We aim to identify factors contributing to the destruction ofβ–cells in diabetes. In this project, we took advantage of publicly available bulk and single cell gene expression datasets from human islets of healthy and diabetic donors(T1D and T2D). We used deep transfer learning platform DEGAS1 combinedwith this data to identify subsets of β–cells within human islet single cell RNA seq data whose transcriptomic signature correlates with that of diabetic islets. We determined the transcriptomic differences between high–and low–scoring β–cells within the human islet scRNAseq data. As part of this project, we identified the geneDLK1asa candidate marker of β–cell resilience under diabetic stress. Our project will take two major approaches. 1)We will use antibodies to stain for the candidate protein DLK1 that we predict is expressed in β–cells that persist in the face of autoimmunity. 2) We will use spatial transcriptomics in non–diabetic, AAB+, and T1Dhuman pancreas to measure the expression of candidate genes identified by DEGAS analysis and the spatial location of cells expressing these genes. By analyzing the gene expression patterns of these cells in we anticipate defining genes associated with different β–cell states, spanning from resilient to vulnerable. We anticipate that this knowledge will help uncover new mechanisms to improve β–cell survival, escape autoimmunity, and explain why some β–cells persist but others do not.
