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

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Rising antibiotic resistance necessitates novel antimicrobial strategies beyond clinical settings, impacting food safety.
• Antimicrobial peptides (AMPs) show promise as alternatives due to broad activity and reduced resistance risk.
• Rational design of AMPs is complex, particularly controlling sequence, function, and length simultaneously.

Purpose of the Study:

• To introduce PLUM (Peptide modeLs for Understanding and engineering antiMicrobial therapeutics), a structured conditional Variational Autoencoder.
• To enable controlled *de novo* and prototype-conditioned generation of antimicrobial peptides (AMPs) with lengths from 5-35 amino acids.
• To disentangle sequence, function, and length in the latent space for precise AMP engineering.

Main Methods:

• Development of PLUM, a structured conditional Variational Autoencoder for AMP generation.
• Generation and evaluation of 45,000 peptides using PLUM.
• Integration of AMP classifiers for robust evaluation of identity and potency.

Main Results:

• PLUM achieved higher AMP yield (0.885, 7% higher than HydrAMP) and increased AMP diversity (14% higher than HydrAMP).
• PLUM maintained a high non-AMP sequence yield (0.895, 19% higher than HydrAMP) and produced 37% more AMPs than HydrAMP in prototype-conditioned generation.
• Generated sequences exhibited low predicted toxicity and closely matched real peptide compositions.

Conclusions:

• PLUM is a scalable and versatile platform for designing antimicrobial peptides (AMPs).
• The model facilitates the engineering of next-generation therapeutics to combat antibiotic resistance.
• PLUM's ability to control peptide characteristics offers a significant advancement in drug discovery.