Network-based Variant Calling — Graph-genome Genotyping
Also known as: graph-genome variant calling, variation graph genotyping, vg-based variant calling, pangenome variant calling
Network-based (graph-genome) variant calling replaces the conventional single linear reference genome with a variation graph — a network in which nodes represent sequence segments and edges represent known alternative paths through the genome. Reads are mapped onto this graph, enabling detection of SNPs, indels, and structural variants with substantially lower reference bias than linear-reference pipelines. Key tools include the Variation Graph Toolkit (vg) and Graphtyper.
Key highlights
- Substantially reduces reference bias, improving genotyping accuracy for variants common in populations underrepresented in the canonical reference.
- Handles repetitive genomic regions and structural variants more accurately than linear-reference pipelines.
- Jointly genotypes known variants across many samples, boosting statistical power in population studies.
- Produces standard VCF output compatible with all downstream association and annotation tools.
- Scales to population-level cohorts when implemented with tools such as Graphtyper or vg giraffe.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use network-based variant calling when the study involves populations or ethnic groups that are poorly represented in the standard linear reference genome, when high accuracy at repeat-rich or structurally complex loci is critical, or when calling structural variants alongside SNPs and indels in a unified framework. It is particularly appropriate for large-scale population genomics where known variants should be re-genotyped rather than discovered de novo. Do not use it when computational resources are very limited (graph workflows are memory-intensive), when the study focuses exclusively on highly conserved coding regions where linear callers already perform near-optimally, or when no suitable population variant panel exists to build a meaningful graph.
Strengths & limitations
- Substantially reduces reference bias, improving genotyping accuracy for variants common in populations underrepresented in the canonical reference.
- Handles repetitive genomic regions and structural variants more accurately than linear-reference pipelines.
- Jointly genotypes known variants across many samples, boosting statistical power in population studies.
- Produces standard VCF output compatible with all downstream association and annotation tools.
- Scales to population-level cohorts when implemented with tools such as Graphtyper or vg giraffe.
- Higher memory and compute requirements than linear-reference pipelines; large population graphs can require hundreds of gigabytes of RAM.
- Quality is bounded by the completeness of the input variant panel; rare or novel variants absent from the graph may still be missed or mis-genotyped.
- Tooling is less mature and documentation is thinner than for established linear callers (GATK, DeepVariant), presenting a steeper learning curve.
- Graph construction and read mapping are slower per sample than equivalent linear workflows, increasing wall-clock time for small cohorts.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does network-based variant calling differ from standard GATK HaplotypeCaller?
GATK HaplotypeCaller maps reads to a single linear reference and assembles local haplotypes de novo at candidate sites. Network-based callers use a pre-built graph that already encodes known variants as explicit edges, so reads carrying those variants align without penalisation. The graph approach reduces reference bias and improves recall in polymorphic regions, but GATK remains competitive for well-characterised coding regions and is simpler to run in clinical pipelines.
Do I need a population-specific variant panel to build the graph?
Yes — the graph quality depends directly on the variant panel used to construct it. A diverse resource such as gnomAD or the 1000 Genomes Project is recommended. If the study population differs substantially from the panel, graph benefits diminish; ideally, a pilot run that adds study-specific variants back into the graph iteratively can further improve performance.
Can graph-based variant calling handle long reads (PacBio, Oxford Nanopore)?
Yes. Tools such as GraphAligner and the vg toolkit support long-read alignment to variation graphs. Long reads can resolve complex structural variants that short reads cannot phase, making the combination of graph genomes and long-read sequencing particularly powerful for SV discovery.
Is the VCF output from network-based callers compatible with standard GWAS pipelines?
Yes. Graphtyper, vg call, and similar tools produce standard VCF files with per-sample genotype fields. These can be imported directly into PLINK, REGENIE, or SAIGE for GWAS. The main caution is to verify that variant positions are lifted to the correct linear reference coordinates when mixing graph-called and linear-called datasets.
What sample sizes are appropriate for building a useful pangenome graph?
For re-genotyping known variants, even a small study can benefit from a publicly available population graph built on thousands of samples. For constructing a new population-specific graph de novo, a few dozen diverse samples can capture the majority of common variants; hundreds or thousands of samples are needed to reliably capture variants at minor allele frequencies below 1%.
Sources
- 1.Garrison, E., Sirén, J., Novak, A. M., Hickey, G., Eizenga, J. M., Dawson, E. T., Jones, W., Garg, S., Markello, C., Lin, M. F., Paten, B., & Durbin, R. (2018). Variation graph toolkit improves read mapping by representing genetic variation in the reference. Nature Biotechnology, 36(9), 875–879.
- 2.Eggertsson, H. P., Jonsson, H., Kristmundsdottir, S., Hjartarson, E., Kehr, B., Masson, G., Zink, F., Hjorleifsson, K. E., Jonasdottir, A., Jonasdottir, A., Jonsdottir, I., Gudbjartsson, D. F., Melsted, P., Stefansson, K., & Halldorsson, B. V. (2017). Graphtyper enables population-scale genotyping using pangenome graphs. Nature Genetics, 49(11), 1654–1660.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Network-based variant calling. ScholarGate. https://scholargate.app/bioinformatics/network-based-variant-calling