DAVI: Deep learning-based tool for alignment and single nucleotide variant identification. Issue 2 (27th May 2020)

• Record Type: Journal Article
• Authors: Gupta, G; Saini, S
• Is Part Of: Machine learning: science and technology. Volume 1:Issue 2(2020)
• Publication Date: 2020-05-27
• Subjects: genome -- deep learning -- single nucleotide variant caller -- genome-assembly -- pipeline -- convolutional neural network -- recurrent neural network; 006.31
• DOI: 10.1088/2632-2153/ab7e19
• Languages: English
• ISSNs: 2632-2153

Abstract

Abstract: Next-generation sequencing (NGS) technologies have provided affordable but errorful ways to generate raw genetic data. To extract variant information from billions of NGS reads is still a daunting task which involves various hand-crafted and parameterized statistical tools. Here we propose a deep neural networks (DNN) based alignment and single nucleotide variant (SNV) identifier tool known as DAVI: deep alignment and variant identification. DAVI consists of models for both global and local alignment and for variant calling. We have evaluated the performance of DAVI against existing state-of-the-art tool sets and found that its accuracy and performance is comparable to existing tools used for bench-marking. We further demonstrate that while existing tools are based on data generated from a specific sequencing technology, the models proposed in DAVI are generic and can be used across different NGS technologies as well as across different species. The use of DAVI will therefore help non-human sequencing projects to benefit from the wealth of human ground truth data. Moreover, this approach is a migration from expert-driven statistical models to generic, automated, self-learning models.