Deep-learning-enabled antibiotic discovery through molecular de-extinction
Problem
Antibiotic resistance requires new antimicrobial candidates; the work mines proteomes of extinct organisms for antibiotic peptides.
Why it matters
Antibiotic-resistant pathogens are a major global health threat needing new drug leads.
What the AI did
Ensembles of deep-learning models (peptide-sequence encoder plus neural networks) predicted antimicrobial activity and mined over 10 million peptides; 69 predicted peptides were synthesized and experimentally confirmed active against bacterial pathogens.
Results
Mined 10,311,899 peptides; predicted 37,176 broad-spectrum sequences (11,035 not in extant organisms); 69 synthesized peptides experimentally confirmed active
Impact
Cited 189 times; produced experimentally confirmed antibiotic peptide candidates, including named lead compounds.
Record
| Venue | Nature Biomedical Engineering |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Fang Wan, Marcelo D. T. Torres, Jacqueline J Peng, César de la Fuente‐Núñez |
| Citations | 189 (OpenAlex, at import time) |
| References | 60 |
| Model or method | deep-learning ensembles with peptide-sequence encoder (unnamed) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41551-024-01201-x |
| Open access | hybrid |
| Keywords | antibiotic discovery, deep learning, antimicrobial peptides, molecular de-extinction, drug resistance |
Importance rating
major — millions of lives, an industry or a field
Abstract frames work as addressing antibiotic resistance, a major global health threat, with experimentally confirmed active peptides and lead compounds validated in mice; no mortality scale stated.