A. I. Luik

Neural Network Studies. 1. Comparison of Overfitting and Overtraining

The application of feed forward back propagation artificial neural networks with one hidden layer (ANN) to perform the equivalent of multiple linear regression (MLR) has been examined using artificial structured data sets and real literature data. The predictive ability of the networks has been estimated using a training/ test set protocol. The results have shown advantages of ANN over MLR analysis. The ANNs do not require high order terms or indicator variables to establish complex structure-activity relationships. Overfitting does not have any influence on network prediction ability when overtraining is avoided by cross-validation. Application of ANN ensembles has allowed the avoidance of chance correlations and satisfactory predictions of new data have been obtained for a wide range of numbers of neurons in the hidden layer.

Neural Network Studies. 3. Variable Selection in the Cascade-Correlation Learning Architecture

Pruning methods for feed-forward artificial neural networks trained by the cascade-correlation learning algorithm are proposed. The cascade-correlation algorithm starts with a small network and dynamically adds new nodes until the analyzed problem has been solved. This feature of the algorithm removes the requirement to predefine the architecture of the neural network prior to network training. The developed pruning methods are used to estimate the importance of large sets of initial variables for quantitative structure−activity relationship studies and simulated data sets. The calculated results are compared with the performance of fixed-size back-propagation neural networks and multiple regression analysis and are carefully validated using different training/test set protocols, such as leave-one-out and full cross-validation procedures. The results suggest that the pruning methods can be successfully used to optimize the set of variables for the cascade-correlation learning algorithm neural networks. Th...

Variable Selection in the Cascade-Correlation Learning Architecture

Recently there has been a growing interest in the application of neural networks in the field of QSAR. It was demonstrated that this method is often superior to the traditional approaches.1 Other studies have shown that prediction ability of such methods can be substantially improved if the number of input variables for neural networks is optimized.2,3

Evaluation of potential HIV-1 reverse transcriptase inhibitors by artificial neural networks

Artificial neural networks were used to analyze the human immunodeficiency virus type 1 reverse transcriptase inhibitors and to evaluate newly synthesized substances on this basis. The training and control set included 44 molecules (most of them are well-known substances such as AZT, dde, etc.). The activities of molecules were taken from literature. Topological indices were calculated and used as molecular parameters. Four most informative parameters were chosen and applied to predict activities of both new and control molecules. We used a network pruning algorithm and network ensembles to obtain the final classifier. The increasing of neural network generalization of the new data was observed, when using the aforementioned methods. The prognosis of new molecules revealed one molecule as possibly very active. The activity was confirmed by further biological tests.<<ETX>>

Anti AIDS drug design with the help of neural networks

Artificial neural networks were used to analyze and predict the human immunodefiency virus type 1 reverse transcriptase inhibitors. Training and control set included 44 molecules (most of them are well‐known substances such as AZT, TIBO, dde, etc.) The biological activities of molecules were taken from literature and rated for two classes: active and inactive compounds according to their values. We used topological indices as molecular parameters. Four most informative parameters (out of 46) were chosen using cluster analysis and original input parameters’ estimation procedure and were used to predict activities of both control and new (synthesized in our institute) molecules. We applied pruning network algorithm and network ensembles to obtain the final classifier and avoid chance correlation. The increasing of neural network generalization of the data from the control set was observed, when using the aforementioned methods. The prognosis of new molecules revealed one molecule as possibly active. It was ...

Application of Self-Organizing Neural Networks with Active Neurons for QSAR Studies

The interest for development of rational methods for investigation of relations between structure and activity of chemical compounds has essentially increased in the last years. Artificial neural networks have became one of the leading methods in this field.1 However, there are some difficulties (such as limitation in speed, local minima, overfitting/overtraining problems,1 etc.) with an application of these methods for analysis of data sets with a large number of input parameters and, particularly, three-dimensional electronic parameters of compounds generated by 3D QSAR approaches, such as CoMFA. The current study analyses a new method, i.e. neural networks with active neurons,2,3 that could be used in such QSAR studies. We also propose to combine this method with Kohonen’s Self-Organizing Maps (SOM) used for preprocessing of 3D QSAR data sets. The performance of new method is compared with that of fixed size neural networks.

Application of topological indexes to predict immunomodulating activity of new peptides

The application of topological indexes to codify the structures of peptide immunomodulators — analogs of taftsin — and to predict the pharmacological activity of new compounds by the κ-neighbor and neural network methods is described. It is shown that if the conformational state of the model compound is chosen correctly, topological indexes can be used successfully for effective screening in the development of new medicinals with corresponding mechanisms of action.

Study of the structure-activity relationship in a series of physiologically active substances with a common trend of action upon the cell signaling system

In the previous works [ I 4 ] we noted that many substances inhibiting the signal transfer in Ca-mobilizing phospholipid (PL) signaling system, blocking the Ca-channels, and activating the adenylate cyclase (AdC) circuit (primarily by blocking the channel and reducing the regulation level) not only exhibit a number of like manifestations of the pharmacological activity, but have a number of common target points in various sites of the cell signaling system (CSS) [1 4]. Each substance, despite a relative selectivity toward certain receptors, enzymes, or effector proteins, exhibited a somewhat lower (but nevertheless sufficiently high) affinity for interaction with the other CSS elements. At the same time, the compounds conform to strictly determined trends of their integral action upon the CSS function. Examples of this polytropic behavior are given below. The fact that a substance may exhibit the same trend in acting upon a number of biopolymers involved in the signaling circuits led us to the conclusion that various CSS elements have similarly organized and regulated centers that serve as targets for the physiologically active substances. Moreover, taking into account a large number of substances capable ofpolytropic PL blocking and AdC activation, on the one hand, while being antagonists of the Ca-channels, on the other hand, we expected that there are certain features of the chemical structure in common for these compounds (despite a considerable general heterogeneity of this group). This assumption was confirmed by analysis of large (more than 300 members) sample sets of compounds using the methods of artificial intelligence and pattern recognition [4]. The structure of compounds was coded using more than 200 descriptors. However, the results of these investigations have proved only the existence of some structural features in common for the

Modified Hopfinger analysis of the phosphodiesterase inhibiting activity of flavonoids

We have obtained an effective correlation equation describing the structure dependence of the activity of a number of hydroxy- and methoxy-substituted flavonoids as cAMP phosphodiesterase inhibitors. We propose an original modification of the Hopfinger method of pairwise molecular shape descriptors which uses two shape reference molecules. The data obtained may be used to predict the activity of new flavonoid molecules.

Use of topological indexes for prediction of the activity of 5-lipoxygenase inhibitors in a series of hydroxamates

An effective system for predicting the inhibiting activity with respect to 5-lipoxygenase in a series of hydroxamic acids was developed. The bioactivity is predicted based on three selected topological indexes with one variation of the method of k-nearest neighbors. The system obtained is suitable for selecting the structures of new 5-lipoxygenase inhibitors which can be used as drugs.