A PLUM Job: Peptide modeLs for Understanding and engineering antiMicrobial therapeutics

Abstract

Motivation: Antibiotic-resistant infections in humans and animals are rising, creating an urgent need for new antimicrobial strategies. This challenge extends from human health to food production, where foodborne pathogens cause substantial animal and human illness annually. Antimicrobial peptides (AMPs) are promising alternatives because of their broad activity and potentially lower resistance risk. However, rational AMP design remains challenging, particularly when generating targeted peptides with desired function and length for downstream experimental testing. Results: We introduce Peptide modeLs for Understanding and engineering antiMicrobial therapeutics (PLUM), a structured conditional variational autoencoder for controlled AMP generation. PLUM organizes the latent space into components associated with function, length, and residual sequence context, enabling de novo and prototype-guided generation across the 5-35 amino acid range. In de novo generation, PLUM achieved the highest classifier-confirmed yields for both AMP and non-AMP targets. PLUM-generated AMPs were also substantially distinct from the training set and exhibited broad internal sequence diversity. They further showed favorable predicted safety, stability, and potency profiles, including low toxicity, the lowest median instability index among the evaluated models, and a large number of candidates predicted to have high antibacterial potency. In prototype-guided generation, PLUM yielded more AMP-classified variants than HydrAMP while also supporting non-AMP-targeted generation from existing prototypes. Together, these results support PLUM as a scalable framework for function- and length-controlled AMP design. Availability: https://github.com/priyamayur/PLUM