A network finds a new antibiotic
On 20 February 2020 Cell published MIT work in which a network trained on 2,335 molecules to inhibit E. coli growth found halicin among known compounds: structurally unlike antibiotics, it killed a wide range of pathogens and cured pan-resistant Acinetobacter baumannii and C. difficile infections in mice.
Why it matters
The antibiotic candidate was chosen by a model and confirmed by the laboratory on bacteria and mice. The same model screened more than 107 million molecules of the ZINC15 database.
The model is directed message passing on the molecular graph. Halicin, formerly known as SU3327, came from the Drug Repurposing Hub; it was active against Mycobacterium tuberculosis and carbapenem-resistant Enterobacteriaceae among others. Of 23 tested predictions from more than 107 million molecules in the ZINC15 database, eight were antibacterial. The record claims no clinical trials of halicin.