Powering Toxicogenomic Studies by Applying Machine Learning to Genomic Sequencing and Variant Detection

Abstract

AbstractSince the advent of massively parallel high-throughput sequencing, also known as next-generation sequencing (NGS), the cost of DNA sequencing has steadily been going down over time. The lowered cost and increased accessibility of NGS have enabled research and clinical scientists from institutions large and small to sequence any samples of interest to interrogate for variants across the whole genome, something unthinkable a decade earlier. With massive amounts of sequencing data being generated and sequencing technologies being refined, the interpretation of the data lies in the cross section of biology, computer science, and big data analytics. Machine learning, including deep learning algorithms, have increasingly become the method of choice for bioinformatics researchers to build efficient and accurate variant detection software, which is critical to power genomic and toxicogenomic studies. The availability of machine learning libraries open-sourced by software giants like Google and Facebook, combined with research consortia’s effects to create publicly available reference samples and training data, have made machine learning in genomic sequencing a reality. In this chapter, we will discuss a number of different variant detection software, e.g., DeepVariant, SomaticSeq, and NeuSomatic, each of which relies on a different machine learning strategy to produce accurate variant calls.