avocado: A Variant Caller, Distributed

In this paper, we present avocado, a distributed variant caller built on top of ADAM and Spark. avocado’s goal is to provide both high performance and high accuracy in an open source variant calling framework. To achieve this, we implement both local assembly and pileup-based single nucleotide polymorphism (SNP) calling. A key innovation presented in our work involves the development of heuristics for when to choose more expensive assembly-based methods instead of pileup-based methods. Additionally, we introduce the concept of “significant statistics,” a tool for performing incremental joint variant calling.