- Variant discovery and genotyping are fundamental processes in bioinformatics that identify genetic differences among individuals, populations, and species and determine which genetic alleles are present in each sample. Genetic variants include single-nucleotide variants, insertions, deletions, structural variations, copy-number variations, and other changes in DNA sequence or chromosome structure. These variations contribute to genetic diversity, evolution, inherited diseases, complex traits, adaptation, and differences in biological characteristics. Variant discovery identifies candidate differences between sequenced samples and a reference genome or among samples, while genotyping determines the alleles carried by each individual at specific genomic positions. Together, these processes provide essential information for population genetics, medical genomics, evolutionary biology, agriculture, conservation genetics, and molecular breeding.
- Genetic variation can occur at several levels. Single-nucleotide variants (SNVs) involve a change in one nucleotide at a specific genomic position. When an SNV is observed as a polymorphism in a population, it is commonly described as a single-nucleotide polymorphism (SNP), although the terms are not strictly interchangeable in every context. Insertions and deletions, collectively called indels, involve the addition or removal of one or more nucleotides. Larger structural variants include deletions, duplications, inversions, insertions, and translocations. Copy-number variations involve differences in the number of copies of particular genomic regions. Variants can occur in coding regions, introns, untranslated regions, promoters, enhancers, and intergenic regions, and their biological effects depend on their location and molecular consequences.
- Variant discovery usually begins with DNA sequencing data. Whole-genome sequencing provides information across most of the genome, while whole-exome sequencing focuses primarily on protein-coding regions. Targeted sequencing examines selected genes or genomic regions, and genotyping arrays measure predefined variants at selected positions. The choice of technology affects the types of variants that can be detected. Short-read sequencing is widely used for SNV and small-indel discovery, but repetitive regions, complex genomic regions, and larger structural variants may be difficult to resolve. Long-read sequencing can improve the detection of structural variants, repeat expansions, complex rearrangements, and variants in regions that are challenging for short reads. Different technologies have different error profiles, coverage characteristics, and analytical requirements.
- The first stage of a sequencing-based variant discovery workflow is quality assessment of the raw reads. Sequencing reads may contain low-quality bases, adapter sequences, contamination, and technical artifacts. Quality-control tools such as FastQC and MultiQC help examine base quality, sequence composition, adapter contamination, duplication levels, and other read characteristics. Tools such as fastp and Cutadapt can remove adapters and filter or trim low-quality reads. Read preprocessing should be performed carefully because excessive trimming can remove useful information or affect alignment and variant detection. The original data, quality reports, and preprocessing parameters should be retained for reproducibility.
- After quality control, reads are aligned to a reference genome. Read alignment places sequencing reads at corresponding genomic positions and provides the information needed to identify differences between the sample and the reference. Common short-read aligners include BWA-MEM2 and Bowtie2, while minimap2 is widely used for long-read alignment and several other sequence-mapping tasks. Spliced aligners such as STAR and HISAT2 are designed for RNA sequencing and are generally used in transcriptomic workflows rather than ordinary DNA variant discovery. The choice of aligner depends on the sequencing technology, read length, organism, and analysis objective.
- Aligned reads are usually stored in SAM or BAM format, or in the compressed CRAM format. SAM is a text-based alignment format, BAM is its binary representation, and CRAM is a reference-oriented compressed format. Alignment files are commonly sorted and indexed using SAMtools. Sorting organizes alignments by genomic coordinate, while indexing allows software to retrieve reads from selected genomic regions efficiently. Alignment quality should be examined before variant calling because poor mapping, contamination, incorrect sample identity, and reference bias can generate false variant candidates.
- A simplified short-read alignment workflow using BWA-MEM2 and SAMtools is shown below:
# Index the reference genome
bwa-mem2 index reference.fasta
# Align paired-end reads
bwa-mem2 mem reference.fasta \
sample_R1.fastq.gz \
sample_R2.fastq.gz \
> sample.sam
# Convert SAM to BAM
samtools view -bS sample.sam > sample.bam
# Sort the BAM file
samtools sort sample.bam -o sample.sorted.bam
# Index the sorted BAM file
samtools index sample.sorted.bam
# Generate alignment statistics
samtools flagstat sample.sorted.bam
- This is an illustrative workflow. Production analyses should also record software versions, reference assembly information, read groups, library metadata, and relevant alignment parameters. In addition, workflows may use CRAM instead of BAM to reduce storage requirements.
- Variant calling is the process of identifying candidate genetic differences from aligned sequencing reads. Variant callers examine the bases observed at genomic positions, the number of supporting reads, base quality, mapping quality, allele balance, local sequence context, and other evidence. They estimate whether an observed difference is likely to represent a true biological variant or a sequencing or alignment artifact. Variant calling can be performed for individual samples or jointly across multiple samples, depending on the study design and software. Common tools include GATK HaplotypeCaller, FreeBayes, bcftools, DeepVariant, and other specialized callers. Their algorithms and assumptions differ, so the appropriate tool depends on the organism, sequencing platform, ploidy, and variant types being investigated.
- GATK HaplotypeCaller uses local haplotype assembly to evaluate candidate variants in genomic regions. It is widely used for germline variant discovery in diploid and other supported settings. FreeBayes uses a Bayesian framework for detecting genetic variants from aligned reads and can support pooled and multisample analyses. Bcftools provides a suite of utilities for variant calling, filtering, and manipulation of sequence variation data. DeepVariant uses deep learning to interpret sequencing evidence and predict variants, with models developed for particular data types and sequencing technologies. No caller is universally optimal, and performance should be assessed for the intended organism, sequencing platform, coverage, and variant classes.
- A simplified example of variant calling with bcftools is:
# Generate a pileup from aligned reads and call variants
bcftools mpileup \
-f reference.fasta \
sample.sorted.bam |
bcftools call \
-mv \
-Ov \
-o sample.vcf
- The -m option selects the multiallelic caller, while -v requests variant sites. The example illustrates a basic calling operation and does not include all recommended quality-control, normalization, or filtering steps. A production workflow should use a validated reference, appropriate sample metadata, suitable ploidy settings, and documented filtering criteria.
- Variant calling results are commonly stored in Variant Call Format (VCF) or its binary representation, BCF. A VCF file contains metadata lines describing the reference, contigs, filters, and information fields, followed by a tab-separated data section. The main fields include chromosome, position, variant identifier, reference allele, alternate allele, quality score, filter status, information, and genotype-related format fields. Sample genotypes may be represented using allele indices such as 0/0, 0/1, and 1/1 for a typical diploid biallelic site. The exact interpretation depends on ploidy and the genotype encoding.
- An illustrative VCF record is shown below:
##fileformat=VCFv4.3
##contig=<ID=chr1>
#CHROM POS ID REF ALT QUAL FILTER INFO
chr1 10583 . G A 99 PASS DP=40
- This record represents a candidate substitution of G by A at position 10583 on chromosome 1. The QUAL field contains a variant quality score, FILTER indicates whether the record passed the specified filters, and INFO contains additional annotations such as read depth. The example is simplified; real VCF files include detailed header definitions for INFO, FORMAT, FILTER, and other fields. A sample-level VCF may also contain genotype information in FORMAT and sample columns.
- Variant discovery and genotyping are related but distinct. Variant discovery identifies candidate variable positions, often by comparing reads with a reference genome. Genotyping determines the allele combination carried by each sample at a specified position. For a diploid organism, a biallelic site may have three possible genotypes: homozygous reference, heterozygous, and homozygous alternate. These are commonly represented as 0/0, 0/1, and 1/1, respectively. A haploid organism typically has one allele at a locus, while organisms with different ploidy levels may require additional genotype states. Polyploid species, such as many plants, present more complex genotyping challenges because several homologous chromosome copies may carry different alleles.
- Genotyping can be performed using sequencing-based methods or genotyping arrays. Sequencing-based genotyping derives allele information from aligned reads and variant-calling models. Genotyping arrays use probes designed to detect known variants at predefined genomic positions. Arrays are generally efficient for large studies involving many samples, but they cannot directly discover all previously unknown variants. Their performance may also depend on how well the selected markers represent genetic diversity in the population being studied. Sequencing offers broader discovery potential but usually requires more extensive computational analysis and data storage.
- Variant quality assessment is essential because raw variant calls may include false positives and false negatives. False positives are reported variants that are not genuine biological differences, while false negatives are true variants that the analysis fails to detect. Common sources of false calls include low sequencing depth, sequencing errors, poor mapping quality, repetitive regions, paralogous sequences, contamination, incorrect sample identity, and reference assembly problems. Variants near insertions, deletions, or complex sequence changes may be especially difficult to call accurately. Quality assessment should therefore consider both the sequencing evidence and the genomic context.
- Important variant-level metrics include read depth, genotype quality, variant quality, mapping quality, allele balance, strand bias, and the number of reads supporting each allele. Read depth describes how many reads cover a genomic position. Genotype quality estimates the confidence in a sample’s genotype call, while mapping quality reflects confidence in the placement of reads. Allele balance describes the proportion of reads supporting each allele and can help identify unusual genotype patterns. For a typical diploid heterozygous variant, approximately equal support for the two alleles may be expected under suitable sequencing conditions, although biological and technical factors can produce deviations.
- Variant filtering removes or flags calls that do not meet predefined quality criteria. Filtering may use depth, genotype quality, mapping quality, allele balance, strand bias, variant quality, and other measures. Thresholds should be selected according to the sequencing technology, organism, ploidy, sample design, and intended analysis. A single universal filtering threshold is not appropriate for all studies. Hard filtering applies predefined thresholds to individual metrics, while statistical or machine-learning approaches may combine multiple quality features. Variant filtering should be documented, and the effects of filtering on variant numbers and sample genotypes should be reported.
- Variant normalization is another important step. The same biological variant can sometimes be represented in different ways, especially for insertions and deletions in repetitive genomic regions. Normalization typically involves aligning alleles to a consistent representation and shifting indels to a standard position according to established conventions. Tools such as bcftools norm and vt can normalize and validate variant records. Consistent normalization is important for comparing variants across samples, databases, and annotation resources. Multiallelic sites may also require decomposition or other processing depending on the downstream analysis.
- An example of variant normalization using bcftools is:
# Normalize variants against the reference genome
bcftools norm \
-f reference.fasta \
-m -any \
input.vcf \
-Oz \
-o normalized.vcf.gz
# Index the normalized VCF file
bcftools index normalized.vcf.gz
- The command uses the reference genome to normalize variant representations and separates multiallelic records into biallelic records. The choice to split multiallelic sites should be made according to downstream analytical requirements, because splitting can affect genotype representation and interpretation.
- After calling and filtering, variants can be annotated to predict their potential biological consequences. Variant annotation identifies the genes, transcripts, coding regions, splice sites, regulatory elements, and other features affected by each variant. Tools such as Ensembl Variant Effect Predictor (VEP), SnpEff, and ANNOVAR can assign consequences including synonymous, missense, nonsense, frameshift, splice-site, untranslated-region, and intergenic effects. These annotations help prioritize variants for further investigation, but predicted consequences do not by themselves establish biological or clinical significance.
- Reference genome selection is particularly important in variant discovery and genotyping. Different genome assemblies may use different chromosome sequences, coordinates, and gene annotations. A variant identified using one assembly may have a different coordinate in another assembly. Researchers should record the assembly accession and version and ensure that the reference genome used for alignment, variant calling, normalization, and annotation is consistent. If variants are converted between assemblies, tools such as CrossMap and Picard LiftoverVcf may be used where appropriate. Liftover can fail or produce ambiguous mappings, especially in rearranged or poorly aligned regions, so converted records should be checked.
- Sample identity and contamination checks are also essential. Incorrect sample labels, sample swaps, cross-contamination, and mixtures of biological samples can lead to misleading genotypes. Genotype concordance, sex chromosome checks, contamination estimation, and relatedness analysis can help identify such problems. Tools such as VerifyBamID2, Picard, and PLINK provide different methods for examining sample quality, identity, and genetic relationships. The appropriate checks depend on the organism and the design of the study.
- Population-level variant discovery involves analyzing genetic variation across multiple individuals. Joint genotyping allows variants to be evaluated across a cohort, improving consistency of genotype assignment and enabling the construction of a combined genotype dataset. Population studies often examine allele frequencies, heterozygosity, linkage disequilibrium, population structure, genetic differentiation, and relatedness. Tools such as PLINK, bcftools, and VCFtools support many of these analyses. Population-level results require careful consideration of sample size, population composition, missing genotypes, relatedness, and ascertainment bias.
- Genotyping arrays and sequencing datasets can also be used together. Sequencing may identify a broad set of variants, while array data provide genotypes at selected known positions. Imputation methods estimate unobserved genotypes using linkage disequilibrium patterns and reference haplotype panels. Tools such as Beagle and Minimac support genotype imputation in appropriate study designs. Imputation accuracy depends on reference panel relevance, allele frequency, sample ancestry, marker density, and the quality of the input genotypes. Imputed genotypes should be accompanied by appropriate quality metrics and should not be treated as equivalent to directly observed genotypes.
- Structural variant discovery requires additional methods because large genomic changes may not be fully represented by ordinary SNV and small-indel callers. Structural variants can be detected through read-pair orientation, split reads, read depth, local assembly, and long-read alignment patterns. Tools such as Manta, DELLY, GRIDSS, Sniffles, and cuteSV support different structural variant detection strategies and sequencing technologies. Copy-number variation can be investigated using read-depth patterns, genotyping arrays, or specialized algorithms. Combining multiple detection signals can improve confidence, although the methods may share limitations and should be evaluated for the relevant variant classes.
- Variant discovery is widely used in medical and clinical genomics to identify genetic changes associated with inherited disorders, cancer, and complex diseases. In cancer genomics, somatic variant calling compares tumor samples with matched normal samples when available to distinguish acquired mutations from inherited variation. Tools such as Mutect2 and other somatic variant callers are designed for this purpose. In population genetics, variant discovery supports the study of ancestry, demographic history, adaptation, and natural selection. In agriculture and animal breeding, genotyping enables genomic selection, marker-assisted selection, parentage verification, genetic diversity assessment, and the identification of variants associated with economically important traits. In conservation genetics, genetic variation data can help evaluate population diversity, inbreeding, and population connectivity.
- Reproducibility is essential throughout variant discovery and genotyping. Researchers should record the sequencing platform, library preparation method, read preprocessing steps, reference genome assembly, aligner and variant caller versions, parameters, filtering criteria, normalization procedures, and annotation database releases. Workflow systems such as Nextflow and Snakemake can organize analysis steps and make computational procedures repeatable. Variant files should retain appropriate metadata, and sample identifiers should be managed consistently. Sensitive human genetic data also require appropriate access controls, secure storage, and compliance with relevant ethical and legal requirements.
- Variant discovery and genotyping provide the foundation for understanding genetic differences at individual and population levels. Reliable results depend on high-quality sequencing data, suitable reference genomes, accurate read alignment, validated variant callers, careful quality filtering, consistent normalization, and appropriate genotype interpretation. By combining robust computational methods with biological evidence and transparent documentation, researchers can generate reliable variant datasets for downstream genomic, evolutionary, clinical, and agricultural analyses.