Comparing the performance of selected variant callers using synthetic data and genome segmentation

Area of Science:

• Genomic medicine
• Bioinformatics
• Cancer research

Background:

• High-throughput sequencing is crucial for precision cancer medicine.
• Validating genomic data analysis is a significant challenge.
• Current manual validation is costly, time-consuming, and selective.

Purpose of the Study:

• To assess the performance of five open-source variant callers.
• To evaluate an ensemble approach for variant calling.
• To compare variant caller accuracy against known ground truth mutations.

Main Methods:

• Utilized four synthetic datasets with known mutations (SNVs, SNPs, SVs).
• Assessed sensitivity, specificity, and balanced accuracy of FreeBayes, VarDict, MuTect, MuTect2, and MuSE.
• Integrated results into an ensemble call set for comparison.

Main Results:

• Ensemble approach showed higher specificity and balanced accuracy with fewer false positives than individual callers.
• MuTect2 performed best, followed by MuSE and MuTect.
• Caller performance generally decreased with increasing data set complexity.

Conclusions:

• Spiking synthetic data with known mutations is an effective, economical validation method.
• This approach offers a viable alternative to expert panel validation.
• Developing multiple lightweight validation methods is key to establishing robust standards for NGS technologies.