Hybrid reduced graph for SAR studies.

R Carrasco-Velar, J O Prieto-Entenza, A Antelo-Collado, J A Padrón-García, G Cerruela-García, Á L Maceo-Pixa, R Alcolea-Núñez, L G Silva-Rojas

Universidad de las Ciencias Informáticas, Havana, Cuba. rcarrasco@uci.cu

SAR and QSAR in environmental research 2013

filter terms:

The purpose of this work is the definition and evaluation of both atomic and local new hybrid indices. Inspired by the Refractotopological State Index for Atoms, the new atomic indices are theoretically supported by graph theory principles. The local indices, named Descriptor Centres (DCs), are obtained from the sum of the atomic values of the atoms in the selected group. Different classifiers were used for structure-activity relationship (SAR) studies, including multilayer perceptron (MLP), support vector machines (SVM) and meta-classifiers. Prediction with SVM and MLP was around 60%, but the best result was obtained with the meta-classifiers, bagging, decorate and others, with more than 92% accurate prediction. These new hybrid descriptors derived from the Refractotopological State Index for Atoms show a low mutual correlation coefficient. The same behaviour is found in the analogously defined Descriptors Centres. The best results are obtained with the inclusion of the distance between DCs with the use of meta-classifiers.

Citation

R Carrasco-Velar, J O Prieto-Entenza, A Antelo-Collado, J A Padrón-García, G Cerruela-García, Á L Maceo-Pixa, R Alcolea-Núñez, L G Silva-Rojas. Hybrid reduced graph for SAR studies. SAR and QSAR in environmental research. 2013;24(3):201-14