Polymer-Agent: Large Language Model Agent for Polymer Design.

Area of Science:

• Materials Science
• Computational Chemistry
• Polymer Science

Background:

• Polymer discovery is crucial for industries like biomedical and materials science.
• Traditional polymer experimentation relies on resource-intensive trial-and-error methods.
• Machine learning accelerates polymer property prediction but lacks accessibility for lab researchers.

Purpose of the Study:

• To develop an accessible, closed-loop polymer structure-property predictor for early-stage discovery.
• To integrate LLM reasoning for property prediction and structure generation.
• To address infrastructure limitations hindering direct access to computational models for researchers.

Main Methods:

• Developed a terminal-integrated framework for polymer structure-property prediction.
• Utilized Large Language Model (LLM) reasoning for core functionalities.
• Incorporated synthetic accessibility and complexity scores to guide polymer generation via SMILES sequences.

Main Results:

• Enabled on-demand polymer property prediction within a terminal interface.
• Facilitated property-guided generation and modification of novel polymer structures.
• Ensured generated polymer structures are synthetically accessible at the monomer level.

Conclusions:

• The framework provides computational insights and accelerates early-stage polymer discovery for laboratory researchers.
• Addresses the challenge of generating novel, synthetically viable polymer structures.
• Enhances accessibility of advanced computational tools for polymer science research.