Diabetes mellitus affects nearly 600 million people worldwide and is a leading cause of blindness, kidney failure, limb amputation, and premature death. Although it is diagnosed by a single biochemical criterion — elevated blood glucose — it encompasses a spectrum of biologically distinct diseases with different causes and very different optimal treatments. Accurate classification of diabetes subtype is therefore clinically important, yet it remains elusive in a substantial proportion of patients. This challenge is most acute in young, non-obese individuals who do not clearly fit the criteria for autoimmune type 1 diabetes (T1D) or obesity-associated type 2 diabetes (T2D). Many of these patients harbour genetic determinants of their disease that, if identified, could guide treatment decisions decisively — for example, permitting a switch from daily insulin injections to oral tablets, or enabling earlier initiation of therapies that can halt disease progression.Genetic testing is increasingly used in diabetes care, but current approaches remain fragmented and incomplete. Standard tests interrogate only a small subset of known disease-causing genes, and focus almost exclusively on the protein-coding portion of the genome — roughly 2% of the total. The remaining 98%, which contains regulatory sequences that control gene activity in pancreatic beta cells, is routinely ignored. Moreover, existing tools treat rare single-gene defects and common polygenic risk as entirely separate problems, when in reality the same patient may carry relevant variants across both categories.GenomeDia was conceived to address these gaps. The project aimed to develop an integrated computational platform that analyses whole genome sequences alongside clinical data to improve the classification of diabetes subtypes in young, non-obese patients. By drawing on extensive functional genomic maps of human pancreatic islet cells generated by our group, as well as genetic findings from the ERC-funded DecodeDiabetes project, GenomeDia sought to expand the interpretable genomic space to the entire genome and to combine rare variant analysis with polygenic risk scores within a single unified workflow.
