A generative deep learning approach to de novo antibiotic design

Area of Science:

• Drug discovery and development
• Artificial intelligence in medicine
• Antimicrobial resistance

Background:

• The growing antimicrobial resistance crisis demands novel antibiotics with distinct structures.
• Current deep learning methods for identifying antibacterial compounds often lack structural novelty.
• There is a critical need for innovative approaches to antibiotic design.

Purpose of the Study:

• To develop a generative artificial intelligence (AI) framework for designing de novo (from scratch) antibiotics.
• To explore novel chemical space for potential antibacterial agents.
• To address the limitations of existing deep learning approaches in generating structurally unique antibiotics.

Main Methods:

• Employed a fragment-based AI method to screen over 10^7 chemical fragments in silico against Neisseria gonorrhoeae and Staphylococcus aureus.
• Utilized an unconstrained de novo compound generation approach.
• Integrated genetic algorithms and variational autoencoders within the generative AI framework.
• Synthesized 24 compounds designed by the AI framework.

Main Results:

• Seven of the 24 synthesized compounds exhibited selective antibacterial activity.
• Two lead compounds demonstrated bactericidal efficacy against multidrug-resistant bacterial isolates.
• These lead compounds showed distinct mechanisms of action and reduced bacterial burden in vivo mouse models for Neisseria gonorrhoeae and methicillin-resistant Staphylococcus aureus infections.
• Structural analogs of both compound classes were validated as antibacterial.

Conclusions:

• The developed generative AI framework successfully designs de novo antibiotics.
• This AI-driven approach provides a powerful platform for exploring uncharted chemical space for novel therapeutics.
• The findings offer a promising strategy to combat the antimicrobial resistance crisis by generating structurally distinct antibiotics.