Designing broad-spectrum antimicrobial peptides based on conditional feedback generation adversarial network

Antimicrobial peptides (AMPs) offer promising alternatives to conventional antibiotics due to their broad-spectrum activity, low toxicity, and reduced risk of resistance development. However, the existing AMP design methods struggle to balance generation efficiency with pathogen-specific targeting, which limits their effectiveness against multi-pathogen infections. To address these challenges, we propose a novel AMP generation method termed CFGAN (Conditional Feedback Generative Adversarial Network), which integrates conditional information with a reinforcement feedback loop. Based on the GAN architecture, CFGAN incorporates two key innovations: (1) conditional information, encoding details about targeted pathogens, minimum inhibitory concentration level, and peptide length, which allows for precise control over antimicrobial specificity against ten representative pathogens, and (2) reinforcement feedback loop, composing a functional analysor and a dynamic reward mechanism that iteratively guides the generator to produce peptide with broad-spectrum targeting and high antimicrobial activity. To improve training stability and convergence speed, we implement a Brownian Motion Controller that dynamically adjusts the learning rates of both the generator and discriminator. Experimental results show that CFGAN outperforms existing AMP generation models in terms of consistency, stability, and antimicrobial properties. Additionally, molecular dynamics simulations confirm that the generated peptides demonstrate strong structural stability and effective membrane-binding capacity. These findings position CFGAN as an efficient framework for designing broad-spectrum AMPs, with significant potential for the development of next-generation anti-infective drugs. The data and Python codes of CFGAN are available at https://github.com/YannanBin/CFGAN6.