Variant calling is a fundamental step in genomic data analysis. It involves identifying differences, or variants, in DNA sequences compared to a reference genome. These variants can range from single nucleotide polymorphisms (SNPs) to larger insertions, deletions, or structural rearrangements. Understanding these variations is crucial for diagnosing genetic diseases, understanding evolutionary relationships, and developing personalized medicine.

The process of variant calling typically involves several key stages, starting from raw sequencing reads and culminating in a list of identified genetic variations.

Before variants can be identified, the raw sequencing reads must be accurately mapped to a reference genome. This process, known as alignment, places each short read into its correct genomic context. Tools like BWA (Burrows-Wheeler Aligner) or Bowtie are commonly used for this purpose. The output is typically a Sequence Alignment Map (SAM) or Binary Alignment Map (BAM) file, which contains the aligned reads and associated quality scores.

Once reads are aligned, specialized software identifies potential variants. These tools analyze the aligned reads at each position, looking for discrepancies. Common variant calling tools include GATK (Genome Analysis Toolkit), FreeBayes, and Samtools/BCFtools. They employ statistical models to assess the likelihood of a variant being present, considering factors like read depth, base quality, and strand bias.

Raw variant calls often contain false positives due to sequencing errors or alignment artifacts. Variant filtering is crucial to remove these unreliable calls. Filtering strategies often involve using quality metrics provided by the variant caller, such as variant quality by depth (QD), mapping quality (MQ), and read position bias. Machine learning approaches are also increasingly used for more sophisticated filtering.

Once variants are called and filtered, annotation provides biological context. This involves mapping the identified variants to known genomic features, such as genes, regulatory elements, and known disease-associated mutations. Tools like VEP (Variant Effect Predictor) or ANNOVAR use databases (e.g., dbSNP, ClinVar, gnomAD) to determine the potential impact of a variant, such as whether it falls within a coding region, causes an amino acid change, or is associated with a particular phenotype.

Variant calling pipelines typically produce files in standard formats. The most common is the Variant Call Format (VCF), which stores information about each variant, including its position, reference and alternate alleles, quality scores, and genotype information for samples. The related file format is the Binary Variant Call Format (BCF).