Generative Design of Functional Metal Complexes Utilizing the Internal Knowledge and Reasoning Capability of Large...

Area of Science:

• Artificial Intelligence in Chemistry
• Computational Materials Science
• Chemical Optimization

Background:

• Large language models (LLMs) show potential in scientific applications but their use in discovery is underexplored.
• Optimizing complex chemical systems often requires extensive computational resources and expertise.

Purpose of the Study:

• To introduce LLM-EO, an integration of LLMs with evolutionary optimization for scientific discovery.
• To demonstrate the efficacy of LLM-EO in optimizing transition metal complexes (TMCs).
• To enhance accessibility of multiobjective optimization for chemists.

Main Methods:

• Integrating LLMs into an evolutionary optimization framework (LLM-EO).
• Utilizing LLMs' intrinsic chemical knowledge and historical data for optimization.
• Employing natural language instructions for multiobjective optimization tasks.

Main Results:

• LLM-EO demonstrated success in optimizing transition metal complexes.
• The approach showed advantages in few-sample learning by leveraging LLM knowledge.
• LLM-EO facilitated novel ligand and TMC generation with unique properties.

Conclusions:

• LLM-EO offers an accessible and efficient method for chemical optimization and discovery.
• The integration of LLMs with evolutionary optimization holds broad potential for chemistry and materials design.
• Advancements in LLMs are expected to further expand the applications of LLM-EO.