Overview
DeepVariant is a deep learning-based variant caller that leverages convolutional neural networks to identify genetic variants from DNA sequencing data. It processes aligned reads (BAM/CRAM), transforms them into pileup image tensors, and classifies these tensors using a CNN. The final output is reported in standard VCF or gVCF files. DeepVariant supports germline variant-calling in diploid organisms and has been adapted for somatic calling via DeepSomatic. It's designed for NGS data (Illumina, Element), PacBio HiFi, and Oxford Nanopore. The tool can be used with various case studies, including whole genome/exome sequencing and RNA sequencing. DeepTrio extends DeepVariant for trio/duo analysis. Using Docker is the recommended deployment method, enabling a consistent and reproducible environment. Experimental GPU support accelerates the call_variants stage.
Common tasks