Deep Learning Model Discovers Antibiotic Drugs in Extinct Organisms Effective Against Drug-Resistant Superbugs

Background

Escalating antibiotic resistance around the globe has emerged as one of the pressing problems of our time, with antibiotic-resistant infections causing approximately 1.27 million deaths annually worldwide. Without effective new drugs, this number could surge to 10 million fatalities by 2050. Traditional drug discovery approaches have been unable to keep pace with the evolution of antibiotic-resistant bacterial strains, resulting in a critical shortage of novel antibiotics to combat these resistant pathogens. These developments can seriously jeopardize public health in the future.

Technology Overview

University of Pennsylvania researchers have developed a deep learning AI technology that analyzes a protein database of long-extinct organisms to rediscover lost, never-before-used antibiotic compounds, which display considerable efficacy against pathogens resistant to more traditional antibiotics.

In their experimentation, the researchers employed the Antibiotic Peptide de-extinction (APEX) approach, which relies on deep learning techniques (). APEX systematically scanned proteomes of extinct organisms, encompassing various species across different time periods. A multitask deep learning model, trained on both public and in-house peptide sequences from natural and synthetic sources. The peptide sequences generated by APEX range from 8 to 50 residues in length, and predicted antibiotic activity against the most relevant pathogenic strains. Based on predicted antibiotic properties, the highest-ranking peptides underwent extensive in vitro characterization, including assessing antibiotic activity, mechanism of action, secondary structure, synergy, and cytotoxicity. Testing the selected peptides using in vivo mouse models validated and showcased the peptides’ potential antibacterial effectiveness.

Further details:

Wan, F et al. Nat Biomed Eng 2024 June 11

Knapton, S. The Telegraph 2024 June 15; 13h30 BST

Stage of Development

• Proof of Concept

Benefits

• Automates the process of finding new antibiotic peptides essential to counter antibiotic-resistant pathogens.
• The deep learning approach allows large databases of ancient proteomes to be mined for relevant molecules.
• Encrypted peptides discovered reduce bacterial load in a mouse model by 2-3 orders of magnitude.
• The encrypted peptides operate via new pathways that circumvent the drug resistance of superbugs.
• Discovered peptides can be used as general antibiotics in non-specialised applications.