Driss Cherqaoui

Use of a neural network to determine the boiling point of alkanes

Back-propagation neural networks (NNs) are useful for the study of quantitative structure–activity relationships or structure–property correlations. Models of relationships between structure and boiling point (bp) of 150 alkanes were constructed by means of a multilayer neural network (NN) using the back-propagation algorithm. The results of our NN were compared with those of other models from the literature, and found to be better. The boiling points of the 150 alkanes were then predicted by removing 15 compounds (test set) and using the 135 other molecules as a training set. Using the same process, all the compounds in the data bank were then predicted in groups of 15 compounds. The results obtained were satisfying.

Support vector machines: development of QSAR models for predicting anti-HIV-1 activity of TIBO derivatives.

The tetrahydroimidazo[4,5,1-jk][1,4]benzodiazepinone (TIBO) derivatives, as non-nucleoside reverse transcriptase inhibitors, acquire a significant place in the treatment of the infections by the HIV. In the present paper, the support vector machines (SVM) are used to develop quantitative relationships between the anti-HIV activity and four molecular descriptors of 82 TIBO derivatives. The results obtained by SVM give good statistical results compared to those given by multiple linear regressions and artificial neural networks. The contribution of each descriptor to structure-activity relationships was evaluated. It indicates the importance of the hydrophobic parameter. The proposed method can be successfully used to predict the anti-HIV of TIBO derivatives with only four molecular descriptors which can be calculated directly from molecular structure alone.

Structure-musk odour relationship studies of tetralin and indan compounds using neural networks

Models of the relationships between structure and musk odour of tetralin and indan compounds were elaborated with a multilayer neural network using the back-propagation algorithm. The neural network was used to classify the compounds studied into two categories (musk or non-musk). The cross-validation procedure was used to assess the predictive power of the network. Each molecule was described by eight global parameters: five steric and three electronic descriptors. The neural networks results were successfully compared to those given by the k-Nearest Neighbours and the Bayesean methods, both in the classification and prediction tests. The contribution of each descriptor to the structure-odour relationships was evaluated. Three out of the eight descriptors were thus found to be the most relevant in the molecular description for the prediction of musk odour. This research points out that neural networks are likely to become a useful technique for structure-odour relationships.

Artificial neural networks: Non-linear QSAR studies of HEPT derivatives as HIV-1 reverse transcriptase inhibitors

Structure-anti HIV activity relationships were established for a sample of 801-[2-hydroxyethoxy-methyl]-6-(phenylthio)thymine(HEPT) using a three-layer neural network (NN). Eight structural descriptors and physicochemical variables were used to characterize the HEPT derivatives under study. The networks architecture and parameters were optimized in order to obtain good results. All the NN architectures were able to establish a satisfactory relationship between the molecular descriptors and the anti-HIV activity. NN proved to give better results than other models in the literature. NN have been shown to be particularly successful in their ability to identify non-linear relationships.

Use of a neural network to determine the normal boiling points of acyclic ethers, peroxides, acetals and their sulfur analogues

Models of relationships between structure and boiling point (bp) of 185 acyclic ethers, peroxides, acetals and their sulfur analogues have been constructed by means of a multilayer neural network (NN) using the back-propagation algorithm. The ability of a neural network to predict the boiling point of acyclic molecules containing polar atoms is outlined. The usefulness of the so-called embedding frequencies for the characterization of chemical structures in quantitative structure–property studies has been shown. NNs proved to give better results than multiple linear regression and other models in the literature.

APPLICATION OF NEURAL NETWORK APPROACH FOR PREDICTION OF SOME THERMOCHEMICAL PROPERTIES OF ALKANES

Abstract A standard feed-forward neural network with one layer of hidden neurons for prediction of boiling points, melting points, critical temperatures, and molar volumes of all alkanes with up to ten carbon atoms has been used. The descriptors (input activities) of alkanes are determined as the embedding frequencies of eight preselected alkane-like substructures. These nonnegative integer entities determine numbers of appearance of the given substructure in an alkane. The obtained results are very encouraging for wider applications of neural networks for prediction, classification and correlation of molecular properties based on their substructure fragments.

APPLICATION OF NEURAL NETWORKS TO STRUCTURE-SANDALWOOD ODOUR RELATIONSHIPS

Neural networks have proved to be particularly successful in their ability to identify non-linear relationships. This paper shows that a three-layer back-propagation neural network is able to learn the relationship between the sandalwood odour and molecular structures of 85 organic compounds belonging to acyclic, cyclohexyl, norbornyl, campholenyl and decalin derivatives. Four steric and three electronic parameters were used to describe each molecular structure. Odour was coded by a binary variable. The neural network was used to classify the compounds into two groups and to predict their odours (sandalwood or non-sandalwood). The results obtained were compared with those given by discriminant analysis, and found to be better. The most important descriptors were revealed on the basis of correlation analysis.

Quantitative structure-activity relationship studies of TIBO derivatives using support vector machines

A quantitative structure-activity relationship (QSAR) study is suggested for the prediction of anti-HIV activity of tetrahydroimidazo[4,5,1-jk][1,4]benzodiazepinone (TIBO) derivatives. The model was produced by using the support vector machine (SVM) technique to develop quantitative relationships between the anti-HIV activity and ten molecular descriptors of 89 TIBO derivatives. The performance and predictive capability of the SVM method were investigated and compared with other techniques such as artificial neural networks and multiple linear regression. The results obtained indicate that the SVM model with the kernel radial basis function can be successfully used to predict the anti-HIV activity of TIBO derivatives with only ten molecular descriptors that can be calculated directly from only molecular structure. The contribution of each descriptor to the structure-activity relationships was evaluated. Hydrophobicity of the molecule was thus found to take the most relevant part in the molecular description.

Neural networks studies : quantitative structure-activity relationships of antifungal 1-[2-(substituted phenyl)allyl]imidazoles and related compounds

Models of relationships between structure and antifungal activity of 1-[2-(substituted phenyl)allyl]imidazoles and related compounds were constructed by means of a multilayer neural network using the back-propagation (BP) algorithm. Each molecule was described by three structural and one physicochemical parameters. The leave-one-out procedure was used to assess the predictive ability of a neural network model. The results obtained were compared to those given in the literature by the multiple linear regression (MLR), and were found to be better. The contribution of each descriptor to the structure-activity relationships was evaluated. Hydrophobicity of the molecule was confirmed to take the most relevant part in the molecular description.

The Hyperconjugation and the Trans Bent Structure of A =A (A =C, Si,..., Pb) Double Bond, from Dimetallene A 2 H 4 to Nanosystems

The trans-bent structure of the A =A (A of 14 th group elements) double bond neighbourhood is reinvestigated at a high level of theory. The hyperconjugation phenomenon is rationalised on going from C to Pb in dimetallene A 2R4 systems. This phenomenon also exhibits conjugated behaviour in larger systems having conjugated double bonds like congeners of polyenes, graphene, and fullerene. Behind this phenomenon is the electronic delocalisation from occupied orbitals σ, π and LP to virtual ones π*, σ* and LP* respectively. In this study we have meticulously analysed the geometrical and electronic structures by performing optimisation at high level of computation like B3LYP/6-311 + G(3df,2p) and B3LYP/LANL2DZ and electronic distribution NBO model.