Researchers at UC San Diego unveiled T1GRS, a new machine learning model that predicts genetic risk for Type 1 diabetes (T1D) more accurately across a broader population. The peer-reviewed study was published April 30, 2026 in Nature Genetics.
What T1GRS does
- Goes beyond high-risk variants: Unlike existing scores that work best for people with known high-risk genes, T1GRS analyzes complex interactions between 199 risk variants across the genome, including the MHC region on chromosome 6.
- Earlier, broader detection: It identifies both children and adults at high risk sooner, including people who develop T1D without the well-known high-risk genetic regions.
- Built on massive data: Trained on genomes from 20,000+ people with T1D and ∼800,000 without, all of European ancestry. Confirmed 79 known loci and found 13 new loci tied to immune function, gene regulation, and blood sugar control.
4 T1D subtypes identified
T1GRS groups people by the genetic features driving their score, each with distinct clinical patterns:
- MHC-driven – Known high-risk MHC variants; earliest childhood onset.
- MHC-enriched – Mix of MHC and non-MHC variants; slightly later onset, intermediate severity.
- T-cell-enriched – Non-MHC variants affecting adaptive immune response; intermediate onset age.
- Pancreas-enriched – Non-MHC variants impacting pancreatic beta cells; later onset but highest rates of complications like kidney disease, nerve damage, heart problems.
How well it works
- Validated externally: Tested on NIH All of Us and nPOD biobank data. Despite smaller samples, it still predicted risk with 87% accuracy.
- Works beyond Europeans: Even though developed on European-ancestry data, it performed well in non-European populations too.
Clinical impact
- Better screening: Captures at-risk people missed by current tools, enabling closer monitoring to reduce complications like diabetic ketoacidosis at diagnosis.
- Personalized prevention: Helps identify candidates for preventive therapies such as teplizumab before T1D fully develops.