The rapid development of next-generation sequencing (NGS) technologies has lowered the barriers to genomic data generation, resulting in millions of samples sequenced across diverse experimental designs. The growing volume and heterogeneity of these sequencing data complicate the further optimization of methods for identifying DNA variation, especially considering that curated high-confidence variant call sets commonly used to evaluate these methods are generally developed by reference to results from the analysis of comparatively small and homogeneous sample sets. We have developed xAtlas, an application for the identification of single nucleotide variants (SNV) and small insertions and deletions (indels) in NGS data. xAtlas is easily scalable and enables execution and retraining with rapid development cycles. Generation of variant calls in VCF or gVCF format from BAM or CRAM alignments is accomplished in less than one CPU-hour per 30× short-read human whole-genome. The retraining capabilities of xAtlas allow its core variant evaluation models to be optimized on new sample data and user-defined truth sets. Obtaining SNV and indels calls from xAtlas can be achieved more than 40 times faster than established methods while retaining the same accuracy.

xAtlas: Scalable small variant calling across heterogeneous next-generation sequencing experiments