Highlights
- TGen scientist Andrea Guarracino co-developed COSIGT, a genotyping method validated on hundreds of genomes at varying coverage levels.
- The method uses cosine similarity to match limited short-read DNA data against the pangenome graph, outperforming existing tools.
- Results published in Genome Biology; prior work applied the method to over 6,000 modern and ancient human genomes.
- Guarracino says the tool could lower genotyping costs for structurally complex, medically important genome regions.
A TGen scientist has co-developed a genotyping tool that holds its accuracy even when DNA data is sparse, a technical advance that could cut costs for researchers working with ancient genomes, low-coverage sequencing runs, or degraded samples.
Andrea Guarracino, Ph.D., a scientist at TGen, part of City of Hope, co-developed the method alongside Davide Bolognini, Ph.D., of Human Technopole. The tool, called COSIGT (COsine SImilarity-based GenoTyper), maps short-read DNA sequences onto a pangenome graph and uses cosine similarity to find the best-matching genotype. Results published in Genome Biology suggest COSIGT outperforms existing pangenome-based genotyping methods when genetic data is limited.
The pangenome is a collection of complete human genome sequences used as a richer reference than any single individual's DNA. Guarracino and Bolognini validated COSIGT on hundreds of genomes at varying coverage levels. In prior work, the team applied the method to over 6,000 modern and ancient human genomes, demonstrating its potential in analyzing structurally complex genetic sequences on a population-level scale.
The tool will help scientists to genotype structurally complex regions of the human genome from limited data, potentially at lower costs, said Guarracino, an assistant professor in TGen's Bioinnovation and Genome Sciences Division.
How does COSIGT work?
COSIGT represents pangenome data as a graph and maps short-read sequences from an individual onto that graph. The researchers quantify how many short-reads map to each node, producing a vector that is compared against vectors representing each genome in the pangenome. The pair of pangenome vectors whose combination most closely matches the individual's short-read vector is selected as the predicted genotype. Because the method compares vector direction rather than raw read depth, it remains accurate when only a small amount of genetic data is available.
"With the pangenome, you have many views of how humans can be," Guarracino said. "We want to leverage this so information from many individuals can be used to genotype other individuals."
The study was posted to the TGen news site Sept. 9, 2026. No commercialization timeline or licensing terms were disclosed in the release.
Sources
Every factual claim in this article traces to one of the sources below. See how we work for the editorial process.
- tgen.org retrieved 10/09/2026 11:54
Authored by The Scottsdale Signal. Drafted by AI from primary-source material under our beat-specific editorial guides; reviewed by humans before publish under our five-gate process. Sources retrieved at 10/09/2026 11:54. Every claim traces to a source.
