Objective
The goal of this project is to develop a new genetic risk score for type 1 diabetes (T1D) that works accurately for people from all populations. A genetic risk score adds up the effects of many common genetic variants to estimate a person’s risk of getting T1D. While current scores perform well in individuals of European ancestry, they are less accurate in people from other ancestries. This limits their usefulness for diagnosis, screening, and research in global populations. We aim to build a next-generation score ’GRS3’ that captures the full range of known and newly identified genetic risk factors for T1D. To do this, we will study genetic data from hundreds of thousands of individuals from diverse populations around the world. We will identify genetic markers that best track important immune-related risk genes in different populations, use advanced machine learning methods to build ancestry-specific and cross-population models, and test the score in large datasets to ensure it works in real-world settings. We will also build a simpler version of the score that can be used with small-scale screening tests such as those performed on newborn blood spots. Finally, we will develop a user-friendly online tool to calculate T1D genetic risk and provide ancestry-adjusted interpretation. By doing this, we aim to make genetic testing for T1D more accurate and equitable, improving how the disease is identified and classified around the world.
Background Rationale
Type 1 diabetes is caused by the immune system attacking the body’s own insulin-producing cells. Scientists have discovered that genes—particularly those in a region which control our immune systems —play a major role in determining who gets T1D. Genetic risk scores (GRS) can combine information from many of these genes to estimate someone’s overall risk of developing the disease. These scores are already used in some countries to help identify children at highest risk or to support diagnosis in adults with unclear diabetes type. However, almost all the existing genetic scores were developed using data from people of European ancestry. Because genes vary significantly between populations, these scores often work poorly in other groups. For example, the same genetic variant might be common in one population and rare in another or may carry different risk depending on how it interacts with other nearby genes. These differences mean that a score built for one population might overestimate or underestimate risk in another. To make genetic risk prediction accurate for everyone, we need to better understand how genetic risk for T1D varies around the world and build a model that accounts for these differences. We will do this by analyzing large datasets that include people from Africa, South Asia, East Asia, the Middle East, and admixed populations like those in the US. This will allow us to improve prediction of Type 1 diabetes and classification in the clinic for all populations.
Description of Project
Type 1 diabetes (T1D) is an autoimmune disease where the immune system attacks insulin-producing cells in the pancreas. While the exact causes remain unclear, genetics play a major role. We know that a person’s risk for T1D can be estimated using a “genetic risk score” (GRS), which adds up the effects of multiple genetic variants associated with the disease. However, current scores are mostly based on studies of people with European ancestry and do not work equally well for people from other backgrounds. Our project aims to create a new, more accurate genetic risk score ‘GRS3’ that works for people of all ancestries. The goal is to improve how T1D is detected, classified, and potentially screened for at an early stage.
Most of the genetic risk for T1D comes from a set of genes in the HLA region. These genes help control the immune system and vary widely by ancestry. Some combinations increase T1D risk a lot, while others protect against it. The exact versions of these genes, and how they interact, differs globally. This means that a score built using European data might miss important signals in other populations. To build a better score, we’ll analyze genetic data from over 950,000 individuals worldwide, including many with confirmed T1D. We’ll identify which genetic variants are best at predicting T1D in each population and how well these variants tag the key HLA risk genes. We will use advance machine learning techniques to build a model that can detect complex genetic interactions that typical models are unable to detect.
We plan to: (1) Identify the best genetic markers for the key T1D risk genes in different ancestry groups. (2) Use those markers to build ancestry-aware models that capture complex genetic interactions. (3) Combine these models into a unified score (GRS3) that works well regardless of ancestry. (4) Create a simpler version that uses fewer genetic markers for use in newborn screening tests from small spots of dried blood alone.
We will build a public web tool where researchers and clinicians can upload genetic data and get back risk scores and interpretation adjusted for ancestry. We’ll also make all the tools, model details, and documentation freely available so they can be used globally.
This work has broad implications. It can help improve diagnosis, reduce misclassification of T1D versus type 2 diabetes (especially in adults and underrepresented groups), and guide screening efforts to catch T1D before symptoms appear. It also supports more equitable use of genetics in medicine by ensuring that people from all backgrounds benefit from advances in genetic risk prediction.
Anticipated Outcome
We expect to deliver a powerful new tool ‘GRS3’ that estimates a person’s genetic risk of developing type 1 diabetes, and that works reliably across global populations. GRS3 will be more accurate than existing scores because it will account for ancestry-specific genetic patterns and interactions, especially within the immune-related HLA region. We also expect to produce a simplified version of GRS3, that uses fewer genetic markers while retaining most of the predictive accuracy. This will make it suitable for use in newborn screening where a limited amount of blood is taken from a heel prick test at birth, or in low-cost healthcare settings. To ensure that these scores can be used easily in practice, we will build a web-based platform that allows researchers and clinicians to upload standard genetic data and receive a clear, interpretable output. This will include ancestry-adjusted percentile rankings and risk estimates calibrated to population-level disease incidence. Our approach will also generate valuable resources for the research community, including a publicly available database linking genetic variants to HLA types in different populations. These resources can support future studies of T1D and other immune-related diseases. Ultimately, we anticipate that our work will lead to better identification of individuals at risk for T1D, more accurate classification of diabetes type in adults, and broader global use of genetic tools in diabetes care and prevention.
Relevance to T1D
This project directly addresses one of the biggest challenges in type 1 diabetes research and care: improving diagnosis and risk prediction across diverse populations. T1D is often difficult to distinguish from other types of diabetes, especially in adults and in people who don’t fit the classic profile seen in children of European ancestry. Genetic risk scores can help, but current scores are less reliable for individuals from many parts of the world. This can lead to missed or delayed diagnoses, inappropriate treatment, or exclusion from important research studies. By building an ancestry-aware genetic risk score, we aim to make risk prediction more accurate and equitable. This will support earlier and more accurate diagnosis, help identify people at risk before symptoms begin, and improve classification in large research databases. It will also allow for better targeting of prevention and treatment strategies, especially in countries or communities where T1D is under-recognized. We are also building tools that will enable other researchers and clinicians to use these scores easily, making the benefits of this work widely available. By improving how we measure and interpret genetic risk, this project supports progress toward personalized medicine in type 1 diabetes and ensures that advances in care and prevention reach everyone, regardless of their ancestry background.