DeepVariant

DeepVariant is a deep-learning analysis pipeline from Google that calls genetic variants from next-generation DNA sequencing data. It reframed variant calling as an image-classification problem and set new accuracy standards for genomics.

Overview

DeepVariant takes aligned reads in BAM or CRAM format, converts the read pileups around candidate sites into image-like tensors, classifies each tensor with a convolutional neural network, and writes the resulting variant calls to standard VCF or gVCF files. It supports germline variant calling in diploid organisms across the major sequencing platforms: Illumina and Element short reads for whole genomes and exomes, PacBio HiFi long reads, Oxford Nanopore R10.4.1 data, and Complete Genomics and Roche SBX data, each with documented case studies. Production workflows are distributed as Docker and Singularity images, and the tool pairs with companion projects for trio analysis and joint genotyping. DeepVariant is open source under the BSD-3-Clause license, has over 3,800 stars on GitHub, and is actively maintained, with releases continuing into 2026.

Key Features

• CNN-based variant calling from sequencing read pileups
• Models for whole-genome, exome, PacBio, and Nanopore data
• Standard VCF and gVCF outputs for downstream pipelines
• Ready-to-run Docker and Singularity workflows
• Documented case studies for each sequencing platform
• Extensively benchmarked accuracy in genomics studies

💳 Pricing

DeepVariant is free and open source under the BSD-3-Clause license. There are no licensing fees; costs are limited to your own sequencing and compute.

Who Is It For

Bioinformatics teams, clinical and research genomics labs, and sequencing providers who need accurate, reproducible germline variant calling at scale.