Application of radial basis function networks and successive projections algorithm in a QSAR study of anti‐HIV activity for a large group of HEPT derivatives

Abstract

A series of 1-[2-hydroxyethoxy-methyl]-6-(phenylthio)thymine] (HEPT) derivatives, as nonnucloside reverse transcriptase inhibitors (NNRTIs), was investigated using a nonlinear quantitative structureanti-HIV-1-activity relationship (QSAR) study. Molecular descriptors derived solely from molecular structure were used to represent molecular structure. Utilizing successive projections algorithm (SPA) and a stepwise backward elimination, a subset of 11 descriptors were selected. Application of SPA minimizes the collinearity between the selected descriptors, which are known to be responsible for the anti-HIV-1 activity. Three layer radial basis function networks (RBFNs) were used to construct the nonlinear QSAR models in all stages of study. The relative standard error percent in anti-HIV activity predictions for the training set by the application of cross-validation (RSECV%) was 9.94%, and for prediction set (RSEP%) was 9.99%. The obtained model outperforms those given in the literature in both the fitting and predicting stages. RBFN analysis yielded predicted activities in the excellent agreement with the experimentally obtained values (cross-validation r ¼ 0.927, prediction r ¼ 0.925).

Correlation weighted successive projections algorithm as a novel method for variable selection in QSAR studies: investigation of anti‐HIV activity of HEPT derivatives

“…The utilized data in this study is from our previous study [34] and are in Table I. The data includes the chemical structure and the observed and calculated values of the activity of 107 compounds studied.…”

“…In this way, especially in the case of QSAR/ QSPR studies, selected variables (vectors) may contain a number of redundant and not informative variables. In our previous study [34], a backward elimination procedure was applied to eliminate the useless variables from the obtained set using SPA.…”

“…Similar to our previous study [34], the RBFNs had n i inputs (number of introduced descriptors, which were less than 100), one output unit (the anti-HIV-1 activity) and one hidden layer of n h units (from 8 to 20). Before transformation to the program which was developed in MATLAB, the input values were normalized to 0.0-1.0.…”

A New Variable Selection Method Based on Mutual Information Maximization by Replacing Collinear Variables for Nonlinear Quantitative Structure-Property Relationship Models

“…31 Recently, there is a growing interest in the use of RBFNN for its short training time and being guaranteed to reach the global minimum of error surface during training. 32 In RBFNN, the input layer does not process the information; it only distributes the input vectors to the hidden layer. The hidden layer consists of a number of RBF neurons and a bias (b k ).…”

Improving the computational efficiency of the successive projections algorithm by using a sequential regression implementation: a case study involving nir spectrometric analysis of wheat samples

“…1 In several applications concerning UV-Vis, 1,2 ICP-OES, 3 FT-IR 4 and NIR spectrometry, [4][5][6][7][8] SPA was found to provide models with good predictive performance. It has also been successfully employed in other fields such as QSAR (quantitative structure activity relationships) 9 and classification. 10,11 A graphic user interface for SPA is freely available at <http://www.ele.ita.br/~kawakami/spa>.…”