Buffio.

A field journal of machine-made science

← All discoveries

Medicine · 20 February 2020

An antibiotic found by screening a hundred million molecules in silico

A deep neural network identified halicin, a molecule structurally unlike conventional antibiotics, with bactericidal activity against a wide range of pathogens, including in mouse models of C. difficile and pan-resistant A. baumannii infection.

Field
Medicine
Date
20 February 2020
AI system
Deep neural network for antibacterial prediction
Institution
MIT, McMaster University, Broad Institute and Harvard
Published in
Cell
Identifier
10.1016/j.cell.2020.01.021

Jonathan Stokes, James Collins, Regina Barzilay and colleagues trained a deep neural network to predict which molecules would inhibit the growth of Escherichia coli, using a training set of 2,335 molecules. They then applied the model to large chemical libraries, including the Drug Repurposing Hub and more than 107 million molecules from ZINC15.

From the Drug Repurposing Hub the model identified SU3327, a compound previously investigated in other contexts, which the authors renamed halicin. Halicin is structurally divergent from conventional antibiotics and showed bactericidal activity against a wide phylogenetic spectrum of pathogens, including Mycobacterium tuberculosis and carbapenem-resistant Enterobacteriaceae. In mouse models, it treated Clostridioides difficile and pan-resistant Acinetobacter baumannii infections.

From 23 empirically tested predictions drawn from ZINC15, the model also identified eight additional antibacterial compounds structurally distant from known antibiotics. Halicin is the headline discovery: a working antibacterial, nominated by a learned model from a library far too large to screen by hand, then confirmed in the laboratory and in animals.