Functional assessment of all ATM SNVs using prime editing and deep learning

Abstract

Ataxia telangiectasia mutated (ATM), a large gene with 63 exons, plays a critical role in the DNA damage response, and its loss of function increases cancer risk and affects the prognosis of cancer patients. However, interpreting the functional impact of ATM variants remains challenging because most are variants of uncertain significance (VUSs). Here, we assessed the functions of all 27,513 possible single-nucleotide variants (SNVs) in ATM. By using prime editing, we experimentally evaluated the effects of 23,092 SNVs on cell fitness in the presence of olaparib, thereby identifying critical residues. Using cancer genetics data and UK Biobank data, we found that our results are useful for estimating both cancer risk and prognosis. We also developed a deep learning model, DeepATM, which predicted such functional effects of the remaining 4,421 SNVs with unprecedentedly high accuracy. This complete evaluation of ATM variants supports precision medicine and provides a framework for addressing VUSs in other genes.

Keywords: ATM; PARP inhibitor; cancer predisposition; deep learning; precision medicine; prime editor; saturation genome editing; variant of uncertain significance.