Design of CO2-philic Molecular Units with Large Language Models

The integration of large language models (LLMs) into chemical sciences presents a transformative approach for molecular design. In this study, we explore the capabilities of LLMs for generating novel molecular structures with enhanced CO₂ affinity for the development of novel physisorptionbased carbon capture technologies. A hybrid workflow was employed, combining pretrained LLMs (GPT-4o, Llama-3, and Gemini 2.0) with computational chemistry techniques. LLMs were prompted to propose molecular candidates, which were subsequently evaluated using density functional theory (DFT) to assess their interaction energies with CO₂. Through iterative model refinement and expert-guided modifications, a dataset of 95 molecular units was generated, revealing several promising candidates with CO₂ interaction energies exceeding the -7 kcal/mol threshold for optimal physisorption. Notably, LLM-generated structures showcased emergent design strategies, such as cooperative binding motifs, that aligned with domain knowledge and experimental precedent. This study highlights the potential of LLMs as powerful tools for molecular discovery while underscoring the need for expert oversight and rigorous validation in AI-driven chemical research.