TGen Researchers Co-Author DeepSomatic Study Improving Cancer Mutation Detection

Highlights

• DeepSomatic, a machine learning method, works with both short-read and long-read genetic sequencing to detect cancer-related mutations.
• The method increased high-confidence benchmark somatic variants by seven times over the previous single tumor-normal cell line set.
• TGen, part of City of Hope, contributed pediatric leukemia and glioblastoma patient samples to test the model.
• The team released an open dataset called CASTLE, covering six matched tumor-normal cell line pairs sequenced across multiple platforms.

A machine learning method called DeepSomatic can significantly improve detection of genetic mutations associated with cancer, according to a study published in Nature Biotechnology and co-authored by researchers at the Translational Genomics Research Institute (TGen), part of City of Hope.

The study was led by UC Santa Cruz Genomics Institute and Google Research. TGen's contribution included patient samples and co-authorship from Floris Barthel, M.D., Ph.D., an assistant professor in TGen's Bioinnovation and Genome Sciences Division.

Somatic variants — genetic mutations acquired during a person's life rather than inherited — are frequently associated with tumors. Detecting them accurately is central to understanding how cancers develop and to guiding treatment decisions based on an individual's tumor profile. Until now, identifying somatic variants was largely limited to older short-read sequencing data, which struggles in complex or repetitive regions of the genome.

DeepSomatic is designed to work with both short-read sequencing, which produces a large number of short DNA pieces (50 to 300 base pairs), and newer long-read sequencing, which generates fewer but much longer pieces (1 to 100 kilobytes of base pairs). That dual compatibility is the method's core technical advance.

Sources

Every factual claim in this article traces to one of the sources below. See how we work for the editorial process.

1. tgen.org retrieved 2026-05-02T08:57:39.899346+00:00

Authored by Claude, drafted from primary-source material with beat-specific editorial guides at The Scottsdale Signal. Sources retrieved at 2026-05-02T08:57:39.899346+00:00. Every claim traces to a source. Reviewed before publish under our five-gate editorial process.