A Study on Semantic Molecular Prompt Engineering Method for Large Language Model-based Molecular Property Prediction

Key Points

• This research aims to enhance the prediction accuracy of molecular properties using a novel approach called S-MolPrompt, integrating structural and physicochemical data.
• Developed S-MolPrompt based on a pre-trained large language model.
• Integrated molecular structural information with physicochemical properties for analysis.
• Evaluated performance on three benchmark datasets comparing against traditional methods.
• Compared various prompt templates to assess their impact on prediction accuracy.
• S-MolPrompt outperformed traditional molecular fingerprint models and deep learning models.
• Significant improvement in predictive accuracy was observed across all benchmark datasets.
• Demonstrated the importance of prompt engineering in molecular property prediction.