A Hierarchical Multimodal Knowledge Graph for Neural Cell-Type-Specific Regulation Integrating Single-Cell...

Area of Science:

• Neuroscience
• Bioinformatics
• Computational Biology

Background:

• The nervous system has diverse cell types with unique molecular regulatory programs.
• Existing regulatory evidence is fragmented across literature, using inconsistent nomenclature, hindering integration and comparison.

Purpose of the Study:

• To construct a standardized, computable neural-cell-centric multimodal knowledge graph.
• To transform scattered regulatory evidence into a unified resource for systematic analysis.

Main Methods:

• Established a three-level hierarchical cell-type taxonomy (79 nodes) based on the Cell Ontology.
• Integrated large-scale human brain single-cell transcriptomic data (4M+ cells) for molecular fingerprints.
• Utilized a large language model to curate 25,812 regulatory evidence records from PubMed abstracts.

Main Results:

• Developed a Neo4j knowledge graph with 41,532 directed relationships.
• Created a training subgraph (19,819 triples, 10,660 entities) for knowledge graph embedding.
• Demonstrated cell-type-specific link prediction for identifying candidate regulators and markers, exemplified by microglia.

Conclusions:

• The framework provides a structured basis for cross-study comparison in neural cell-type-specific regulation.
• Enables hypothesis generation and knowledge-guided reasoning for understanding complex regulatory networks.
• Facilitates systematic integration and analysis of fragmented neurobiological data.