PanKbase Integrated Single-Cell Map: A Comprehensive Atlas of Human Pancreatic Islets

Abstract Aims/hypothesis Single-cell RNA sequencing (scRNA-seq) of pancreatic islet tissue is a powerful tool for investigating Type 1 Diabetes (T1D). However, individual datasets are limited in size and fragmented across donors, laboratories, and experimental conditions, highlighting the need for a unified single-cell atlas. This study aimed to construct a comprehensive, integrated scRNA-seq map of human isolated pancreatic islets by collating data from diverse sources. Methods Publicly available scRNA-seq datasets derived from isolated pancreatic islets, generated and/or provided by the Human Pancreas Analysis Program (HPAP), Prodo Labs, and the Integrated Islet Distribution Program (IIDP), were collected. Systematic quality controls were implemented to select high-quality samples, reads and cells. Data integration was conducted, accounting for important variables such as age, sex, body mass index (BMI), origin study, treatments, islet data/distribution resources, and sequencing chemistry. Results We generated a comprehensive single-cell atlas of human pancreatic islets comprising 191 high-quality assays from 140 donors (59 female, 81 male) across five phenotypic groups: controls without diabetes (69 donors), autoantibody-positive donors without diabetes (12), pre-diabetic donors (11), donors with type 1 diabetes (12), and donors with type 2 diabetes (36). The atlas also includes experimentally perturbed samples, including those exposed to SARS-CoV-2 infection and pro-inflammatory cytokines. In total, the atlas contains 448,935 cells, capturing major endocrine islet populations, such as alpha cells (43.3%) and beta cells (26.8%), as well as non-endocrine populations such as endothelial cells (0.75%) and immune cells (0.6%). Conclusions/interpretation By uniformly harmonizing and integrating data from multiple sources, we have developed a comprehensive single-cell atlas of isolated human pancreatic islets, which is publicly available at www.pankbase.org. The atlas provides a platform for hypothesis-driven investigation of diabetes pathophysiology and, given rigorous quality control at the read, barcode, and sample levels alongside careful metadata curation, is well suited for downstream machine-learning applications.