NGS & Sequencing Published: July 23, 2026 By VigyanLLM Research Team

Variant Calling in NGS — A Practical Guide for 2026

Variant calling identifies genetic variants — SNPs, indels, and structural variants — from next-generation sequencing data. A practical guide to the complete workflow, tools, and best practices for 2026.

What is variant calling in NGS?

Variant calling is the computational process of identifying genetic variants — single nucleotide polymorphisms (SNPs), insertions and deletions (indels), and structural variants — by comparing aligned sequencing reads to a reference genome. It is a core step in every NGS analysis pipeline, from clinical diagnostics to population genomics.

What Is Variant Calling?

Variant calling is the bioinformatics process of identifying genetic differences between a sequenced sample and a reference genome. These differences — or genetic variants — include single nucleotide polymorphisms (SNPs), small insertions and deletions (indels), copy number variants (CNVs), and structural variants (SVs). Variant calling is the central analytical step in every next-generation sequencing (NGS) pipeline, transforming raw sequencing data into actionable biological insights.

The process involves comparing aligned sequencing reads to a reference genome, identifying positions where the sample differs from the reference, and applying statistical models to distinguish true variants from sequencing errors. Modern variant callers use Bayesian inference, machine learning, or heuristic methods to make these calls with high accuracy.

Key Concept

Variant calling is NOT the same as variant annotation. Calling identifies where the genome differs from the reference. Annotation determines what the functional consequence of that difference is — missense, nonsense, splice site, regulatory, or intergenic.

The Complete Variant Calling Workflow

1. Read Alignment (Mapping)

Raw sequencing reads are aligned to a reference genome using aligners like BWA-MEM, Bowtie2, or Minimap2. The output is a BAM or SAM file containing read positions, mapping quality scores, and alignment information. Proper alignment is critical — poor mapping leads to false variant calls.

2. Pre-Processing

Aligned reads undergo several quality improvement steps: Mark Duplicates (Picard MarkDuplicates or samtools markdup) tags PCR and optical duplicate reads. Base Quality Score Recalibration (BQSR) recalibrates base quality scores using known variant sites. Indel Realignment realigns reads around indels to reduce false positives. These steps are standard in the GATK Best Practices pipeline.

3. Variant Discovery

The core step. Variant callers analyze the aligned read pileup at each genomic position and apply statistical models to determine if a variant is present. Key tools include:

GATK HaplotypeCaller — Industry standard for germline SNP/indel calling. Uses a de Bruijn graph-based local assembly and Bayesian likelihood model.
Mutect2 — GATK's somatic caller, designed for tumor-normal paired analysis. Accounts for cross-sample contamination and tumor heterogeneity.
FreeBayes — Bayesian genetic variant detector that calls variants from multiple samples simultaneously. Handles complex multi-allelic sites well.
DeepVariant — Google's deep learning-based caller that uses a convolutional neural network on pileup images. Produces highly accurate calls, especially in difficult regions.
Strelka2 — Fast and accurate small variant caller for germline and somatic analysis. Uses a novel log-likelihood model with efficient filtering.

4. Filtering & Quality Control

Raw variant calls contain false positives. Standard filters include: QUAL score (keep variants with Phred-scaled quality > 30), read depth (DP > 10), genotype quality (GQ > 20), strand bias (FS < 60 for SNPs, < 200 for indels), and mapping quality (MQ > 40). The GATK VQSR (Variant Quality Score Recalibration) tool uses machine learning to filter variants based on multiple annotations simultaneously.

5. Variant Annotation

Filtered variants are annotated with functional impact using tools like ANNOVAR, SnpEff, or VEP. Annotations include: gene name and transcript ID, variant type (missense, nonsense, synonymous, splice site), predicted protein change, population allele frequency from gnomAD, and clinical significance from ClinVar.

Germline vs Somatic Variant Calling

Feature	Germline Calling	Somatic Calling
Sample type	Single sample (blood, saliva)	Tumor-normal paired or tumor-only
Expected VAF	50% (heterozygous) or 100% (homozygous)	1-50% (tumor heterogeneity + stroma)
Key variant callers	GATK HaplotypeCaller, DeepVariant, FreeBayes	Mutect2, Strelka2 somatic, VarScan2
Reference databases	gnomAD, 1000 Genomes, dbSNP	COSMIC, TCGA
Filtering strategy	Hard filters + VQSR	Panel of normals, contamination filtering, orientation bias
Primary applications	Inherited disease, pharmacogenomics, population genetics	Cancer genomics, liquid biopsy, tumor evolution

Variant Calling Tools Compared

Tool	Method	Strengths	Limitations	Best For
GATK HaplotypeCaller	Bayesian + local assembly	Industry standard, excellent documentation, best practices framework	Slow for large cohorts, complex workflow	Germline SNP/indel calling
Mutect2	Bayesian with contamination model	Excellent somatic calling, handles FFPE data well	Requires matched normal, memory intensive	Somatic mutation detection
DeepVariant	CNN deep learning	Highest accuracy, low false positives, simple pipeline	GPU required for large datasets	High-precision germline calling
FreeBayes	Bayesian haplotype-based	Fast, handles multi-allelic sites, multi-sample calling	Less accurate in high-depth regions	Rapid screening, microbial variant calling
Strelka2	Likelihood model	Fast, accurate, low memory usage	Limited structural variant detection	Production pipelines, both germline and somatic

Key Applications

Clinical Diagnostics

Variant calling is used to identify disease-causing mutations in patients with suspected genetic disorders. Clinical workflows require high sensitivity and specificity, with orthogonal validation (Sanger sequencing) of all reported variants. The AMP/ASCO/CAP guidelines provide standards for clinical variant interpretation.

Cancer Genomics

Somatic variant calling from tumor sequencing identifies driver mutations, guides targeted therapy selection, and monitors minimal residual disease. Key challenges include tumor heterogeneity, low tumor purity, and clonal hematopoiesis artifacts. Liquid biopsy variant calling requires ultra-deep sequencing (100,000x) and specialized error suppression techniques.

Population Genomics

Large-scale variant calling across thousands of samples (e.g., gnomAD, 1000 Genomes, UK Biobank) powers GWAS, rare variant association studies, and understanding human genetic diversity. Joint calling across all samples simultaneously improves low-frequency variant detection.

Infectious Disease Genomics

Variant calling from pathogen sequencing identifies drug resistance mutations (e.g., HIV, hepatitis, tuberculosis), tracks transmission chains, and monitors viral evolution. Pathogen variant calling often uses specialized tools due to high genetic diversity and variable coverage.

Pharmacogenomics

Identifying known pharmacogenetic variants (e.g., CYP2C19, CYP2D6, VKORC1, TPMT) from sequencing data enables personalized drug dosing and prevention of adverse drug reactions. Allele-specific genotyping from variant calls is translated into actionable prescribing guidelines.

Common Challenges & Troubleshooting

High false positive rate: Often caused by sequencing errors in homopolymer regions, misalignment around indels, or poor base quality scores. Mitigate with BQSR, hard filtering, and VQSR.

Low sensitivity in difficult regions: GC-rich regions, segmental duplications, and pseudogenes are poorly covered by short reads. Use long-read sequencing or targeted amplicon sequencing with deep coverage.

Batch effects: Variant calls differ across sequencing runs, capture kits, and bioinformatics pipeline versions. Joint calling and consistent pipeline versions minimize batch effects in cohort studies.

Copy number variant calling: Short-read NGS has limited sensitivity for CNVs. Use dedicated tools (CNVnator, Delly, Manta) or array-based methods for robust CNV detection.

Design Primers for Targeted Variant Validation with VigyanLLM

Use VigyanLLM free automated primer design to validate called variants by PCR and Sanger sequencing. Design primers around any genomic coordinate with automated specificity checking via BLAST.

Design Primers Free →

About the Author — VigyanLLM Research Team

The VigyanLLM Research Team combines expertise in molecular biology, bioinformatics, and artificial intelligence. Our team develops NGS analysis pipelines, variant calling workflows, and bioinformatics tools at VigyanLLM.

Frequently Asked Questions

What is variant calling in bioinformatics?

Variant calling is the process of identifying differences between a sample genome and a reference genome from next-generation sequencing (NGS) data. It detects single nucleotide variants (SNVs), insertions, deletions (indels), and structural variants. Tools like GATK HaplotypeCaller, FreeBayes, and DeepVariant compare aligned reads to a reference to identify genomic variations.

What is the difference between SNVs, indels, and structural variants?

SNVs (single nucleotide variants) are changes to a single base pair, such as A to G. Indels are small insertions or deletions of 1-50 base pairs. Structural variants are larger rearrangements involving 50+ base pairs, including duplications, inversions, translocations, and copy number variations. Each type requires different detection algorithms.

What tools are used for variant calling?

GATK HaplotypeCaller is the most widely used tool for germline variant calling and is the industry standard for clinical genomics. FreeBayes is a popular open-source alternative for haplotype-based calling. DeepVariant uses deep learning for variant detection. For somatic variants, Mutect2 (GATK) and VarScan2 are commonly used. Strelka2 handles both germline and somatic calling.

What is VCF format and how is it used in variant calling?

VCF (Variant Call Format) is the standard file format for storing variant calls. It contains a header section with metadata (reference genome, caller info, filter descriptions) and a data section with tab-delimited columns: CHROM, POS, ID, REF, ALT, QUAL, FILTER, INFO, and FORMAT plus sample genotypes. VCF files can be compressed to .vcf.gz and indexed with .tbi for efficient querying.

How much sequencing depth is needed for accurate variant calling?

For germline variant calling, 30x coverage is considered the minimum standard for whole-genome sequencing. Clinical applications typically require 60-100x for reliable heterozygous variant detection. For somatic variant calling in cancer, 200x or higher tumor coverage is recommended due to tumor heterogeneity and contamination with normal cells.

What is the GATK Best Practices pipeline?

The GATK Best Practices pipeline is a step-by-step workflow for variant calling developed by the Broad Institute. Key steps include: BWA-MEM alignment to reference, MarkDuplicates to remove PCR duplicates, Base Quality Score Recalibration (BQSR), HaplotypeCaller for variant calling, Variant Quality Score Recalibration (VQSR) for filtering, and joint genotyping for multi-sample analysis.

What is variant allele frequency (VAF)?

Variant allele frequency is the proportion of sequencing reads supporting a variant allele divided by total reads at that position. A VAF of 50% suggests a heterozygous germline variant, while low VAF (1-5%) indicates a somatic or subclonal variant. The minimum VAF detectable depends on sequencing depth and base quality.

How do I filter false-positive variant calls?

Filter variants by read depth (min 20x for germline, min 100x for somatic), base quality (min Q20), mapping quality (min 30), strand bias (Fisher test p > 0.05), and variant allele frequency. Remove variants in homopolymer runs and repetitive regions. Use population frequency databases (gnomAD) to filter common polymorphisms.

What is the difference between SNPs and indels in variant calling?

Single nucleotide polymorphisms (SNPs) change one base and are easier to call accurately. Insertions and deletions (indels) are harder to call because they require realignment around the variant site. GATK's HaplotypeCaller uses local de novo assembly to improve indel calling, but indels still have higher false-positive and false-negative rates than SNPs.

How do I call variants from RNA-seq data?

RNA-seq variant calling requires a splice-aware aligner (STAR or HISAT2) to handle exon-exon junctions. Use GATK's HaplotypeCaller in RNA-seq mode which accounts for splicing. Filter variants in splice sites and consider allele-specific expression bias. RNA-seq variant calling is less sensitive than DNA-seq due to expression level variation.

What is the importance of panel of normals (PoN) in somatic calling?

A Panel of Normals (PoN) is a collection of sequencing data from normal samples that captures systematic sequencing artifacts and recurrent false positives. Using a PoN with Mutect2 dramatically reduces false-positive somatic calls by modeling site-specific error rates. Build a PoN from at least 30 normal samples processed identically to your tumor samples.

What is the difference between germline and somatic variant calling?

Germline variant calling detects inherited variants present in all cells, typically using paired normal samples and requiring 30-100x depth. Somatic variant calling identifies mutations present only in tumor or diseased cells, requiring matched normal-tumor pairs and higher depth (200x+) to distinguish true somatic variants from sequencing noise and germline polymorphisms.

Why This Matters for Variant Calling

Understanding variant calling is essential for bioinformaticians, clinical genomicists, and molecular biology researchers. The choice of variant caller, filtering strategy, and quality control approach directly impacts the accuracy of genetic findings. VigyanLLM supports researchers with automated primer design for variant validation and targeted resequencing.

References

DePristo M.A. et al. (2011). A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nature Genetics, 43(5), 491-498.
McKenna A. et al. (2010). The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Research, 20(9), 1297-1303.
Poplin R. et al. (2018). A universal SNP and small-indel variant caller using deep neural networks. Nature Biotechnology, 36(10), 983-987.
Cibulskis K. et al. (2013). Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nature Biotechnology, 31(3), 213-219.
Garrison E. & Marth G. (2012). Haplotype-based variant detection from short-read sequencing. arXiv preprint, arXiv:1207.3907.

References

Quail M.A. et al. (2012). A tale of three next generation sequencing platforms: comparison of Ion Torrent, Pacific Biosciences and Illumina MiSeq sequencers. BMC Genomics, 13, 341.
Bybee S.M. et al. (2011). Targeted amplicon sequencing (TAS): a scalable next-gen approach to multilocus, multitaxa phylogenetics. Genome Biology and Evolution, 3, 1317-1323.
Samorodnitsky E. et al. (2015). Evaluation of hybridization capture versus amplicon-based methods for whole-exome sequencing. Human Mutation, 36(9), 903-914.
Hodkinson B.P. & Grice E.A. (2015). Next-generation sequencing: a review of technologies and tools for wound microbiome research. Advances in Wound Care, 4(1), 50-58.
Meredith M.M. et al. (2012). Dramatic enhancement of HIV-1 detection by whole genome amplicon sequencing. Journal of Virological Methods, 179(1), 214-221.