Faculty Opinions recommendation of A deep learning approach to antibiotic discovery.
Problem
Rapidly emerging antibiotic-resistant bacteria create an urgent need for new antibiotics with novel structures.
Why it matters
Antibiotic resistance threatens treatment of tuberculosis and drug-resistant pathogens like carbapenem-resistant Enterobacteriaceae.
What the AI did
A trained deep neural network predicted antibacterial activity across chemical libraries, discovering halicin plus eight other antibacterial compounds structurally unlike known antibiotics.
Results
From 23 empirically tested predictions out of >107 million ZINC15 molecules, 8 antibacterial compounds identified; halicin active against M. tuberculosis and carbapenem-resistant Enterobacteriaceae and treated C. difficile and pan-resistant A. baumannii in murine models.
Impact
Deep learning yielded structurally novel antibiotic leads validated in vitro and in mouse models.
Record
| Venue | Faculty Opinions – Post-Publication Peer Review of the Biomedical Literature |
|---|---|
| Published | 2020 |
| Work type | peer-review |
| Authors | Serge Mostowy, Ana T. López-Jiménez |
| Citations | 0 (OpenAlex, at import time) |
| Model or method | deep neural network |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3410/f.737402248.793574353 |
| Open access | closed |
| Keywords | halicin, antibiotic resistance, deep learning, drug discovery, ZINC15 |
Importance rating
major — millions of lives, an industry or a field
Abstract frames antibiotic resistance as rapidly emerging with growing need for new antibiotics; compounds active against M. tuberculosis, carbapenem-resistant Enterobacteriaceae, pan-resistant A. baumannii; no scale numbers.