Consensual classification of drug/nondrug compounds for drug design

Because of the costly, lengthy and laborious drug development process, the aim of the pharmaceutical industry has shifted from traditional trial-and-error process of drug discovery to a structure-based drug design. Although many potential drug candidates may look promising in in vitro studies, they can fail during in vivo stages. Therefore, it is very important to know if the compound is druglike or nondruglike for selection and development of new potential drug candidates. In this thesis, a special consensual approach is proposed for this purpose. The constructed consensual model has a preprocessing unit which consists of transformation of input patterns by random matrices and median filtering to generate independent errors for a single type of classifier, and a postptocessing unit for consensus. Three different combining methods, genetic algorithm, pseudoinverse and equal weight, for postprocessing and four different neural network methods, general regression neural network, adaptive general regression neural network, supervised self organizing map and self organizing global ranking for individual classification were used.Murcia-Soler and Cherkasov data sets which consist of drug and nondrug compounds were used with descriptors calculated with the Molecular Operating Environment tool.This thesis presented that the best results to discriminate between the drug-like and the nondrug-like compounds were obtained with the proposed consensus approach with the genetic algorithm which is used for the first time for combining. On the other hand, the self organizing global ranking algorithm which has been applied for the first time on chemical data set also produced high classification results. The consensus approach presented is also used for the first time with chemical data.In this thesis, on the basis of results obtained from the experiments, new approach is provided for choosing potential drug candidates by using molecular data automatically. Hence, it can be stated that this narrowing considerably reduces the time and effort for the discovery of novel pharmaceuticals.