Agentic Curation of the Pathophysiome (GlySpace Alliance Webinar)

Slides (PDF and PPTX) from a talk given online on 27 October 2026 in the GlySpace Alliance Webinar series, hosted by GlyGen. Gene–phenotype links say what a disease does but not how. The talk argues for curating the mechanism in between as a causal graph, a pathograph, and presents DisMech, a knowledge base of disease mechanisms: 8,676 diseases and subtypes in 3,753 causal graphs, with 111,189 nodes and 85,259 causal edges. Using Fabry disease as a worked example, it walks from the GLA mutation through α-galactosidase A deficiency and Gb3 storage to tissue-specific branches in kidney, heart and nerves, with each node grounded in an ontology (HGNC, GO, ChEBI, CL, UBERON, HPO). Treatments, disease models, competing hypotheses and open knowledge gaps attach to the nodes they concern, and every claim carries a quoted receipt from the literature (290,315 quotes across the knowledge base). It shows how a pathophysiological event ontology links diseases, with examples from congenital disorders of glycosylation, and how pathographs support drug repurposing, variant prioritization and knowledge graph enrichment. The second half covers how DisMech is built: a LinkML schema that constrains each slot to the right ontology branch; general-purpose agents working inside a domain harness of skills, tools and validators; deterministic checks of identifiers and quotations with linkml-term-validator and linkml-reference-validator; and people steering the agents, including curators, a drug developer and parents of rare disease patients. It closes with ways glycobiologists could contribute: closing glyco knowledge gaps, grounding glycan nodes to GlyGen, and reviewing entries.