What problem does it solve? Constructing a domain-specific knowledge graph requires combining curated databases with unstructured scientific literature, which is tedious and error-prone to do manually. This Skill provides a three-phase pipeline that ingests structured data sources, extracts claims from PubMed abstracts with LLMs, and generates ranked research hypotheses from the resulting graph. ## Core Features & Use Cases - Structured Data Ingestion: Import concepts and relations from NeuroNames, MeSH, DisGeNET, Cognitive Atlas, and Nilearn brain atlases into a NetworkX DiGraph with typed edges and provenance. - Literature Claim Extraction: Search PubMed, deduplicate candidates against the existing graph, extract structured claims (subject, predicate, object, p-value, sample size, conditions, population) via LLM prompts, and resolve entities with a 5-level cascading matcher. - Hypothesis Engine: Batch-generate cross-domain hypotheses, detect contradictions and gaps, find trending evidence, and rank results by confidence, novelty, evidence, and testability. - Use Case: A neuroimaging researcher runs batch extraction over 10 diseases across 27 years of PubMed abstracts with checkpoint/resume support, then queries the graph for novel hippocampus-to-disease paths to prioritize experiments. ## Quick Start Ask the agent to build a knowledge graph by ingesting atlas data and extracting claims from PubMed papers about Alzheimer's disease neuroimaging studies.
