Prompt-based multimodal representation learning for drug repurposing.

Area of Science:

• Computational chemistry
• Drug discovery
• Artificial intelligence in medicine

Background:

• Drug repurposing accelerates drug discovery by reducing costs and timelines.
• Current deep learning methods often use static molecular structures, limiting the capture of dynamic interactions.
• Accurate prediction of compound-protein interactions is crucial for effective drug repurposing.

Purpose of the Study:

• To develop an innovative prompt-based multimodal representation learning framework for enhanced drug repurposing.
• To address the limitations of static representations in predicting dynamic compound-protein interactions.
• To improve the accuracy of identifying drug candidates for specific therapeutic targets.

Main Methods:

• Introduced a prompt-based multimodal representation learning framework.
• Developed a dynamic prompt generation module for adaptive, receptor-specific prompts.
• Implemented a prompt calibration module for multimodal feature integration and optimization.
• Applied the framework to identify FDA-approved drug candidates targeting G-protein-coupled receptors.

Main Results:

• Achieved a 7.4% improvement in mean absolute error compared to state-of-the-art methods.
• Demonstrated up to a 25.1% improvement for specific drug targets.
• Successfully identified potential non-opioid pain management treatments.

Conclusions:

• The proposed framework enhances the accuracy of compound-protein interaction predictions for drug repurposing.
• This approach can significantly advance drug discovery, offering solutions for various therapeutic needs, including safe pain management.
• The dynamic encoding of contextual information is key to improving prediction accuracy.