UC San Diego researchers say a new machine-learning tool is giving them a sharper, more nuanced view of who is at risk for type 1 diabetes. The genetic score, called T1GRS, picks up a broader and subtler range of risks than earlier tests, confirming dozens of known risk loci while surfacing new variants and sorting people with the disease into four genetic subtypes tied to age of onset and complications. The hope is that this tool will tighten up screening, cut down on misdiagnoses and flag more people who might qualify for prevention trials.
Study Lands In Nature Genetics
The peer-reviewed work behind T1GRS was published last Thursday in Nature Genetics, detailing how the team built and tested the model. Researchers trained it on genetic data from more than 20,000 people with type 1 diabetes and roughly 800,000 without the disease, giving them the statistical muscle to probe complex, non-linear interactions across the genome.
Untangling The MHC Reveals New Signals
One of the trickiest regions to analyze, the major histocompatibility complex (MHC) on chromosome 6, got special attention. Using data from more than 29,000 people, the group mapped that region and picked out novel variants that appear to influence immune function and gene activation, as reported by UC San Diego Today. Co-first author Emily Griffin said clusters of MHC "blocks" raise diabetes risk, while their absence makes the disease much less likely. That mapping helped the team validate 79 risk loci and pinpoint 13 that had not previously been linked to type 1 diabetes.
How T1GRS Works
T1GRS itself is a gradient-boosting machine-learning classifier that combines 199 risk variants, including 70 established HLA-associated alleles, and models nonlinear interactions among them to outperform simple additive scores, according to Nature Genetics. By capturing how MHC and non-MHC loci interact, the tool can spot people who go on to develop type 1 diabetes even if they lack the classical high-risk variants, expanding the group of patients who might merit closer monitoring and possible prevention.
Clinical Implications And Early Tests
The team ran T1GRS on external datasets, including the NIH All of Us program and the nPOD biobank. Performance dipped in smaller samples, but in one validation, the tool still predicted risk with roughly 87 percent accuracy, researchers told The San Diego Union‑Tribune. Crucially, the model grouped patients into four subtypes: MHC-driven, MHC-enriched, T-cell-enriched and pancreas-enriched. That pancreas-enriched group showed higher rates of complications such as kidney disease and nerve damage. The authors say those genetic subtypes could steer who gets tighter follow-up or enrollment in preventive trials, including studies of teplizumab.
What This Could Mean For San Diego Patients
UC San Diego already runs extensive clinical and research programs focused on type 1 diabetes, and investigators hope T1GRS can eventually slide into existing screening pipelines to catch at-risk children and adults earlier, according to UC San Diego. The researchers stress that genes are only one piece of the puzzle. The next step, they say, is to combine genetic scores like T1GRS with biomarker data so they can sharpen prediction even further and better match people to the prevention strategies most likely to help them.
