This research introduces PEIT (Property Enhanced Instruction Tuning), a two-step framework that significantly improves large language models' ability to generate molecules with specific biochemical properties.

• Addresses the challenge of limited labeled data for molecule generation tasks
• Enhances LLMs' capabilities in handling complex multi-property constraints
• Leverages instruction tuning techniques specifically designed for molecular applications
• Demonstrates potential for accelerating drug discovery processes

For the medical field, this breakthrough enables more efficient development of candidate molecules with desired therapeutic properties, potentially reducing drug development timelines and costs while improving success rates.

Property Enhanced Instruction Tuning for Multi-task Molecule Generation with Large Language Models