Machine-Learning Genetic Risk Score Improves Early Prediction of Type 1 Diabetes

Predicting who will develop the disease remains difficult, as many genetic risk scores focus on a limited set of high-risk variants. Expanding risk stratification to capture broader populations could enable earlier monitoring and intervention in both children and adults. A new study shows that a machine learning–derived genetic risk score improves prediction across more diverse groups.

T1GRS is a machine learning model designed to estimate an individual’s genetic risk for type 1 diabetes (T1D). It generates a personalized score by integrating signals from disease-associated loci across the genome, including the major histocompatibility complex (MHC). The approach aims to improve accuracy beyond existing tools, which perform best primarily in individuals with well-characterized high-risk variants.

Investigators at the University of California San Diego (La Jolla, CA, USA) developed T1GRS to capture non-linear interactions among 199 risk variants spanning the genome. In analyses of more than 20,000 individuals of European ancestry with T1D and nearly 800,000 without the disease, the team confirmed known risk variants at 79 loci and identified 13 additional loci linked to gene regulation, immune function, and glycemic control. Separate mapping in more than 29,000 individuals revealed novel MHC variants associated with T1D that influence immune function and gene activation.

The model demonstrated high accuracy across a broader population, including individuals with more complex genetic risk, and identified children and adults at elevated risk earlier than current methods. To evaluate performance beyond the training dataset, the team applied T1GRS to external cohorts from the National Institutes of Health (NIH) All of Us Research Program and the National Pancreatic Organ Donor (nPOD) biobank. Although accuracy declined in these smaller datasets, the model still predicted risk with 87% accuracy.

Analysis of the genetic features most strongly influencing each individual’s score enabled the classification of people with T1D into four subtypes with distinct clinical profiles. The MHC-driven group is defined by well-known high-risk variants and is typically associated with early onset in childhood. The MHC-enriched group reflects a combination of variants within and outside the MHC region, with slightly later onset and intermediate disease severity.

The T-cell-enriched group is largely influenced by non-MHC variants affecting adaptive immune responses and shows a similar intermediate onset pattern. The pancreas-enriched group is driven primarily by non-MHC variants affecting pancreatic cells, including insulin-producing beta cells, and although associated with later onset, it carries the highest risk of complications such as kidney disease, nerve damage, and cardiovascular conditions.

The findings, published April 30, 2026, in Nature Genetics, showed that these four subtypes were consistently reproduced across independent cohorts. The authors suggest that the results support broader clinical screening to enable earlier detection and monitoring, with potential to guide more individualized care pathways.

“Genetic risk scoring allows us to capture a broader pool of both children and adults who are at high risk for T1D but who might otherwise be missed. This supports close monitoring to reduce the risk of complications such as diabetic ketoacidosis at diagnosis and helps identify individuals eligible for preventative therapies like teplizumab,” said Carolyn McGrail, Ph.D., senior associate consultant at L.E.K. Consulting.

“We were able to do a really good job of predicting risk in non‑European populations as well, even though T1GRS was developed in individuals of European descent,” said co-first author Emily Griffin, PhD, postdoctoral fellow at UC San Diego School of Medicine. 

Related Links
UC San Diego School of Medicine