Leveraging Large Language Models for Molecular Property Prediction in Scientific Research

Recent advancements in large language models (LLMs) have seen their implementation across various domains, providing assistance in numerous tasks from text classification and translation to complex document searches. The adaptable nature of general-purpose Large Language Models (LLMs) prompts a pivotal question: Can these models be effectively fine-tuned to serve as artificial intelligence assistants in scientific research, particularly in predicting molecular properties? In this work, we conducted two experiments focusing on fine-tuning OpenAI’s Ada model to predict both the antimicrobial potential of peptides and the binding affinity of ligands to target proteins. The results show that a well-designed prompt-completion can successfully empower the Ada model to predict molecular functions of peptides or small molecules.