Bridging data and discovery: a survey on knowledge graphs in AI for science.

Area of Science:

• Computer Science
• Artificial Intelligence
• Bioinformatics
• Cheminformatics
• Materials Informatics

Background:

• Knowledge graphs are crucial for organizing scientific data and enabling AI-driven research.
• Scientific knowledge graphs (SciKGs) are increasingly vital for accelerating discovery across various scientific domains.
• The synergy between SciKGs and large language models (LLMs) is a rapidly developing area.

Purpose of the Study:

• To provide a comprehensive survey of scientific knowledge graphs (SciKGs).
• To examine the construction methodologies and applications of SciKGs in biology, chemistry, and materials science.
• To highlight the integration of SciKGs with LLMs for advancing AI for science.

Main Methods:

• Literature review and synthesis of existing research on SciKGs.
• Analysis of SciKG applications in drug development, omics analysis, reaction prediction, and materials design.
• Exploration of the integration framework combining SciKGs and LLMs.

Main Results:

• SciKGs effectively support complex scientific tasks like drug discovery and materials design.
• The integration of SciKGs with LLMs creates a knowledge- and language-driven discovery framework.
• Key challenges and opportunities for developing auditable, interoperable, and self-evolving SciKGs were identified.

Conclusions:

• SciKGs, coupled with LLMs, form a foundational infrastructure for autonomous scientific discovery.
• Future SciKG-centered ecosystems will feature self-updating graphs and AI scientists for accelerated research.
• The development of advanced SciKGs is essential for the next generation of AI-driven scientific breakthroughs.