Pangenome-aware DeepVariant.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Population-scale genomics provides crucial prior knowledge for genomic analyses.
• The human pangenome reference enhances read mapping and structural variant genotyping.
• Existing variant callers often rely on linear references, which can limit accuracy.

Purpose of the Study:

• To introduce pangenome-aware DeepVariant, a novel variant caller.
• To leverage pangenome references for improved variant calling accuracy.
• To evaluate the performance of pangenome-aware DeepVariant across different sequencing platforms and read mappers.

Main Methods:

• Developed pangenome-aware DeepVariant, a variant caller utilizing a pangenome reference and sample-specific read alignments.
• Generated pileup images of reads and pangenome haplotypes near potential variants.
• Employed a Convolutional Neural Network (CNN) for genotype inference.
• Assessed performance on various short-read sequencing platforms and read mappers.

Main Results:

• Pangenome-aware DeepVariant consistently outperformed linear-reference-based DeepVariant across all tested settings.
• Error rates were reduced by up to 25.5% using the pangenome-aware approach.
• Element reads combined with pangenome-aware DeepVariant achieved 23.6% higher accuracy than existing methods.

Conclusions:

• Pangenome-aware DeepVariant effectively distinguishes true variant signals from sequencing or alignment noise.
• The use of pangenome references significantly enhances variant calling accuracy.
• This method represents a substantial advancement in population genomics and personalized medicine.