The world-leading Journal of Cheminformatics (impact factor 4.2) recently published the results of a study conducted by Dr Noureddin Sadawi, a Research Fellow at the Department of Computer Science, Brunel University London.
The study aims to improve the understanding of proteins, predict how a potential drug (i.e. a small molecule) would bind to that protein, and thereby treat a disease. Gathering molecule activity information with classic research (including lab experiments) is highly costly and time consuming. Therefore, this research attempts to reduce costs by building robust predictive models that can be used to forecast which molecules are likely to work well with target protein(s).
The technique proposed by Dr Noureddin Sadawi and his collaborators leads to significant improvement in performance levels and yields high accuracy predictive models. It exploits the similarity of drug targets in a multi-task learning (MTL) setting, where existing data about well studied proteins is employed, to improve the quality of predictive models built for proteins with scarce data (small numbers of known activity molecules). This work is done with the main purpose of advancing the understanding of such proteins in order to produce better drugs. The full article can be accessed here.
![mtl-article-image](/computer-science/images/mtl-article-image.png)