Toward an Optimal Procedure for PC-ANN Model Building:  Prediction of the Carcinogenic Activity of a Large Set of Drugs

Abstract

The performances of the three novel QSAR algorithms, principal component-artificial neural network modeling method combining with three factor selection procedures named eigenvalue ranking, correlation ranking, and genetic algorithm (ER-PC-ANN, CR-PC-ANN, PC-GA-ANN, respectively), are compared by application of these model to the prediction of the carcinogenic activity of a large set of drugs (735 drugs) belonging to a diverse type of compounds. A total number of 1350 theoretical descriptors are calculated for each molecule. The matrix of calculated descriptors (with 735 x 1350 dimension) is subjected to PCA. 95% of the variances in the matrix are explained by the first 137 principal components (PC's). From the pool of 137 PC's, the factor selection methods (ER, CR, and GA) are employed to select the best set of PC's for PC-ANN modeling. In the ER-PC-ANN, the PC's are successively entered into the ANN based on their decreasing eigenvalue. In the CR-PC-ANN, the ANN is first employed to model the nonlinear relationship between each one of the PC's and the carcinogen activity separately. Then, the PC's are ranked based on their decreasing correlating ability and entered to the input layer of the network one after another. Finally, a search algorithm (i.e. genetic algorithm) is used to find the best set of PC's. Both the external and cross-validation methods are used to validate the performances of the resulting models. One is able to see that the results obtained by the PC-GA-ANN and CR-PC-ANN procedures are superior to those resulted from the EV-PC-ANN. Comparison of the results reveals that the results produced by the PC-GA-ANN algorithm are better than those produced by CR-PC-ANN. However, the difference is not significant.

Prediction of Melting Point for Drug-like Compounds Using Principal Component-Genetic Algorithm-Artificial Neural Network

“…21 Hence, selecting the significant and informative PC's is the main problem in all of the PCA-based calibration methods. [22][23][24][25] Different methods have been addressed to select the significant PC's for calibration purposes. The simplest and most common one is a top-down variable selection where the PC's are ranked in the order of decreasing eigenvalues and the PC's with highest eigenvalue is considered as the most significant one and, subsequently, the PC's are introduced into the calibration model.…”

“…However, the magnitude of an eigenvalue is not necessarily a measure of its significance for the calibration. 25 In the other method, which is called correlation ranking, the PC's are ranked by their correlation coefficient with the property and selected by the procedure discussed for eigenvalue ranking. 22,23 Better results are often achieved by this method.…”

Prediction Partial Molar Heat Capacity at Infinite Dilution for Aqueous Solutions of Various Polar Aromatic Compounds over a Wide Range of Conditions Using Artificial Neural Networks

“…The theoretical descriptors were reduced by the following methods: 1) descriptors that are constant or nearly constant have been eliminated, because these descriptors can not define the variation of the property with structure; 2) in order to decrease the redundancy existing in the descriptors data matrix, the correlation coefficients for all pairs of remaining descriptors were determined. If a correlation coefficient was higher than 0.91, the descriptor with lower correlation with the heat capacity was eliminated; 44,45 3) the method of stepwise multi-parameter linear regression was used to select the most important descriptors and to calculate the coefficients relating the heat capacity to the descriptors. 15 The MLR models were generated using spss/pc software package release 10.0.…”

Exploring QSARs for Inhibitory Activity of Cyclic Urea and Nonpeptide-Cyclic Cyanoguanidine Derivatives HIV-1 Protease Inhibitors by Artificial Neural Network

“…In the case of each MLR model, a feed-forward neural network with back-propagation of error algorithm was constructed to model the activity-structure relationships between the descriptors on one hand and inhibitory activity on the other hand. The model development in ANN and the network architecture is fully described by us [13] and others [14]. The data set was divided into training and external test sets.…”

“…Both linear and nonlinear mapping functions can be modeled by configuring the network properly. To obtain powerful and accurate ANN models, one should train a subset of descriptors instead of all generated descriptors [10][11][12][13][14][15].…”