Discovering highly potent antimicrobial peptides with deep generative model HydrAMP

Area of Science:

• Computational biology
• Medicinal chemistry
• Drug discovery

Background:

• Antimicrobial resistance (AMR) poses a significant global health threat.
• Novel computational methods are needed to accelerate the discovery of new antimicrobial peptides (AMPs).

Purpose of the Study:

• To introduce HydrAMP, a conditional variational autoencoder for generating antimicrobial peptides.
• To optimize peptide generation for diverse applications, including analogue and unconstrained design.

Main Methods:

• Utilizing a conditional variational autoencoder (HydrAMP) to learn continuous peptide representations.
• Implementing parameter-controlled creativity and disentangling peptide properties from antimicrobial conditions.
• Employing a preselection strategy with peptide ranking and molecular dynamics simulations for experimental validation.

Main Results:

• HydrAMP outperforms existing methods in unconstrained and analogue peptide generation.
• Experimental validation confirmed high antimicrobial activity for nine analogues of clinically relevant prototypes and six analogues of an inactive peptide.
• Generated peptides demonstrated potent activity against five bacterial strains.

Conclusions:

• HydrAMP effectively generates diverse and potent antimicrobial peptides.
• This computational approach represents a significant step towards addressing the antimicrobial resistance crisis.
• HydrAMP facilitates the design of novel peptide therapeutics for combating bacterial infections.