Drug design by machine learning: Modelling drug activity

This paper describes an approach to modelling drug activity using machine learning tools. Some experiments in modelling the quantitative structure-activity relationship (QSAR) using a standard, Hansch, method and a machine learning system Golem were already reported in the literature. The paper describes the results of applying two other machine learning systems, Magnus Assistant and Retis, on the same data. The results achieved by the machine learning systems, are better then the results of the Hansch method; therefore, machine learning tools can be considered as very promising for solving that kind of problems. The given results also illustrate the variations of performance of the different machine learning systems applied to this drug design problem.