AI-powered Immune Cell Knowledge Graph (ICKG) with granular immune contexts enables immune program interpretation.

Area of Science:

• Immunology and Computational Biology
• Leveraging advanced computational methods for biological data analysis
• Integrating multi-omics data for immune system research

Background:

• Single-cell and spatial omics have advanced immune cell profiling but knowledge remains fragmented.
• Synthesizing immune cell states, functions, and gene regulation from literature is challenging.
• Existing databases lack context-aware relationships crucial for mechanistic understanding.

Purpose of the Study:

• To construct Immune Cell Knowledge Graphs (ICKGs) for integrating fragmented immune cell knowledge.
• To enable context-aware reasoning about immune cell functions and gene regulation.
• To facilitate mechanistic hypothesis generation in immune omics research.

Main Methods:

• Utilized large language models (LLMs) to process over 24,000 PubMed abstracts focused on cancer immunotherapy.
• Developed four cell type-specific knowledge graphs with human-verifiable validation.
• Validated ICKGs against perturbation datasets from cytokine stimulation and CRISPR experiments.

Main Results:

• ICKGs capture directed, literature-supported relationships among genes, pathways, and immune functions.
• ICKGs provide more accurate and immunologically coherent contexts than canonical databases.
• Identified interpretable pathway annotations, including novel signatures for immuno-oncology.

Conclusions:

• ICKGs offer a scalable framework for functional interpretation in immune omics.
• The developed interactive portal enables community access to ICKG-based pathway annotations.
• This work enhances mechanistic hypothesis generation by integrating literature-derived immune cell knowledge.