Refinement of genetic variants needs attention

TL;DR

This paper presents VariantTransformer, a Transformer-based deep learning model that automates the refinement of genetic variants from low-coverage sequencing data, significantly improving accuracy over traditional methods.

Contribution

Introduction of VariantTransformer, a novel deep learning model that enhances variant calling refinement directly from VCF files in low-coverage data, outperforming heuristic filters.

Findings

Achieved 89.26% accuracy and 0.88 ROC AUC on low-coverage data.

Outperformed traditional heuristic filters in variant refinement.

Approached the performance of state-of-the-art AI variant callers.

Abstract

Variant calling refinement is crucial for distinguishing true genetic variants from technical artifacts in high-throughput sequencing data. Manual review is time-consuming while heuristic filtering often lacks optimal solutions. Traditional variant calling methods often struggle with accuracy, especially in regions of low read coverage, leading to false-positive or false-negative calls. Here, we introduce VariantTransformer, a Transformer-based deep learning model, designed to automate variant calling refinement directly from VCF files in low-coverage data (10-15X). VariantTransformer, trained on two million variants, including SNPs and short InDels, from low-coverage sequencing data, achieved an accuracy of 89.26% and a ROC AUC of 0.88. When integrated into conventional variant calling pipelines, VariantTransformer outperformed traditional heuristic filters and approached the performance of state-of-the-art AI-based variant callers like DeepVariant. Comparative analysis demonstrated VariantTransformer's superiority in functionality, variant type coverage, training size, and input data type. VariantTransformer represents a significant advancement in variant calling refinement for low-coverage genomic studies.