Modelling the Pharmaceutical Action of Drug Mixtures

Medication that uses a mixture of pharmaceuticals often has an effect that differs from the simple sum of the constituent components, and this can be the case both for desired effects (efficacy) of the medication, as well as undesired side-effects. This project aims to build on previous research in the Bender group, which has established modelling approaches for studying drug action, that can be based either on features of molecular structure/properties, and/or features from biological studies, such as from transcriptomics data. Given the availability of large-scale clinical screening data, in particular in cancer and antibiotic treatments, this project will aim to make the step from qualitative modelling of compound combinations to quantitative modelling, i.e. to predicting the quantitative effects of using drug combinations. Data for this purpose will be taken both from public sources as well as company-internal sources (via the company sponsor of the project, Unilever). Various chemical and biological descriptors, as well as data-mining algorithms (such as Random Forests, Support Vector Machines, and Deep Learning) will be employed and evaluated as to their ability to quantitatively predict the net action of drug mixtures. The goal of the project is to arrive at a validated framework for quantitative prediction, which will be validated against clinical data in areas of cancer and antibacterial bioactivities, but with the aim of being more generally applicable.