Virtual High Throughput Screening Using Machine Learning Methods

Progresses in the field of biotechnology permitted the emergence of an effective screening technique, High Throughput Screening (HTS). In a typical HTS campaign, the main objective consists of the identification of active compounds, called hits. We discuss the possibility of using machine learning methods to predict experimental HTS measurements. Such a virtual HTS analysis will be based on the results of real HTS campaigns carried out with similar compound libraries and similar drug targets. In this way, we analyze an experimental HTS assay from McMaster Data Mining and Docking Competition (Elowe et al. 2005) by means of decision trees, neural networks and support vector machines. First, we study separately the molecular and atomic descriptors in order to establish which of them provide a better discrimination. We present and discuss the results provided by machine learning methods in terms of identification of false positive and false negative hits.