Repolode
Submit
google

google/deepvariant

A deep learning variant caller that identifies genetic variants from aligned next-generation DNA sequencing reads and outputs VCF/gVCF files.

View on GitHub
Stars
3,827
Forks
808
Open issues
5
Last commit
12 days ago

What is it?

DeepVariant takes aligned sequencing reads in BAM or CRAM format, converts them into pileup image tensors, classifies each tensor with a convolutional neural network, and writes the results as a standard VCF or gVCF file. It is aimed at germline variant calling in diploid organisms.

The project covers a range of sequencing technologies, including Illumina and Element short reads, PacBio HiFi, Oxford Nanopore R10.4.1, Complete Genomics, and Roche SBX, with separate models and case studies for whole genome, whole exome, and RNA data. Its bundled models are trained on human data. A companion tool, DeepTrio, extends the approach to trios and duos, and DeepSomatic adapts it for somatic calling.

DeepVariant is distributed as a Docker image, prebuilt binaries, and build-from-source scripts for Ubuntu 20.04. It runs on Unix-like systems with Python 3.10, and supports GPU and TPU acceleration. It depends on Nucleus, Google's library for reading and writing genomics file formats.

Key features

  • 01Calls germline SNVs and indels from BAM or CRAM alignments and outputs VCF or gVCF.
  • 02Provides pretrained models for WGS, WES, PacBio, ONT R10.4.1, and hybrid PacBio plus Illumina data.
  • 03Runs as a Docker image or prebuilt binaries, with GPU and TPU support.
  • 04Includes an optional HTML VCF statistics report.
  • 05Offers DeepTrio for trio and duo calling, with GLnexus for merging outputs.

Use cases

  • Calling variants from whole genome or whole exome sequencing of human samples.
  • Analyzing PacBio HiFi or Oxford Nanopore reads for variant discovery.
  • Detecting inherited variants in family trios with DeepTrio.
  • Running variant calling on cloud or on-premise hardware with accelerators.

Topics

#genomics#bioinformatics#deep-learning#variant-calling#dna-sequencing#vcf#tensorflow#deep-neural-network#deepvariant#dna

FAQ

What is deepvariant?
A deep learning variant caller that identifies genetic variants from aligned next-generation DNA sequencing reads and outputs VCF/gVCF files. DeepVariant takes aligned sequencing reads in BAM or CRAM format, converts them into pileup image tensors, classifies each tensor with a convolutional neural network, and writes the results as a standard VCF or gVCF file. It is aimed at germline variant calling in diploid organisms.
Is deepvariant free to use? What license does it have?
deepvariant is open source under the BSD-3-Clause license. Check the license terms before using it in commercial products.
Is deepvariant still maintained?
Actively maintained. The last commit to deepvariant was on Oct 26, 2026.
What can I use deepvariant for?
Calling variants from whole genome or whole exome sequencing of human samples. Analyzing PacBio HiFi or Oxford Nanopore reads for variant discovery. Detecting inherited variants in family trios with DeepTrio. Running variant calling on cloud or on-premise hardware with accelerators.
Where can I get deepvariant?
The source code and installation instructions are on GitHub: https://github.com/google/deepvariant

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects