Machine Learning to Predict Drug Penetration into Gram-negative Bacteria

An alarming rise in pathogens that show antibiotic resistance has been observed
over recent years. In the case of Gram-negative bacterial pathogens, the resistance crisis has started to go out of control. Due to the lower permeability of the Gramnegative cell envelope for antibiotics, these pathogens are inherently more difficult to treat. The lower cell penetration of new drug candidates is also reflected in the failure of medicinal chemistry to advance novel classes of compounds with Gram-negative activity. This substantially increases the costs required for Gram-negative drug discovery. Therefore, the development of therapeutics against Gram-negative infections and their treatment faces a number of key bottlenecks, most of which are related to poor drug uptake.

The structural and chemical properties of antibiotics that determine if it can, or
cannot, overcome the cell wall barrier are still insufficiently understood. We
hypothesise that the use of artificial intelligence will be instrumental in helping to
distinguish drug candidates that will be taken up by the bacteria from those that will be held back or expelled from the pathogens through efflux pumps. We will work on data sets containing information on compound d permeation into microbes, chemical structure, and physical-chemical properties. The available data sets will be expanded by recording our own data on antibiotic efficiency, using the novel SLIC technique.

We will then develop and make use of machine-learning approaches to analyse
these data to find out which properties turn a chemical compound into a well permeating drug candidate, and which properties may help in avoiding efflux. The
outcome of this work is anticipated to have a crucial role in the search for new and better antibiotics.