Title: Skyhawk: an artificial neural network-based discriminator for reviewing clinically significant genomic variants
Authors: Ruibang Luo; Tak-Wah Lam; Michael C. Schatz
Addresses: Department of Computer Science, The University of Hong Kong, Hong Kong, China ' Department of Computer Science, The University of Hong Kong, Hong Kong, China ' Department of Computer Science, Johns Hopkins University, Baltimore, MD, 21211, USA
Abstract: Genome sequencing has become an important tool in clinical practice. However, variant interpretation remains the bottleneck and may take a specialist several hours of work per patient. On average, one-fifth of this time is spent on visually confirming the authenticity of the candidate variants. We developed Skyhawk, an artificial neural network (ANN)-based discriminator that mimics the process of expert review on clinically significant genomics variants. Skyhawk runs in less than 1 min to review 10,000 variants, and about 30 min to review all variants in a typical whole-genome sequencing sample. Among the false positive singletons identified by GATK HaplotypeCaller, UnifiedGenotyper and 16GT in the HG005 GIAB sample, 79.7% were rejected by Skyhawk. Worked on the variants with unknown significance (VUS), Skyhawk marked most of the false positive variants for manual review and most of the true positive variants no need for review. Skyhawk is freely available at https://github.com/aquaskyline/Skyhawk.
Keywords: clinical decision support; variant validation; ANN; artificial neural network; third-generation sequencing; variant calling; variant interpretation.
International Journal of Computational Biology and Drug Design, 2020 Vol.13 No.5/6, pp.431 - 437
Received: 23 Sep 2019
Accepted: 21 Apr 2020
Published online: 31 Mar 2021 *