Introduction and Objective: Type 1 Diabetes (T1D) is a lifelong autoimmune disease with major impacts on health, quality of life, and healthcare systems worldwide. Despite decades of research, progress is often slowed by fragmented knowledge spread across isolated databases, publications, and experimental platforms. These resources are rich but heterogeneous, making it difficult to connect genetic, molecular, cellular, clinical, and imaging evidence into a coherent view. Methods: Here we present PanKgraph, a comprehensive knowledge graph for Type 1 Diabetes that integrates diverse data types into a unified, machine-readable framework. Unlike traditional databases, which store data in fixed schemas and silos, a knowledge graph explicitly represents entities and their relationships, enabling flexible integration of multi-modal experimental data, biomedical literature, and curated domain knowledge. This structure supports cross-scale reasoning, hypothesis generation, and discovery of non-obvious connections across biological systems. Results: To serve the broader research community, we provide an open, web-based, question-driven interface that allows users to explore T1D knowledge through interactive graphs, concise textual summaries, and visual representations linked to underlying experimental evidence. We further expose an application programming interface (API) designed for large language models, enabling joint analysis of scientific images, publications, and raw experimental data. Conclusion: By lowering barriers to data integration and interpretation, PanKgraph aims to shift T1D research beyond single-axis thinking, foster collaboration across disciplines, and accelerate innovative, systems-level insights into disease mechanisms and therapies. Disclosure Y. Wang: None. R. Mao: None. F. Feng: None. H.T. Vu: None. Y. Huang: None. Z. Han: None. A.K. Huber: None. J. Cartailler: None. S. Chen: Stock/Shareholder; Current; iOrganBio Inc. Stock/Shareholder; Ended; Oncobeat. M. Brissova: None. S. Parker: Research Support; Current; Pfizer Inc. Consultant; Ended; Novo Nordisk. J. Liu: None. Funding National Institute of Diabetes and Digestive and Kidney Diseases (5U24DK138515-02)

Introduction and Objective: Type 1 diabetes (T1D) is a chronic autoimmune disease that imposes a substantial burden on medical systems. T1D and pancreas related data are distributed across multiple portals. This dispersion complicates mechanism-focused analyses and creates barriers for T1D research AI agent. Additionally, most existing initiatives use conventional tabular databases, which struggle to represent and query higher-order regulatory and mechanistic relationships in an intuitive and efficient way. Overall these limitations decrease utility for modern AI and system biology workflows. Methods: PanKgraph addresses these challenges with a knowledge graph based representation that enables direct traversal of biologically meaningful relationships, allowing users and AI agents to explore how genetic variants, genes, and traits interact around the pancreas. Its schema is designed to accommodate heterogeneous data while unifying information from multiple programs into a reusable framework. This structure supports systematic evidence aggregation, and provides AI-ready data for downstream research. Results: As large language models and other AIs are increasingly used in biology, their lack of verifiable evidence remains a major concern. PanKgraph provides a structured, provenance-aware data source for agentic AI, supporting queries that return interpretable subgraphs with rich metadata. By constraining reasoning to curated entities and biologically meaningful relations, it reduces hallucinations, improves reproducibility, and preserves transparent evidence chains. Programmatic API access further allows external AI agents to interoperate with PanKgraph, establishing it as a shared platform for AI-assisted T1D research. Conclusion: Together, PanKgraph provides an agent-ready, provenance-aware framework that unifies pancreas-centered resources for transparent interrogation of variant-gene-trait relationships. This resource establishes a scalable foundation for evidence-driven AI workflows in T1D and related diabetes research. Disclosure R. Mao: None. Y. Wang: None. H.T. Vu: None. F. Feng: None. Z. Han: None. Y. Huang: None. A.K. Huber: None. S. Chen: Stock/Shareholder; Current; iOrganBio Inc. Stock/Shareholder; Ended; Oncobeat. M. Brissova: None. J. Cartailler: None. J. Liu: None. S. Parker: Research Support; Current; Pfizer Inc. Consultant; Ended; Novo Nordisk. Funding National Institute of Diabetes and Digestive and Kidney Diseases (5U24DK138515-02)

In type 1 diabetes (T1D), α cell function is dysregulated with loss of glucagon secretion in response to hypoglycemia that precedes a secretory impairment of major α cell stimulus, epinephrine, produced by adrenal chromaffin cells. Recurrent hypoglycemic episodes, which correlate with progressive β cell loss in T1D, lead to an impairment of sympathoadrenal responses and the risk of a life-threatening hypoglycemia unawareness syndrome. To discover new molecules that activate signaling pathways in human α cells and promote glucagon secretion during hypoglycemia, we used our primary human pseudoislet system for screening of 1027 FDA-approved compounds from ion channel and GPCR compound library (APExBIO). The compound screen was optimized for pseudoislet formation in 96-well format including state-of-the-art liquid-handling platforms for compound addition and analysis of glucagon secretion from a single pseudoislet in the presence of carefully selected internal standards. Using this approach, we identified several ‘hit’ compounds targeting both ion channels and GPCRs that increased glucagon secretion above positive controls including 1.7 mM glucose + 20 mM arginine. Notably, nearly 30 compounds targeting GPCRs including opioid, oxytocin, histamine H1, dopamine, 5-HT, adrenergic, and muscarinic receptors elicited robust glucagon responses across multiple human islet donors (N=3). Additionally, we leveraged our single cell and bulk RNA-seq datasets to evaluate target specificity and facilitate the compound selection process based on target expression changes in T1D α cells. Overall, the primary human pseudoislet system is ideal for the screen because it maintains 3-D islet cell arrangement, islet microenvironment, and allows for generation of α cell-enriched, T1D-like pseudoislets as a model system for further compound validation. This could lead to repurposing of FDA-approved drugs for treatment of hypoglycemia and rapid translation for clinical testing. Disclosure T.S.R. Bate: None. C. Reihsmann: None. R. Aramandla: None. C. Davis: None. S. Mei: None. A. Bradley: None. J. Bauer: None. S. Parker: Research Support; Pfizer Inc. D.C. Saunders: None. A.C. Powers: None. M. Brissova: None.

Pancreatic islet endocrine cell phenotype and function are tightly regulated by transcription factor (TF) regulatory networks. In a mouse model of pancreas organogenesis, the key islet-enriched TF NKX2.2 governs both β and α cell specification. Further, loss of NKX2.2 in adult mouse β cells results in glucose intolerance, reduced insulin content, and polyhormonal β cells, demonstrating its importance in the maintenance of β cell phenotype. To understand the undefined role of NKX2.2 in adult human islets, we utilized adenoviral delivery of CRISPR and shRNA constructs in our primary human pseudoislet system to knockout (gNKX2-2) or knockdown (shNKX2-2) NKX2-2 in islet cells, respectively (n=4-6 donors without diabetes). Immunofluorescence analyses of gNKX2-2 pseudoislets demonstrated loss of NKX2.2 signal across β (81 ± 2.0%; p<0.001), α (58 ± 5.5%; p=0.001), and δ cells (83 ± 2.2%; p<0.001) compared to scrambled controls (gSCR). Surprisingly, functional assessment by dynamic perifusion revealed increased insulin secretion in gNKX2-2 versus gSCR pseudoislets measured as area under the curve in response to 16.7 mM glucose (6.64 ± 1.53 vs. 3.55 ± 1.53 ng/100 islet equivalents (IEQs); p=0.007) and cAMP-evoked potentiation with 16.7 mM glucose + 100 μM isobutylmethylxanthine (IBMX) (5.43 ± 0.50 vs. 3.15 ± 0.59 ng/100 IEQs p=0.003) without changes in total insulin content (gNKX2-2 vs. gSCR: 3.64 ± 0.75 vs. 4.78 ± 0.87 ng/IEQ, p=0.45). Furthermore, these findings were recapitulated in shNKX2-2 versus shSCR human pseudoislets (16.7 mM glucose: 10.17 ± 2.46 vs. 6.50 ± 2.51 ng/100 IEQs, p=0.02; 16.7 mM glucose + IBMX: 8.43 ± 1.31 vs. 4.21 ± 1.33 ng/100 IEQs, p=0.0002; insulin content: 6.18 ± 1.05 vs. 7.53 ± 1.26 ng/IEQ, p=0.36), where we observed a reduction in NKX2.2hi β (65 ± 10%; p=0.008), α (52 ± 5.5%; p=0.003), and δ cells (83 ± 4.6%; p=0.003). Collectively, our data indicate a critical role of NKX2-2 in the maintenance of primary adult human β cell function and suggest differential roles in islet function across species. Disclosure Y.D.Pettway: None. S.Parker: Research Support; Pfizer Inc. A.C.Powers: None. M.Brissova: None. J.Walker: None. C.Dai: None. R.Aramandla: None. A.L.Hopkirk: None. C.Reihsmann: None. C.Davis: None. R.Jenkins: None. L.Sussel: None.

be extracted from the VLM.The course of FEV1 will be evaluated in relation to treatment with antibiotics and compliance to daily tests will be assessed. Conclusion:The effect of two weeks of intravenous antibiotics on FEV1 will be assessed.Compliance to daily measurements will be compared to quality of life and health literacy.Patients' experiences with home monitoring will be used to further develop the setup for home monitoring.347 Cystic Fibrosis CareMessage (CFCM) short message service C. Landon 1 .