Wshah, S. | bioRxiv Science

bioRxiv2020

AMPGAN v2: Machine Learning Guided Discovery of Anti-Microbial Peptides

Antibiotic resistance is a critical public health problem. Each year ~2.8 million resistant infections lead to more than 35,000 deaths in the U.S. alone. Antimicrobial peptides (AMPs) show promise in treating resistant infections. However, applications of known AMPs have encountered issues in development, production, and shelf-life. To drive the development of AMP-based treatments it is necessary to create design approaches with higher precision and selectivity towards resistant targets. Previously we developed AMPGAN and obtained proof-of-concept evidence for the generative approach to design AMPs with experimental validation. Building on the success of AMPGAN, we present AMPGAN v2 a bidirectional conditional generative adversarial network (BiCGAN) based approach for rational AMP design. AMPGAN v2 uses generator-discriminator dynamics to learn data driven priors and controls generation using conditioning variables. The bidirectional component, implemented using a learned encoder to map data samples into the latent space of the generator, aids iterative manipulation of candidate peptides. These elements allow AMPGAN v2 to generate of candidates that are novel, diverse, and tailored for specific applications--making it an efficient AMP design tool.

biophysics↗

bioRxiv2020

A Generative Approach toward Precision Antimicrobial Peptide Design

As the emergence of bacterial resistance is outpacing the development of new antibiotics, we must find cost-effective and innovative approaches to discover new antibacterial therapeutics. Antimicrobial peptides (AMPs) represent one promising solution to fill this void, since they generally undergo faster development, display rapid onsets of killing, and most importantly, show lower risks of induced resistance. Despite prior success in AMP design with physics- and/or knowledge-based approaches, an efficient approach to precisely design peptides with high activity and selectivity is still lacking. Toward this goal, we have invented a novel approach which utilizes a generative model to predict AMP-like sequences, followed by molecular modeling to rank the candidates. Thus, we can identify peptides with desirable sequences, structures, and potential specific interactions with bacterial membranes. For the proof of concept, we curated a dataset that comprises 500,000 non-AMP peptide sequences and nearly 8,000 labeled AMP sequences to train the generative model. For 12 generated peptides that are cationic and likely helical, we assessed the membrane binding propensity via extensive all-atom molecular simulations. The top six peptides were promoted for synthesis, chemical characterizations, and antibacterial assays, showing various inhibition to bacterial growth. Three peptides were validated with broad-spectrum antibacterial activity. In aggregate, the combination of AMP generator and sophisticated molecular modeling affords enhanced speed and accuracy in AMP design. Our approach and results demonstrate the viability of a generative approach to develop novel AMPs and to help contain the rise of antibiotic resistant microbes.

biophysics↗