“Performance Evaluation of Variant Calling Tools for Human and Microbial Genomes”

Analyzing the data deluge generated by next generation sequencing technologies is a challenging task. Variant calling aids in characterizing phenotype of an organism. The objective of this work is to compare popular variant calling protocols available in terms of their accuracy for human and microbial genomes. Three popular read aligners viz. BWA-MEM, Bowtie2 and NovoAlign were used for the human genome benchmark dataset of Genome in a Bottle (GIAB). Variant calling was performed using four popular variant callers viz. samtools-bcftools, VarScan2, GATK-HC and DeepVariant. For microbial genomes, BWA-MEM was used for reference mapping followed by variant calling using above-mentioned tools (excluding DeepVariant) along with FreeBayes and PILON. The predicted variants were assessed in terms of true positives, false positives and false negatives. Consensus approach was used for identifying SNPs (predicted by more than one variant calling pipeline) in the false positive dataset that satisfied the threshold criteria to be included as true positives. Precision and Recall were measured for evaluating each of the variant calling pipelines. Variant calling pipeline using BWA-MEM for read mapping along with either VarScan2 or samtools-bcftools predicted optimal number of single nucleotide polymorphisms (SNP) for high coverage human datasets. For Indel calling, GATK-HC with any aligner was preferred over VarScan2 and samtools-bcftools. Variant calling using bcftools, Pilon and GATK-HC with default parameters for heterogeneous microbial datasets revealed optimal performance. SNP calling with minimum errors was evident when
$\sim\mathbf{95}\%$
of the samples from a population was taken into cognizance for variant prediction. Predicted variants were validated by comparing with gold standard data which aided in improved identification of actionable high confidence variants, reduced prediction of false positives and this approach has proved to be valuable for prioritization of actionable variants.