> For the complete documentation index, see [llms.txt](https://docs.neurolabai.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.neurolabai.io/engineering-and-scientific-assurance/knowledge-graph.md). # Knowledge Graph The Neurolab knowledge graph is a governed substrate for model input, retrieval, evidence organization, and exploratory reference layers. It connects compounds with typed biological, chemical, clinical, cognitive, and brain-related entities. ## Two release identities The predictive model and the facts-and-evidence layer advance independently: * **v43.262 predictive model:** the checkpoint-bound graph projection and known-compound model family. * **v44.16 facts and evidence:** the latest candidate facts-and-evidence package for governed identities, records, evidence coverage, receptor context, and report support. The candidate package shows how provenance and report context can improve without retraining the predictive model or changing its raw output heads. Its availability in source does not establish that a beta deployment is serving it. Reports should preserve the actual model and facts-and-evidence identities returned by the environment in which they were generated. ## Four roles of the graph ### 1. Model substrate The known-route checkpoints consume an allowlisted projection of typed relations among compounds, receptors, genes, regions, and ontology concepts. Relations used as supervision targets for the de-leaked heads are excluded from message passing. ### 2. Retrieval substrate Graph-derived artifacts help resolve identities, find chemical or mechanism context, and retrieve records for report composition. Retrieval relevance does not automatically establish query-specific support. ### 3. Evidence artifact The application can receive a filtered, report-local graph containing nodes, edges, paths, sources, evidence records, and support links. This response-safe artifact is not the complete graph and is not a literal trace of GNN computation. ### 4. Reference substrate Atlases, receptor densities, nuclei, tracts, population and pathway overlays, dosage records, and circuit frameworks can provide context. Reference presence does not make a layer a model feature or a validated biological transition weight. ## Binding and receptor evidence The facts-and-evidence layer distinguishes observed binding, measured non-binding, derived or inferred relationships, and unresolved coverage. It also preserves affinity measurements and receptor-level trust metadata when available. Dual activity can be represented when supported by compatible evidence. Conflicting or incomplete records remain explicitly unresolved rather than being forced into one action label. Negative, reverse-traversal, and noncausal relations cannot silently become positive receptor support. ## Provenance and visibility Each surfaced fact should carry enough provenance to identify its source, transformation, evidence state, and release. Source policy determines whether a record can be shown directly, summarized, or withheld. Visibility is not the same as scientific quality. A visible record can still be weak or irrelevant, while a restricted record can still be scientifically informative within its permitted use. Provenance, relevance, licensing, and evidentiary strength remain separate dimensions. When a required source or release identity is missing or mismatched, user-facing evidence should fail closed rather than hydrate an uncertain record. ## Edge meaning Every relation has typed semantics. An edge weight is meaningful only under the rule for that relation and is not generally comparable across relation families. Graph proximity is not causality. Reverse edges can support navigation without asserting that the biological relation is symmetric. ## Brain and mechanism reference layers Circuit, nucleus, tract, atlas, receptor-density, cell-type, and population/pathway material supports exploration without becoming a running virtual brain: * reference paths do not imply compound-specific activation; * receptor density is not functional engagement; * a local operator sign is not a global endpoint sign; * cross-species evidence applies only under its reviewed transfer claim; and * reference paths are not validated simulation weights. ## What users see The Knowledge Graph Explorer shows the support and context available for the current report. Users can search and filter the artifact, inspect nodes, edges, evidence, sources, and paths, and export Graph JSON plus SVG or PNG visualizations. Its 3D Evidence Constellation depicts abstract topology, not anatomy. Next: [Model Training](/engineering-and-scientific-assurance/model-training-overview.md) --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.neurolabai.io/engineering-and-scientific-assurance/knowledge-graph.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.