Quantitative Structure-Activity Relationships in Carboquinones and Benzodiazepines Using Counter-Propagation Neural Networks

Abstract

Counter-propagation neural networks are used to model and predict activities of carboquinones and of benzodiazepines from physicochemical parameters. For carboquinones, networks with one hidden layer processing element (PE) for each compound achieved significantly better training set RMSE values than corresponding back-propagation and multiregression results and test set RMSE values as good or slightly worse than back-propagation. Test set results improved by 10-15% using networks with fewer hidden layer PEs than carboquinones; the smallest test set RMSE values are between 0 and 10% better than back-propagation values, about 1.3 times greater than corresponding training set values, and occur when there are about as many competitive layer PEs as there are compounds in the data set. Training set RMSE values increase with decreasing number of competitive layer PEs and approach those of test sets. Both counter-propagation and back-propagation networks, however, have worse predictive capability than multiregression. For benzodiazepines, networks with one hidden layer PE for each compound achieved significantly better training set RMSE values than back-propagation and multiregression results and test set RMSE values slightly worse than back-propagation. Test set results improved by 10-15% using fewer hidden layer PEs than benzodiazepines; the smallest test set RMSE values are 0-10% better than back-propagation values, about 1.3 times greater than training set values, and occur when there are about half as many competitive layer PEs as there are compounds in the data set. Training set RMSE values increase with decreasing number of competitive layer PEs and approach those of test sets. Counter-propagation, back-propagation, and multiregression all have similar predictive capabilities.

Multivariate Regression Outperforms Several Robust Architectures of Neural Networks in QSAR Modeling

“…The author stated that 296 weights were optimized in the counter-propagation NNs. 35 From the comparison of these counter-propagation NN models with back-propagation NN 34 and MR models, 33 Peterson concluded that all have similar predictive capabilities. 35 (3) One NN model for the carboquinone data set from the paper of Tetko et al 36 was obtained with a feed-forward NN trained by the back-propagation algorithm.…”

“…32 In this report we will analyze the usefulness of the multivariate regression (MR) method in modeling the antileukemic activity of carboquinones against lymphoid leukemia in mice 33,34 and the tranquilizer activity of benzodiazepine derivatives. 34 The aim of this paper is (i) to present a new effective way for selecting the best possible descriptors in MR models, applicable in the cases when it is needed to select one or several models with six descriptors from a set of 100 descriptors; (ii) to present a new way for a stepwise selection of descriptors in the next-step model by addition of one, two, three, ... i new descriptors to all descriptors selected up to the previous-step model, where i is limited by the total number of descriptors and by the computer resources at one's disposal; (iii) to compare the performance of these MR models with other QSAR models such as earlier MR models obtained using approximate MR procedures, 33 models obtained using NN algorithms, [34][35][36] and the FUN-CLINK method. 37…”

Quantitative Structure−Activity Relationship of Flavonoid p56lck Protein Tyrosine Kinase Inhibitors. A Neural Network Approach

“…Although CP-ANNs have proved to be efficient in modeling response surfaces in various fields, , they have thus far not been extensively used in QSAR . The implementation of this approach in our case study of QSAR in a large bioactive flavonoids data base confirms that such methodology could provide one with a viable and superior alternative to the error back propagation and standard Kohonen algorithms.…”

Neural networks as a method for elucidating structure–property relationships for organic compounds

“…The salient advantages of networks trained by counterpropagation include a relatively small (of the order of several hundreds) number of interactions required for the training of networks 35,36 and the possibility of finding a global minimum of the error function for any starting setup of the weighting coefficients. 37,38 One disadvantage is a weaker approximating capacity of such networks in comparison with other architectures.…”

“…Today, simulation of biological activities of organic compounds presents considerable practical interest: the overwhelming majority of publications devoted to the use of artificial neural networks for elucidating structure ± property relationships are related to this particular area (Table 4). 31 Some publications describe the results of successful quantitative neural network simulation of affinities of organic compounds to various receptors 20,145,146 and transport proteins, 29,30,99 simulation of inhibition constants of enzymes, 20,35,55,74,75,80,83,145,147 antitumour, 20,38,43,105,150 carcinogenic, 36,76,152,153 mutagenic 69, 145, 154 ± 156 and antibacterial activities, 159 etc.…”