Variant calling is where the findings start
Variant calling is the step where a sequenced sample is compared against a reference genome and every position where it differs is identified, scored, and filtered. Everything upstream exists to get here cleanly, and everything downstream depends on getting it right.
Types of variants
Single nucleotide variants
The simplest — one base swapped for another.
Small insertions and deletions
A few bases added or removed.
Structural and copy number variants
Larger rearrangements, duplications, or deletions, requiring different detection methods than the read level approaches used for SNVs and indels.
Most standard pipelines handle SNVs and indels well. Structural variant calling is harder, less standardized, and usually the place where a team's specific needs shape the tooling most.
Germline versus somatic
The two main workflows are built for different questions.
Germline calling
Identifies the variants a person was born with, present in every cell. The reference is the standard genome, and the expectation is that each variant appears in roughly half or all of the reads at a position.
Joint calling across many samples improves accuracy because rare variants gain evidence from the cohort.
Somatic calling
Identifies variants acquired by a tumor, present only in a subset of cells. The comparison is usually a tumor sample against a matched normal from the same patient, and the variant may appear in a small fraction of reads.
Sensitivity at low variant allele frequency, with false positives from sequencing artifacts a constant risk.
Which workflow applies determines the pipeline design, the filtering strategy, and the interpretation — so it has to be settled before anything is built.
Filtering and annotation
Raw variant calls are noisy. Filtering removes calls that are more likely artifact than biology, either through hard thresholds on quality metrics or through GATK's variant quality score recalibration, which learns from known variant sets.
Annotation then attaches meaning: which gene a variant falls in, whether it changes the protein, how common it is in population databases, whether it has been reported as pathogenic.
The choice of annotation sources matters as much as the tool.
Where biomarker analysis fits
In research and clinical settings, the goal is often not the full variant set but a specific set of biomarkers — variants or signatures with established relevance to a disease, a prognosis, or a treatment response.
Biomarker analysis narrows the pipeline output to those signals and applies the statistical rigor needed to report on them.
Statistical Analysis for Biological Data →Our NGS pipeline work using BWA, GATK, and SAMtools covers variant calling for both genomic research and biomarker analysis, with the germline or somatic workflow chosen and engineered to fit the question the team is actually asking.
A modular mutation calling pipeline, alignment through reporting, validated stage by stage in TypeScript, with 98 percent automated test coverage on the web application.
Read the case study →