ActivityDiff: A diffusion model with Positive and Negative Activity Guidance for De Novo Drug Design

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Generative models for molecular design

Background:

• De novo drug design aims to create novel molecules with desired target activity while avoiding off-target interactions.
• Current generative models often prioritize positive activity, neglecting negative activity data crucial for safety and selectivity.
• Off-target effects are a significant challenge, compromising drug selectivity and safety.

Purpose of the Study:

• To develop a generative framework that explicitly incorporates negative activity information for improved molecular design.
• To enhance drug discovery by simultaneously optimizing for desired target activity and minimizing undesired interactions.
• To provide a computational tool for generating safer and more selective drug candidates.

Main Methods:

• ActivityDiff, a classifier-guided diffusion framework for activity-controlled molecular generation.
• Utilizes separately trained drug-target classifiers for both positive and negative guidance.
• Incorporates explicit negative guidance to suppress harmful off-target interactions during generation.

Main Results:

• ActivityDiff effectively supports various drug design tasks, including single- and dual-target generation.
• Demonstrates successful fragment-constrained dual-target design and selective generation for improved target specificity.
• Achieves significant reduction in off-target effects, enhancing overall drug safety and selectivity.

Conclusions:

• Classifier-guided diffusion with explicit negative guidance is an effective strategy for molecular design.
• ActivityDiff enables joint optimization of drug efficacy and safety by considering both positive and negative activity.
• This approach represents a significant advancement in developing safer and more selective therapeutic agents.