Abstract
Minimizing false positives is a critical issue when variant calling as no method is without error. It is common practice to post-process a variant-call file (VCF) using hard filter criteria intended to discriminate true-positive (TP) from false-positive (FP) calls. These are applied on the simple principle that certain characteristics are disproportionately represented among the set of FP calls and that a user-chosen threshold can maximize the number detected. To provide guidance on this issue, this study empirically characterized all false SNP and indel calls made using real Illumina sequencing data from six disparate species and 166 variant-calling pipelines (the combination of 14 read aligners with up to 13 different variant callers, plus four ‘all-in-one’ pipelines). We did not seek to optimize filter thresholds but instead to draw attention to those filters of greatest efficacy and the pipelines to which they may most usefully be applied. In this respect, this study acts as a coda to our previous benchmarking evaluation of bacterial variant callers, and provides general recommendations for effective practice. The results suggest that, of the pipelines analysed in this study, the most straightforward way of minimizing false positives would simply be to use Snippy. We also find that a disproportionate number of false calls, irrespective of the variant-calling pipeline, are located in the vicinity of indels, and highlight this as an issue for future development.
Highlights
Minimizing false positives is a critical issue when variant calling, when the presence of a given variant can inform a clinical decision
Machine-learning approaches to bacterial true-p ositive (TP)/false-p ositive (FP) classification, which could obviate this need for hard filters, are not yet widely available due to the lack of truth sets on which they may be trained
The aim of this study was to identify which positional characteristics – that is, statistics recorded for each position, such as read depth – were disproportionately associated with bacterial FP calls and to produce generalizable recommendations for hard filters broadly applicable across a range of datasets
Summary
Minimizing false positives is a critical issue when variant calling, when the presence of a given variant can inform a clinical decision (for instance, when diagnosing disease [2] or disease susceptibility [3], or genotyping bacterial isolates [4]). Neither circumstance is uncommon when variant calling from bacterial sequencing data. It is routine practice to post-process variant-c all files (VCFs) using hard filter criteria intended to discriminate false-p ositive (FP) from true-positive (TP) calls [11,12,13,14,15]. Hard filters apply the simple principle that certain characteristics are disproportionately represented among the set of false-positive calls and that an empirically determined threshold can maximize the number detected. Machine-learning approaches to bacterial TP/FP classification, which could obviate this need for hard filters, are not yet widely available due to the lack of truth sets on which they may be trained
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
