A Generative Approach toward Precision Antimicrobial Peptide Design

Abstract

Antimicrobial peptides (AMPs) are peptides with promising applications for healthcare, veterinary, and agriculture industries. Despite prior success in AMP design using physics- or knowledge-based approaches, there is still a critical need to create new methodologies to design peptides with a low false positive rate and high AMP activity and selectivity. Toward this goal, we invented a cost-effective approach which utilizes a generative model to produce AMP-like sequences and molecular simulations to select peptides based on their structures and interactions. For a proof of concept, we curated a dataset that comprises 500,000 non-AMP peptide sequences and 8,000 labeled AMP sequences to train the generative model, which generated novel and diverse AMP candidates to potentially target a wide variety of microbes. Following a screening process to select peptides that are cationic and likely helical, we assessed 12 candidates by simulating their membrane-binding tendency to a lipid bilayer model. With the umbrella sampling technique, we determined the free energy change during transfer from the solution to the membrane environments for each peptide. Accordingly, we selected the six peptides with the best membrane-binding tendency, synthesized them, and characterized through spectroscopies and biological assays. Three novel peptides were validated with activity to inhibit bacterial growth. In aggregate, the combination of AMP generator and molecular simulations afford an enhanced accuracy in AMP design. Towards future precision AMP design, our methodology and results demonstrate the viability to design novel AMP-like peptides to target selected pathogens and mechanisms.

AMPGAN v2: Machine Learning Guided Design of Antimicrobial Peptides

“…Furthermore, little validation was done to ensure that AMPGAN v2 is responsive to manipulations of the target microbe and target mechanism conditioning elements, or that the generated sequences feature those attributes. A previous version of AMPGAN v2 was able to generate novel AMPs, 1 but that still leaves the target microbe and target mechanism conditioning elements fairly untested.…”

“…AMPGAN v2 builds on our previous experience with AMPGAN v1, 54 though there are several differences in the implementation and evaluation procedure that make direct comparison of the two difficult. Full implementation details for AMPGAN v2 can be found in our GitLab repository.…”

“…To drive the development of AMPbased treatments it is necessary to create design approaches with higher precision and selectivity towards resistant targets. In this paper we present AMPGAN v2, a generative adversarial network (GAN) based approach for rational AMP design. Like AMP-GAN, 1 AMPGAN v2 combines data driven priors and controlled generation. These elements allow for the generation of AMP candidates tailored for specific applications.…”

Deep generative models for peptide design

“…Through experimental validation, the authors further showed that PepGAN was able to generate an AMP twice as strong as the conventional antibiotic ampicillin. In addition, AMPGAN 39 and the follow-up model AMPGAN v2 40 adopted a bidirectional CGAN framework to generate peptides with desired targets, mechanisms, and minimum inhibitory concentration (MIC) values. By introducing an encoder neural network to map peptide sequence to latent representations, the learnt latent space can be more structural and consequently leads to improved data modeling.…”

AMPGAN v2: Machine Learning-Guided Design of Antimicrobial Peptides

“…Recently, we provided a proof of concept for such an application with AMPGAN and tested its ability to design antibacterial peptides. 54 For 12 generated peptides that are cationic and likely helical, we assessed the membrane-binding propensity via extensive molecular simulations. The top six peptides were promoted for synthesis, chemical characterizations, and antibacterial assays.…”