Variant calling in genomics: A comparative performance analysis and decision guide.

Abstract

The accurate detection of genetic variants is critical for advancing genomics research and precision medicine. However, this task remains challenging due to pervasive sequencing errors and complex genomic regions. The choice of variant calling software significantly influences results, creating a need for clear, evidence-based guidance. This study aims to provide a performance evaluation and a clear, evidence-based guide for selecting variant callers by benchmarking seven widely used tools, GATK, FreeBayes, DeepVariant, Samtools, Strelka2, Octopus, and Varscan2, highlighting their algorithmic trade-offs. The well-characterized NA12878 genome from the Genome in a Bottle consortium was analyzed. High-coverage whole-genome sequencing data was processed with each variant caller, and the resulting variant calling files were benchmarked against a gold-standard reference. Performance was assessed using precision, recall, and F1-score on a chromosome 20 subset and on full whole-genome data. The analysis revealed that DeepVariant's deep learning approach achieved the highest precision (0.7869) and F1-score (0.8754) on chromosome 20. For whole-genome analysis, Strelka2 excelled in precision (0.8326), while Octopus demonstrated superior recall (0.9838). FreeBayes exhibited high sensitivity but lower precision, underscoring a key trade-off. There is no universally superior variant caller; the optimal choice depends on the specific research objectives, whether prioritizing precision, recall, or computational efficiency. This study serves as a crucial evidence-based resource for researchers and clinicians, enabling informed tool selection.