within the exonuclease domain (n = 39), we identified 3 new driver variants outside this domain. Additionally, 7.2% (n = 42) of the PPD samples did not harbor any driver variants, highlighting the need to classify tumors as PPD rather than relying solely on observed driver variants when making treatment decisions.