google/deepvariant
Deepvariant
DeepVariant is a deep learning genotyping tool and DNA sequence analysis pipeline used to identify single nucleotide polymorphisms and indels from next-generation sequencing data. It functions as a convolutional neural network genetic variant caller that treats genomic read alignments as multi-channel image tensors to determine genotypes.
The system supports specialized analysis workflows including long-read variant calling for circular consensus sequencing and trio-based variant calling to identify inherited or de novo mutations. It enables model optimization for new species or genome contexts through variant model retraining and the use of pre-trained checkpoints.
The project covers a broad range of genomic analysis capabilities, including multiallelic variant resolution, genomic error correction, and haploid genotype adjustment for sex-linked chromosomes. It also provides utilities for consensus sequence determination and haplotype-aware read sorting to increase detection precision.
Features
- Deep Learning Genotyping Tools - Provides a neural network based system for calling single nucleotide polymorphisms and insertions or deletions in genomic data.
- Genomic Variant Classifiers - Uses convolutional neural networks to classify genomic variants by treating read alignments as multi-channel image tensors.
- DNA Sequence Analysis Pipelines - Provides a computational workflow for processing DNA sequencing reads to determine consensus sequences and genotype mutations.
- Long-Read Variant Callers - Provides a specialized analysis tool for identifying genetic mutations from circular consensus sequencing and other long-read technologies.
- Tensor-Based Genotype Representation - Represents overlapping DNA reads as structured multidimensional arrays to allow deep learning models to analyze spatial patterns.
- Trio Variant Callers - Provides a genomic analysis tool that predicts inherited and de novo mutations by comparing related sample sets.
- Genomic Input Encoding - Encodes base qualities and sequence identities into separate image channels for processing by deep neural networks.
- Genomic Model Checkpoints - Provides pre-trained model checkpoints to initialize variant calling for new species via weight fine-tuning.
- Genomic - Optimizes models for new genome contexts or species using specific data and silver standard callsets to improve accuracy.
- Sequence Model Training - Implements training procedures to adjust model weights using labeled examples for optimizing variant calling across different data types.
- Genomic - Optimizes neural network weights for specific species or genome contexts using labeled sequencing data.
- Trio Genomic Analysis - Predicts inherited or de novo mutations by calling variants across related family samples.
- Multiallelic - Resolves complex multiallelic genotypes by processing tensors for each possible combination of alternate alleles.
- Phased Haplotype Calling - Increases precision for indel detection by sorting reads according to their haplotype to provide better context.
- Trio - Predicts genomic variants across related samples to identify inherited or de novo mutations.
- Consensus Sequence Determiners - Identifies correct DNA bases by analyzing matrices of normalized base counts from overlapping reads.
- Genomic Error Correction - Predicts and fixes missing or low-quality bases in DNA sequences by leveraging reference genome patterns.
- Genomic Read Error Correctors - Predicts missing or low-quality bases in DNA sequences by training neural networks on reference genome patterns.
- Haplotype-Aware Read Sorting - Organizes genomic reads by their shared ancestral origin to provide cleaner contextual signals for indel detection.
- Biological Analysis - Identifies genetic variations from DNA sequencing data.
- Variant Calling - Deep learning-based variant caller.
