A synthetic-diploid benchmark for accurate variant-calling evaluation.

Area of Science:

• Genomics
• Bioinformatics

Background:

• Existing benchmark datasets for variant calling accuracy are often biased.
• These datasets are derived from consensus calls of short-read algorithms.
• This bias favors genomic regions easily accessible by current short-read sequencing technologies.

Purpose of the Study:

• To develop a novel benchmark dataset for evaluating variant-calling accuracy.
• To provide a less biased and more realistic estimation of small variant calling error rates.
• To overcome limitations of existing consensus-based benchmark datasets.

Main Methods:

• Utilized de novo PacBio assemblies of two fully homozygous human cell lines.
• Generated a new benchmark dataset from these high-quality assemblies.
• Applied this dataset to estimate small-variant-calling error rates.

Main Results:

• The new dataset offers a more accurate estimate of variant-calling error rates.
• It provides a less biased evaluation compared to existing methods.
• Demonstrates the utility of long-read sequencing assemblies for benchmark creation.

Conclusions:

• De novo PacBio assemblies of homozygous cell lines are valuable for creating unbiased genomic benchmarks.
• This new dataset improves the evaluation of variant-calling algorithms.
• Highlights the need for benchmark datasets that reflect real-world genomic complexity.