Because of the fact that many bacteria have become drug-resistant, there is an ongoing need to discover novel antibiotics. The researchers from the MIT have created a deep neural network model, which predicts the efficiency of various chemical substances, available from the open database: Drug Repurposing Hub, in order to fight pathogenic bacteria.
The process proposed for the purpose of the discovery of novel antibiotics is divided into three phases. First, deep learning network which serves the purpose of predicting of the growth suppression of E. coli, which was trained based on the collection of over 2000 molecules. Next, the resulting model, was used for the identification of potential lead compounds, which disclose anti-bacterial features against E. coli, based upon the data of chemical compound libraries, containing over 107 000 000 compounds. After ordering the outcome molecules against the effect predicted by the model, a small subset of the antibiotic candidate substances was selected, upon the preselected prediction threshold, chemical structure and accessibility.
In such a way, halicin molecule was discovered. Structurally, it differs from conventional antibiotics. During laboratory testing, it was shown that it is a potent drug which can fight E. coli but also other similar and drug-resistant bacteria (malignant bacteria such as Streptococcus pneumoniae, Staphylococcus or Pseudomonas aeruginosa). The experimental treatment of Acinetobacter baumannii infection was performed on mice, where halicin cleared the rodent organism within 24 hours, with very promising results.
More information:
