AI-assisted workflows for integrating data using LinkML

AI-assisted workflows for integrating data using LinkML Chris Mungall Lawrence Berkeley National Laboratory Open Data in Neurophysiology (ODIN) Symposium 2025 2025-04-09 https://linkml.io If only all data was shaped like this… https://sos.noaa.gov/catalog/datasets/climate-model-surface-temperatur e-change-ssp1-sustainability-2015-2100/ …in our world, data looks like this Rübel, Tritt, Ly, .. (2022) The Neurodata Without Borders ecosystem for neurophysiological data science eLife 11:e78362 https://doi.org/10.7554/eLife.78362 ● Heterogeneous ● Linked ● Numerical + Categorical ● Rich metadata …and like this ● Multi-omics, multi-modal ● Multiple coordinate systems ● Rich metadata Linked and labeled data required for training and tuning foundation models AI Applications Ontologies help standardize and label data… Examples: ● dopamine receptor activity, dendrite, synaptic vesicle cycle… Uses: ● Unifying functional descriptions ● Interpreting omics data Examples: ● GABAergic inhibitory neuron, … Uses: ● Unifying cell-type vocabulary ● Single-cell annotation Examples: ● primary motor cortex, CA1 field of hippocampus Uses: ● Standard for animal (neuro)anatomy ● Mapping anatomical atlases ● Gene expression annotation Examples: ● Parkinson disease, rolandic epilepsy-speech dyspraxia syndrome Uses: ● Integrating clinical and genetic data ● Diagnosis and variant prioritization https://obofoundry.org Gene function Disorders Cell types (metazoa) Gross anatomy (metazoa) Ontologies: necessary but not sufficient Incompatible Schemas ! We have many frameworks to structure data SHACL Semantic web architect Developers Data Scientist Scientists, Clinicians, .. SQL DDL HDF5 NetCDF Pandas Parquet Excel ISO-11179 CDEs FHIR Clinical modeler JSON Schema Ontologist OWL Pydantic ProtoBuf GraphQL ShEx LinkML is a converter box https://linkml.io Example of use: Microbiome multi-omics Data collection app Metadata standards to enable microbiome analysis ● Environmental sample data ● Omics data ● Community development model Data portal Core concepts: Study Environmental Sample Workflow Analysis (genomic, metabolomic, ..) Data Object https://microbiomedata.github.io/nmdc-schema/ LinkML: The community Open and inclusive community https://github.com/linkml/linkml/graphs/contributors https://linkml.io/linkml/faq/contributing.html Making multidimensional data AI ready Rübel et al, https://doi.org/10.1101/523035 ??? Hey, LinkML looks great for semantics. How do we do arrays? v1.6.0 (2023) Making multidimensional data AI ready Rübel et al, https://doi.org/10.1101/523035 ??? Hey, LinkML looks great for semantics. How do we do arrays? Umm, you can’t* *at least not as first-class structures, they need to be pivoted such that cells are objects... v1.6.0 (2023) Adding first-class arrays to LinkML linkml.io/linkml/schemas/arrays.html AI for data and knowledge extraction Papers Lab notebooks Clinical notes Databases Datasets Ontology enabled structured data lakes Knowledge Graphs Semantic Schemas 2020-22 2023 2024 2025 ● NER/CR ● Ontology APIs Schema Automator ● Zero-shot LLM ● LinkML extraction OntoGPT ● RAG-based CurateGPT ● Agentic ● Tools ● MCP Aurelian AI for data extraction and integration Caufield et al 10.1093/bioinformatics/btae104 Toro et al 10.1093/bioinformatics/btae104 gh: monarch-initiative/aurelian linkml.io/schema-automator Aurelian LinkML Agent (via MCP) monarch-initiative.github.io/aurelian/agents/linkml_agent AI ready polymorphic data lakehouses with LinkML Store Create Read Update Delete Search Inference Validate database contents using my schema Ask questions using RAG Make inferences (LLM, ML) from data linkml.io/linkml-store AI Readiness: Beyond FAIR # Muñoz Torres & Clark, Bridge2AI Leadership Meeting 2024-12-06 Standards WG Available Resources Clark et al, https://doi.org/10.1101/2024.10.23.619844