MHNfs: Prompting In-Context Bioactivity Predictions for Low-Data Drug Discovery.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Drug discovery increasingly utilizes computational and machine learning (ML) methods.
• A significant hurdle in these approaches is the scarcity of high-quality data.
• Addressing data limitations is crucial for advancing computational drug discovery.

Purpose of the Study:

• To introduce MHNfs, an application designed for molecular activity prediction in low-data settings.
• To provide an accessible tool for researchers to leverage few-shot learning in drug discovery.
• To enable precise activity predictions using minimal known molecular data.

Main Methods:

• Development of the MHNfs application, incorporating a state-of-the-art few-shot learning model.
• Utilizing the MHNfs model, which demonstrated strong performance on the FS-Mol benchmark dataset.
• Simulating real-world drug discovery scenarios by adapting PubChem bioassays into few-shot prediction tasks.

Main Results:

• The MHNfs model achieved strong performance in few-shot activity prediction tasks.
• The application provides an intuitive interface for users to obtain molecular activity predictions.
• Evaluation using adapted PubChem bioassays confirmed the application's efficacy in low-data scenarios.

Conclusions:

• MHNfs offers a streamlined and accessible solution for deploying advanced few-shot learning models.
• The application effectively addresses the challenge of data scarcity in computational drug discovery.
• MHNfs serves as a valuable tool for accelerating the identification of novel drug candidates.