A deep learning solution for detection of homologous recombination deficiency in ovarian cancer using low pass whole-genome sequencing: Evaluation of the analytical performance.

Abstract

e17599 Background: Poly (ADP-ribose) polymerase inhibitors (PARPi) induce synthetic lethality in cells with homologous recombination deficiency (HRD). PARPi treatment revolutionized management of patients with cancer, particularly in types where HRD is common, such as ovarian and breast cancer. However, challenges in the implementation of available methods currently limit adoption of HRD testing in the clinics. Here we present the analytical performance evaluation of the Genomic Integrity Index (GII) (SOPHiA GENETICs SA). Methods: GII is a deep-learning based solution for identification of HRD positive tumors from low-pass whole genome sequencing (lpWGS) data (X1 fold coverage). The analytical performance of the GII was evaluated as positive (PPA), negative (NPA) and overall (OPA) percentage of agreement with the HRD status determined by Myriad myChoice CDx (Myriad Genetic Laboratories, Inc.; US FDA-approved). We generated whole genome sequencing libraries for DNA extracted from 139 ovarian cancer Formalin-Fixed Paraffin- Embedded samples, in 4 independent clinical laboratories. We sequenced to the equivalent of 1X coverage (̃10 million reads, 150 base paired end reads, Illumina) and performed HRD classification data using GII using manufacturer’s recommended thresholds. Results: The GII demonstrated high analytical concordance with Myriad myChoice CDx with 91.7% PPA, 95.5% NPA and 94.5% OPA (Table). The HRD status obtained by GII from a subset of positive and negative samples was 100% reproducible (n = 4) between runs. Conclusions: The HRD status obtained by GII is highly concordant with that obtained from standard method, supporting that GII allows accurate and reproducible detection of HRD from lpWGS data. LpWGS is amongst the most cost-effective and easy to implement genome profiling methods, making it uniquely suited for clinical applications. [Table: see text]