Yoshikatsu Miyashita

GA Strategy for Variable Selection in QSAR Studies: GA-Based PLS Analysis of Calcium Channel Antagonists⊥

The GAPLS (GA based PLS) program has been developed for variable selection in QSAR studies. The modified GA was employed to obtain a PLS model with high internal predictivity using a small number of variables. In order to show the performance of GAPLS for variable selection, the program was applied to the inhibitor activity of calcium channel antagonists. As a result, variables largely contributing to the inhibitory activity could be selected, and the structural requirements for the inhibitory activity could be estimated in an effective manner.

Automated odor-sensing system based on plural semiconductor gas sensors and computerized pattern recognition techniques

Abstract Although odor is an inherent characteristic of chemical substances, no methods to measure or identify odor objectively have been available. A computerized method for classification and identification of odor is reported. Odors are detected by using eight semiconductor gas-sensor elements which have different sensing properties for gases; the output from the odor-detecting apparatus for a gas sample is represented by an 8-dimensional vector. The vectors for 30 substances were examined by clustering analysis and four obvious clusters were observed. These clusters corresponded to ethereal, etherealminty, ethereal-pungent and pungent substances.

Chemical pattern recognition and multivariate analysis for QSAR studies

Abstract Chemical pattern recognition (CPR) and quantitative structure-activity relationships (QSAR) studies based on multivariate analysis and chemometric techniques are reviewed. In particular, applications of the SIMCA classification method to structure-taste problems are discussed. Cluster significance analysis (CSA) is compared with modelling powers for feature selection of asymmetric data sets. A concentric hypersphere model is used to predict candidate new sweeteners. Partial least squares (PLS) modelling methods are employed to antiarrhythmic data of phenylpyridines and fungicidal and herbicidal data of thiocarbamates, respectively. The CoMFA approach to 3-dimensional QSAR using PLS modelling is described as well. In practice, QSAR is an important branch of chemometrics and enhances rational drug design and new agent development. The chemometric techniques described in the article not only work well for QSAR but also are very helpful for solving the problems related to analytical characteristics-chemical structure relationships.

Computer-assisted structure/taste studies on sulfamates by pattern recognition methods

Abstract Structure/taste relationships for fifty carbosulfamates were investigated by means of pattern recognition. The SIMCA method was used to classify the compounds into sweet or unsweet categories. As a result of feature selection, four descriptors were shown to be significant in modelling the sweet compounds. A compound predicted to be sweet by the model equation was synthesized and found to be three times sweeter than sucrose.

Synthesis and spectroscopic properties of metal complexes of 5,7-dimethyl[1,2,4]triazolo[1,5-a] pyrimidine. X-ray structure of the cobalt(II) and cadmium(II) complexes

Abstract Metal complexes of 5,7-dimethyl[1,2,4]triazolo[1,5-a]pyrimidine (dmtp), a purine derivative, have been prepared, and the crystal structures of the cobalt(II) and cadmium(II) complexes, [M(dmtp)2(H2O)4](NO3)2 (M = cobalt(II) or cadmium(II)), have been determined by single-crystal X-ray diffraction. The complexes are isostructural with the metal ion, which rides on an inversion centre, and octahedrally coordinated by two dmtp ligands through the usual N(3) site and by four water ligands. A pair of intramolecular hydrogen bonds between the water ligand and N(4) of the base stabilize the structure. A nitrate anion, which is hydrogen-bonded to water ligands, intercalates between the successive dmtp rings. Metal bonding preference to the N(3) site for dmtp is consistent with the electronic structure of the ligand.

Quantitative Structure−Activity Relationships of the Synthetic Substrates for Elastase Enzyme Using Nonlinear Partial Least Squares Regression

Eighty-nine synthetic substrates for elastase enzyme and its activities (log 1/Km, log kcat, and log Kcat/Km) are treated using partial least squares (PLS) and quadratic partial least squares (QPLS). Chemical features of synthetic substrates are described using principal properties (PPs). By using the QPLS method, we obtain the nonlinear model equations for three properties (log 1/Km, log kcat, and log kcat/km) with the correlation coefficient 0.736, 0.918, and 0.868, respectively. Also, the predictive correlation coefficients for these model equations are 0.640, 0.865, and 0.793, respectively. By this study, it becomes clear that the z2 value of the amino acid residue on position A and the size of the side chain for amino acid residues on position B are related to the properties of the synthetic substrates.

Nonlinear partial least squares modeling of phenyl alkylamines with the monoamine oxidase inhibitory activities.

The nonlinear partial least squares (PLS) method is a nonlinear version of PLS. In this approach, a quadratic inner relation is used instead of the linear inner relation in PLS. Since the quadratic inner relation can be extended to a general form, it is said that the nonlinear PLS method has a high potential for nonlinear modelling. However, few applications of the method have been appeared in quantitative structure--activity relationships (QSAR) studies, because the mathematical descriptions underlying the algorithm were not so clear. In this paper, we have carried out the QSAR analysis of four monoamine oxidase (MAO) inhibitory activities using the nonlinear PLS method. The in vitro and in vivo MAO inhibitory activities were analyzed separately. From the PLS loadings, the structural requirements could be estimated and the utility of the nonlinear PLS method was demonstrated.

A structure-biological activity study based on cluster analysis and the nonlinear mapping method of pattern recognition

Abstract Cluster analysis is used in a study of structure—activity relationships of biologically active compounds. A hierarchal clustering technique was applied to 29 typical antibiotics using 27 antibacterial activities. These antibiotics were of various types; penicillins, cephalosporins, aminoglycosides, macrolides, tetracyclines, and peptides. The result was obtained as a branching tree diagram. The technique allowed the antibiotics to be distributed into 6 clusters, each cluster mostly consisting of compounds with a similar structure. Nonlinear mapping was used to display the 27-dimensional data structure of the antibiotics. The nonlinear map was compared with the clusters obtained by cluster analysis.

Artificial neural network studies in quantitative structure-activity relationships of antifungal azoxy compounds

Summary Artificial neural networks (ANN) based on the back-propagation algorithm (BP algorithm) were applied to a quantitative structure-activity relationship (QSAR) study for 30 azoxy compounds with antifungal activity. The ANN model could well explain the variance of the antifungal activity owing to its ability to deal with a nonlinear tendency in the data set. A modified BP algorithm proposed by the authors has provided the ANN model with a more enhanced predictive capability. Finally, a transformation of the final ANN model to a polynomial of original physico-chemical parameters was shown to be useful to elucidate the structural requirements for the antifungal activity.

Quantitative structure-activity relationship study of antiarrhythmic phenylpyridines using multivariate partial least squares modelling

Abstract The partial least squares (PLS) method was applied to a study of antiarrhythmic activity data of phenylpyridines. A two-component PLS model based on three physicochemical parameters successfully separated active from inactive compounds. Since the modelling power of each physicochemical parameter is great enough, these coefficients of the PLS model equation are interpreted as measures of the contribution of physicochemical parameters. From the PLS discriminant function, a more potent antiarrhythmic agent can be produced by increasing the molar refractivity value of a substituent R on the benzene ring, increasing the torsion angle between amide function and the benzene ring and decreasing the proton affinity of an amino group of a phenylpyridine. The PLS model equation corresponds to the modulated-receptor hypothesis proposed by Hille as a mechanism for the antiarrhythmic agent-action for sodium channels. The PLS method has proved to be more predictive and promising in multivariate statistical methods for quantitative structure-activity relationship analysis than a linear learning machine approach.