RESUMO
Identification and analysis of clinically relevant strains of bacteria increasingly relies on whole-genome sequencing. The downstream bioinformatics steps necessary for calling variants from short-read sequences are well-established but seldom validated against haploid genomes. We devised an in silico workflow to introduce single nucleotide polymorphisms (SNP) and indels into bacterial reference genomes, and computationally generate sequencing reads based on the mutated genomes. We then applied the method to Mycobacterium tuberculosis H37Rv, Staphylococcus aureus NCTC 8325, and Klebsiella pneumoniae HS11286, and used the synthetic reads as truth sets for evaluating several popular variant callers. Insertions proved especially challenging for most variant callers to correctly identify, relative to deletions and single nucleotide polymorphisms. With adequate read depth, however, variant callers that use high quality soft-clipped reads and base mismatches to perform local realignment consistently had the highest precision and recall in identifying insertions and deletions ranging from1 to 50 bp. The remaining variant callers had lower recall values associated with identification of insertions greater than 20 bp.