Benchmarking within-sample minority variant detection with short-read sequencing in M. tuberculosis

Area of Science:

• Genomics
• Bioinformatics
• Microbiology

Background:

• Low-frequency variants are vital in research and health.
• Distinguishing these variants from errors requires optimal bioinformatics.
• 700 Mycobacterium tuberculosis strains were used to benchmark variant callers.

Purpose of the Study:

• To benchmark variant callers for accuracy in detecting low-frequency variants.
• To identify the best bioinformatic approach for variant calling.
• To develop an error model for filtering low-frequency variant calls.

Main Methods:

• Benchmarking seven variant callers on simulated short-read whole genome sequencing data.
• Simulating variants across different genomic regions, allele frequencies, and sequencing depths.
• Developing a new low-frequency error model using read mapping and quality metrics.

Main Results:

• FreeBayes achieved the highest pooled F1 score (0.86) in drug resistance regions.
• Variant caller performance was lower in repetitive regions with reference bias.
• The developed error model excluded 49% of false variants and <1% of true variants when paired with FreeBayes.

Conclusions:

• The study provides evidence for best practices in low-frequency variant calling.
• FreeBayes is recommended as the optimal tool for low-frequency variant detection.
• A novel error model effectively reduces false positive low-frequency variant calls.