HMAMP: Designing Highly Potent Antimicrobial Peptides Using a Hypervolume-Driven Multiobjective Deep Generative...

Area of Science:

• Biomaterials Science
• Computational Chemistry
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) show promise against multidrug-resistant bacteria.
• Current generative models often neglect the multiobjective nature of AMP discovery, leading to high candidate attrition.
• Optimizing multiple AMP attributes simultaneously is crucial for effective drug development.

Purpose of the Study:

• To introduce a novel approach, hypervolume-driven multiobjective AMP design (HMAMP), for simultaneous optimization of multiattribute AMPs.
• To address the limitations of existing generative models in AMP discovery by incorporating multiobjective optimization.
• To improve the effectiveness and diversity of generated antimicrobial peptides.

Main Methods:

• Synergistic integration of reinforcement learning and gradient descent algorithms based on hypervolume maximization.
• Utilizing HMAMP to bias generative processes and mitigate pattern collapse.
• Employing a knee-based decision strategy for rapid screening of promising candidates.

Main Results:

• HMAMP significantly outperforms state-of-the-art methods in AMP effectiveness and diversity.
• The approach successfully biases generative processes and addresses pattern collapse.
• A knee-based strategy efficiently identifies candidates with desirable properties.

Conclusions:

• HMAMP offers an effective strategy for navigating complex search spaces in AMP design.
• The method facilitates the discovery of antimicrobial peptides balancing ideal properties with realistic constraints.
• HMAMP enhances the development pipeline for novel antimicrobial biomaterials.