A comparative study using different topological representations in pattern recognition based drug activity characterization

The use of certain machine learning and pattern recognition tools for automated pharmacological drug design has been recently introduced. Different families of learning algorithms have been applied to the task of associating observed chemical properties and pharmacological activities to certain kinds of representations of the candidate compounds. In this work, several families of molecular descriptors are considered in order to establish the appropriateness of these families for a particularly challenging drug design task consisting of characterizing the analgesic properties of a relatively large number of compounds. As a second goal, the composite use of descriptors from different families and a first attempt to select the best attributes from these families is considered. As a conclusion, relatively good discrimination results can be obtained by combining the best descriptors of the different families considered.