BioMedGraphica: An All-in-One Platform for Joint Textual Biomedical Prior Knowledge and Numeric Graph Generation.

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Artificial intelligence (AI) revolutionizes scientific discovery by integrating large datasets.
• Foundation models like large language models (LLMs) and large vision models (LVMs) are crucial for general AI.
• Biomedical data is fragmented across diverse databases with inconsistent nomenclature, challenging AI integration in Precision Health and Medicine (AI4PHM).

Purpose of the Study:

• To develop an integrated platform, BioMedGraphica, for foundation AI model development in AI4PHM.
• To create a unified text-attributed knowledge graph (TAKG) harmonizing diverse biomedical data.
• To support AI-driven knowledge discovery and the development of novel AI models for biomedical research.

Main Methods:

• Constructed BioMedGraphica, a TAKG with 3,131,788 entities and 56,817,063 relations from 43 biomedical databases.
• Harmonized 11 entity types and 29 relation types, assigning unique IDs and textual features to all elements.
• Developed a graphical user interface (GUI) for accessing the knowledge graph and generating knowledge-multi-omic signaling graphs.

Main Results:

• BioMedGraphica integrates heterogeneous biomedical data into a cohesive knowledge graph.
• The platform supports zero-shot or few-shot knowledge discovery through relation prediction.
• Researchers can access and utilize the graph for AI model development in AI4PHM.

Conclusions:

• BioMedGraphica addresses the challenge of fragmented biomedical data for AI applications.
• The platform facilitates AI-ready data access for discovering disease pathogenesis, therapeutic targets, and drugs.
• It enables the development of advanced AI models for accelerating scientific discovery in precision health and medicine.