The search for new drug candidates in databases is of paramount importance in
pharmaceutical chemistry. The selection of molecular subsets is greatly optimized and
much more promising when potential drug-like molecules are detected a priori. In this
work, about one hundred thousand molecules are ranked following a new methodology: a
drug/non-drug classifier constructed by a consensual set of classification trees. The classification
trees arise from the stochastic generation of training sets, which in turn are used to
estimate probability factors of test molecules to be drug-like compounds. Molecules were
represented by Topological Quantum Similarity Indices and their Graph Theoretical counterparts.
The contribution of the present paper consists of presenting an effective ranking
method able to improve the probability of finding drug-like substances by using these
types of molecular descriptors.
Keywords: Drug-like molecules, Recursive partition trees, Topological quantum similarity indices, Topological indices, Consensus model.
Rights & PermissionsPrintExport