Machine learning model for antibiotic discovery
Updated on 20 October 2020
Researchers at the University of Dundee have developed a computational protocol that predicts the likelihood for any given compound to act as a broad-spectrum antibiotic against both Gram-positive and Gram-negative bacteria.
- Gram-negative antibiotic discovery
- Novel curation process that addresses cell wall permeation
- Rapid virtual library screening
- Identification of molecular features that drive permeation
- Reduce time, cost and failure rates
- Protocol could extend to antivirals
There is an urgent need for new drug candidates to combat antibiotic resistance. The majority of totally or extremely drug-resistant priority pathogens, as defined by the WHO, are Gram-negative. It is widely understood that there is insufficient drug permeation into Gram-negative organisms due to their complex and poorly permeable cell envelope.
Dundee researchers have developed an AI technology based on data mining, chemoinformatics and machine learning. The technology has been trained on compounds curated to address the Gram-negative cell wall obstacle. The algorithms are designed to detect and predict broad spectrum antibacterial activity.
The technology especially targets novel chemical space by removing known antibiotic compounds and similar structures from the training datasets. In recent tests of the technology, a virtual screening of compound databases has identified about 1,000 new molecules, from around 3M readily synthesisable molecules, which score high on probability for broad spectrum activity. The major molecular features that contribute to the scores are also identified. The method has the potential to be applied to antivirals and other areas of unmet clinical need by varying the training datasets.
This exciting technology is able to pre-screen proprietary and public compound libraries to increase the success rate of anti-infectious drug discovery programmes, thereby reducing time, cost and attrition rates. The University is seeking a development partner for this methodology and contact is welcomed from organisations interested in collaboration for this opportunity.
Research and Innovation Services
+44 (0)1382 firstname.lastname@example.org