Using artificial intelligence to discover new treatments for superbugs
Antimicrobial resistance is an emerging threat to healthcare systems worldwide. As a consequence of the spread of drug-resistant bacteria, also called “superbugs,” medical treatments could become ineffective for an increasing number of people in the next years. To fix this huge problem, chemists are asked to find new effective antibiotics.

Drug discovery is an expensive and time-consuming process during which pharmaceutical chemists look for new candidate molecules to interact with a particular target protein or pathway causing the disease. Chemists screen large libraries of thousands to millions of molecules, looking for compounds with specific biological effects and low toxicity. However, these screenings are not very efficient: if chemical libraries don’t include molecules with enough structural diversity, chemists will fail to discover antibiotics with molecular structures different from the ones already tested in laboratories or clinical trials.
Now machine learning is flanking chemoinformatics through innovative deep neural network approaches to find new drugs. An example of how this approach works can be seen in a recent study by James Collins and coworkers at MIT. First, researchers trained a neural network model to predict growth inhibition of Escherichia coli using a set of 2335 diverse molecules; then, they applied the optimized neural network model to screen large chemical libraries with more than 107 million molecules.
They ended up with a list of candidate molecules structurally different from known antibiotics, and ranked them based on their predicted biological activity. Among those candidates, they found that halicin, a compound under investigation as a treatment for diabetes, displayed high efficacy against E. coli and a large spectrum of pathogens such as Acinetobacter baumanii, at the top list of resistant bacteria which urgently requires new antibiotics.
Research groups are currently developing similar deep learning approaches to find new compounds that could fight the COVID-19 virus. This suggests how recent improvements in machine learning can assist chemists’ work to speed up and lower the costs of the drug discovery process.
This story originally appeared on Massive Science, an editorial partner site that publishes science stories by scientists. Subscribe to their newsletter to get even more science sent straight to you.

Enjoy reading ASBMB Today?
Become a member to receive the print edition four times a year and the digital edition monthly.
Learn moreGet the latest from ASBMB Today
Enter your email address, and we’ll send you a weekly email with recent articles, interviews and more.
Latest in Science
Science highlights or most popular articles

Influenza gets help from gum disease bacteria
Scientists discover that a protease from Porphyromonas gingivalis enhances viral spread. Read more about this recent Journal of Biological Chemistry paper.

How bacteria fight back against promising antimicrobial peptide
Researchers find a mutation in E. coli that reduces its susceptibility to a potential novel antibiotic. Read more about this recent Journal of Biological Chemistry paper.

New clues reveal how cells respond to stress
Redox signaling protein may help regulate inflammasome and innate immune activation. Read more about this recent Journal of Biological Chemistry paper.

Innovative platform empowers scientists to transform venoms into therapeutics
Scientists combine phage display and a “metavenome” library to discover new drugs that bind clinically relevant human cell receptors. Read about this recent Molecular & Cellular Proteomics paper.

Meet Shannon Reilly
The JLR junior associate editor discusses the role of adipocytes in obesity at Weill Cornell Medical School.

Meet Donita Brady
Donita Brady is an associate professor of cancer biology and an associate editor of the Journal of Biological Chemistry, who studies metalloallostery in cancer.