Deep optimized generation for antimicrobial peptide discovery

Introduction

The DOG-AMP project will develop cutting-edge methods for Deep Optimized Generation (DOG) and use them to transform the emerging field of AntiMicrobial Peptide (AMP) discovery.

Background

Continuous overuse of antibiotics fuels the outgrowth and spread of multi-drug resistant microbial strains. Increasing antimicrobial resistance is already a major health and economic hazard, and is expected to account for 10 million deaths globally per year by 2050, exceeding deaths caused by cancer.

Importance of AMPs

AMPs are short peptides that can actively and selectively kill antibiotic-resistant pathogens, and as such are considered the most promising strategy for fighting antimicrobial resistance. Still, intensive research on AMP did not translate to their success in the clinic, mostly due to their lower activity and safety compared to existing antibiotics.

Objectives of the DOG-AMP Project

Deep optimized generation has the potential to radically advance AMP discovery, but only once unsolved problems are attacked and open research directions in this area are further explored to reach three major objectives of the DOG-AMP project:

1. Develop a novel model, geared for deep optimized generation, combining the variational autoencoder framework with probabilistic modeling and algorithms for Pareto (conflicting multi-target) optimization, dealing with data scarcity and bias, generation diversity, and model interpretability.

2. Combine the deep optimized generation model into a framework tailored for the specific needs of AMP design, e.g., accounting for AMP clustering, or conflicting features that make AMPs active or toxic.

3. Apply the newly developed framework to explore and navigate the space of peptides to select and experimentally validate the best candidates that will supersede existing AMPs and antibiotics in their activity against hazardous microbes and safety.

Potential Impact

DOG-AMP has the potential to bring breakthroughs in the broad research areas of deep generative modeling, sequence optimization, and AMP discovery.