This research explores how Large Language Models can convert standard graphs into text-attributed graphs (TAGs), enabling powerful cross-domain analysis previously hindered by feature compatibility issues.

• Solves the challenge of analyzing graphs with different feature spaces
• Leverages LLMs to create universally compatible text-attributed nodes
• Enables transfer learning across diverse graph domains
• Particularly valuable for cross-domain applications like drug discovery

Medical Impact: By enabling more effective cross-domain graph analysis, this approach could significantly accelerate drug discovery processes, allowing researchers to transfer insights between molecular structures and identify novel therapeutic candidates more efficiently.

Can LLMs Convert Graphs to Text-Attributed Graphs?