Bayesian Variant Calling — Probabilistic SNP and Indel Detection
Bayesian Statistical Variant Calling from Sequencing Data · Also known as: Bayesian genotyping, probabilistic variant calling, GATK HaplotypeCaller, Bayesian SNP/indel detection
Bayesian variant calling is a computational pipeline that uses probabilistic inference to identify single-nucleotide polymorphisms (SNPs), insertions, and deletions in a genome by treating sequencing data as evidence and computing posterior probabilities over candidate genotypes. Unlike deterministic threshold-based callers, Bayesian approaches explicitly model sequencing error, mapping uncertainty, and prior genotype frequencies to produce calibrated genotype likelihoods that can be used for downstream filtering and association testing.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Bayesian variant calling whenever you need to identify germline or somatic SNPs and indels from next-generation sequencing data and require calibrated confidence scores rather than binary calls. It is the standard approach for whole-genome sequencing (WGS), whole-exome sequencing (WES), and targeted panel sequencing in human genetics and clinical genomics. Sequencing depth of at least 20x is generally required for reliable germline calling; somatic calling from tumor-normal pairs requires higher depth (50–100x). Do NOT apply to RNA-seq data without specialist tools designed for transcript-aware calling, and do not use standard Bayesian germline callers for somatic variants without switching to somatic-mode pipelines (e.g., Mutect2), as the priors and error models differ fundamentally.
Strengths & limitations
- Provides calibrated, interpretable genotype likelihoods and quality scores that downstream tools can use to filter and weight calls.
- Explicitly models sequencing error and mapping uncertainty, dramatically reducing false-positive rates compared to naive coverage-threshold approaches.
- Haplotype-aware local assembly captures complex indels and nearby variants that pileup-based callers miss.
- Population-level priors (e.g., from gnomAD) improve genotyping accuracy at low-coverage sites and common variants.
- Well-supported by established toolkits (GATK, Strelka2, DeepVariant) with extensive documentation and community benchmarking.
- Computationally intensive — WGS analysis of a single human sample requires hours to days on a typical compute cluster without parallelisation.
- Default genotype priors assume a specific population genetic model; calling in highly inbred organisms, polyploids, or non-human species requires re-tuned priors.
- VQSR requires a large training set of validated variant sites; it is not reliable for small sample sizes (fewer than ~30 samples) or for non-model organisms lacking curated reference variant databases.
- Structural variants (SVs) and copy number variants (CNVs) require separate specialist callers — SNP/indel Bayesian callers are not designed to detect them.
Frequently asked
What is the difference between GATK HaplotypeCaller and DeepVariant?
Both produce Bayesian-style genotype likelihoods, but HaplotypeCaller uses a hand-engineered probabilistic model (PairHMM + Bayes' theorem), while DeepVariant trains a convolutional neural network on images of read pileups to estimate the same genotype probabilities. DeepVariant often achieves lower error rates on Illumina data; HaplotypeCaller remains more flexible for custom priors and non-standard applications. Both output VCF files with phred-scaled likelihoods.
How much sequencing depth do I need?
For germline SNP calling, 30x mean coverage is the standard recommendation for WGS; 20x is often used but increases false-negative rates for heterozygous variants. For somatic calling from tumor-normal pairs, 50–100x on the tumor is typical. Targeted panels often run at 200–500x to ensure sensitivity at clinically relevant loci.
Can I use Bayesian variant calling for RNA-seq data?
Standard DNA-seq variant callers require modification for RNA-seq because splicing creates read-alignment artefacts at exon boundaries, and allelic expression imbalance mimics somatic mutations. GATK provides an RNA-seq variant calling workflow that includes SplitNCigarReads preprocessing, but it is less validated than DNA-seq pipelines and should not be used interchangeably.
What is VQSR and when should I use hard-filtering instead?
Variant Quality Score Recalibration (VQSR) trains a Gaussian mixture model on high-confidence variant sites to distinguish true variants from artifacts. It requires at least 30 samples (ideally hundreds) and validated reference resources (HapMap, 1000 Genomes, dbSNP). For small cohorts, non-human organisms, or targeted panels without suitable training data, hard-filtering using fixed annotation thresholds (QD, FS, SOR, MQ) is the recommended fallback.
My samples are from a non-human polyploid organism. Can I still use Bayesian variant calling?
Yes, but default tools assume diploidy. For polyploids you must specify the ploidy level (GATK supports this via --sample-ploidy) and adjust the genotype priors accordingly. Population-level resources for non-model organisms are rarely available, so VQSR is usually not feasible; rely on hard-filtering and manual inspection of high-priority calls.
Sources
- McKenna, A., Hanna, M., Banks, E., Sivachenko, A., Cibulskis, K., Kernytsky, A., ... & DePristo, M. A. (2010). The Genome Analysis Toolkit: A MapReduce framework for analyzing next-generation DNA sequencing data. Genome Research, 20(9), 1297–1303. DOI: 10.1101/gr.107524.110 ↗
- Rimmer, A., Phan, H., Mathieson, I., Iqbal, Z., Twigg, S. R., WGS500 Consortium, ... & McVean, G. (2014). Integrating mapping-, assembly- and haplotype-based approaches for calling variants in clinical sequencing applications. Nature Genetics, 46(8), 912–918. DOI: 10.1038/ng.3036 ↗
How to cite this page
ScholarGate. (2026, June 3). Bayesian Statistical Variant Calling from Sequencing Data. ScholarGate. https://scholargate.app/en/bioinformatics/bayesian-variant-calling
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Bayesian GWASBioinformatics↔ compare
- Copy Number Variation AnalysisBioinformatics↔ compare
- RNA-seq Differential ExpressionBioinformatics↔ compare
- Sequence AlignmentBioinformatics↔ compare
- Single-cell variant callingBioinformatics↔ compare
- Variant CallingBioinformatics↔ compare