Comparative Evaluation of Mutect2, Strelka2, and FreeBayes for Somatic SNV Detection in Synthetic and Clinical Whole-Exome Sequencing Data

TL;DR

This study compares three tools for detecting genetic mutations in cancer, finding that each performs differently and suggesting that combining tools may improve results.

Contribution

The study provides a comparative evaluation of three somatic SNV detection tools using both synthetic and clinical data, emphasizing the benefits of ensemble approaches.

Findings

Mutect2 had the highest recall in synthetic data with ~99.9% precision and 63.1% recall.

FreeBayes detected the most variants in real samples, but only 5.1% of SNVs were shared across all three tools.

Ensemble approaches using SomaticSeq improved variant detection by leveraging stronger allelic signals.

Abstract

Somatic variant calling is a critical step in cancer genome analysis, but the performance of available tools can vary depending on their underlying algorithms and filtering strategies. We compared three widely used variant callers—Mutect2, Strelka2, and FreeBayes—for their performance in somatic single-nucleotide variant (SNV) detection using both synthetic and real whole-exome sequencing (WES) data. Synthetic data were generated by introducing 4709 SNVs into a variant-free BAM file, while real data consisted of tumor and matched normal WES samples from five ovarian cancer (OC) patients. All callers were run using the nf-core/sarek pipeline with default settings and appropriate filtering. In the synthetic dataset, all tools showed high precision (~99.9%), with Mutect2 achieving the highest recall (63.1%), followed by Strelka2 (46.3%) and FreeBayes (45.2%). In real samples, FreeBayes detected the most variants, and only 5.1% of SNVs were shared across all three tools. We then integrated calls with SomaticSeq in consensus mode (Mutect2 + Strelka2) and kept variants with stronger allelic signals—showing higher VAFs and, typically, higher coverages relative to single-caller only. Caller-exclusive variants showed significant differences in allele frequency and sequencing depth. These results highlight substantial variability in SNV detection across tools. While all showed high specificity, differences in sensitivity and variant profiles underscore the need for context-specific caller selection or ensemble approaches in cancer genomics.

Figures