#6161st of 80 in Open modelsFor science, health and research, science, health and medicine
deepvariant
CNN-based germline variant caller that turns BAM/CRAM reads into VCF/gVCF
- Stars
- 3.8k
- License
- BSD-3-Clause
- Last commit
- Mar 2026
- Last release
- Mar 2026
- Language
- Python
3,654 on 22 Apr 20263,829 on 11 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
DeepVariant builds pileup image tensors from aligned reads (BAM or CRAM), classifies each with a convolutional neural network, and writes VCF or gVCF. Pretrained models cover Illumina/Element WGS and WES, PacBio HiFi, ONT R10.4.1 and hybrid PacBio+Illumina, all trained on human data. It also includes DeepTrio for trio and duo calling. It is research code, not intended for clinical use.
Who it is for: Bioinformaticians calling germline SNPs and indels from sequencing data
Strengths
- Model types for WGS, WES, PACBIO, ONT_R104 and HYBRID_PACBIO_ILLUMINA via one run_deepvariant command
- Documented case studies per sequencing platform, plus pangenome-aware and RNA-seq modes
- Won 2020 PrecisionFDA Truth Challenge V2 in several categories, per the README
- Official Docker image; GPU optional, with an experimental CPU/GPU pipelined mode
Weaknesses
- Diploid organisms only; genotypes limited to hom-ref, het and hom-alt
- Bundled models trained on human data only; other species need extra care
- Germline only; somatic calling lives in the separate DeepSomatic repo
- Unix-like systems only; source build scripts target Ubuntu 20.04; external PRs not merged directly
What it needs
- GPU optional
- Docker
- Needs Docker, TensorFlow, Nucleus, GLnexus (DeepTrio merging)
- Models: BAM, CRAM, VCF, gVCF, WGS
What changed
From the release notes. No flag means the note does not say so, not that the upgrade is safe.
Also in Open models
See all 80| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier returning typed decisions with calibrated odds | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 81 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | open_clipPyTorch library for training and running CLIP-style image-text models | 78 out of 100 |