In the validation of biodegradation (Q)SAR, the prediction failure often occurs for the chemical substances containing the partial structures for possible abiotic hydrolysis reactions. Actually in the biodegradation test under the Chemical Substance Control Law (OECD 301C), it is sometimes observed the abiotic hydorolysis in the test. It is necessary to pursue the quantitative prediction of hydrolysis for the improvement of accuracy for the biodegradation prediction. We first picked up the low molecular weight ester compounds and physicochemical parameters such as activataion energy are calculated by the 3D reaction analysis utilizing MOPAC. By comparing the calculated activation energy values with the experimental values, which are obtained by the group of the Yamaguchi University in the collaboration with us, the reliabilities of the calculation results are discussed. We also try to formulate the prediction equation for hydrolysis reaction velocity coefficient on the basis of Arrenius equation and made its validation. As descriptors in the prediction equation, we picked up the parameters in relation to frequency factor in the Arrenius equation such as molecular size, molecular weight and partial charge of reaction site, in addition to the activation energy.

View full abstract

DNA microarray data analysis is one of the core technologies which are extensively utilized in the field of genome science. Since the dimensionality and the number of such data set are so large that the knowledge discovery from the data is quite difficult, recently, developing and studying the methods for analyzing such data set have been being active. In the case of the multivariate data analyses of huge data sets such as DNA microarray ones, it is necessary to extract the feature using noise reduction as well as variable selection. In addition, non-linear methods are sometimes required in order to analyze complicated relations among the variables. In the previous study, we suggested the fingerprint verification type Self-Organizing Map (FvSOM), of which the reference vectors of all grid points on the resulted SOM map are included for verifications. In this study, we suggest the weighted FvSOM, which is an advanced version of the FvSOM. We applied nonlinear multivariate classification procedures including weighted FvSOM to the DNA microarray data set which was obtained from the tissues of the embryonic tumors of the central nervous system. Comparing their classification results, the results of our advanced procedure were satisfactory.

View full abstract

The Ordinary gene finding method by the Markov model utilizes gene data identified experimentally. Escherichia coli are adopted to build parameters of the Markov model. For genes which belong to other families than E.coli or a newly found ones such as nanoarchaeum equitans, the ordinary Markov model is of no use. An automatic model building method is needed for gene findings for every kinds of genomes. Audic and Claverie(1998) have presented a new method to predict coding regions in prokaryote genomes. In genomes there are six kinds of stochastically different regions. They are three direct coding classes based on reading frames, three reverse (shadow) coding classes. The Markov model with six kinds of transition probabilities is adopted to find ORFs. Starting with randomly selected sequences, repeated improvements of transition probabilities of the Markov model could discriminate two kinds of coding regions. After trimming those gene candidates to ORFs, 70% of genes were found for E.coli. 80% ORFs with a stop codon and without a start codon were found. These ORFs should be extended to get true genes. For this purpose the signal sequence, ribosome binding site which lies between -15 and -7 upstream from a starting codon, could be utilized.

View full abstract

Artificial neural networks (ANN) have widely been used in chemical analysis of mixtures. In this work, ANN is applied to determining concentrations of tartrate diastereomers (i.e. meso and racemic) which are very similar in structure to each other, and so it is quite difficult to do that. In this case, absorption spectra appear in the similar magnitude. Therefore, we used sensor molecules (i.e. cobalt(III) complexes) which show charge-transfer bands owing to ion-pair formation with tartrate anions. Charge transfer absorptions upon addition of the meso isomer are somewhat different from the racemic isomer but are not extremely stereospecific. Multilayer neural networks were chosen for pattern-recognition analysis. The architecture consists of input, output, and hidden layers, which are where data are processed through a back propagation training algorithm. Of the 16 spectra obtained, 15 of these were used to train the data set (charge-transfer absorbance) by using 20 wavelengths from each spectrum. When the remaining one unknown is processed, an absolute error of a few % was obtained for the output concentrations for both meso and racemic tartrates.

View full abstract

Fast and accurate calculation of the predicted results of Chemical Vapor Depositions (CVD) is very helpful to the R&D of the CVD processes. Therefore, we have developed a novel calculation method, by which accurate calculations along with reduced computing cost were achieved, to reproduce deposition profiles in a macroscopic cavity (macrocavity). Boundary value problems for estimating diffusion-reaction equations by iterations of numerical integrations were changed into problems of synthesizing a few basis functions and then finding the linear combinations consisted of the functions. We optimized the coefficients of the linear combinations by Genetic Algorithms (GA). We introduced the theoretical details of the calculation method. In additions, we could demonstrate the validity of the proposed method using an example of the reaction mechanism and conditions.

View full abstract

For the sake of realizing the easy undastandable discriminant flow chart on plastics in the field of the near infrared spectral region, we combined the decision tree (CART, classification and regression trees) and the genetic algorithm. CART has advantages because it is able to clarify the conditions for the discriminant analysis. However, it is not easy to have one decision tree, that should separate a lot of plastic species. That is why we tried the small decision tree, which distinguish a A groupe and non A groupe, and the optimized arranging the orders of small trees by the genetic algorithm. In this case, we tried self chromosome cross for the application of the concept, chromosome cross. This is the exchange of the parts of its own chromosome. As a result, we got better results than the so-called random search.

View full abstract

I developed an online data book for studying higher order structure of biomolecule. With the use of Web browser and a Java molecular viewer Jmol, a user can study biomolecular data registered in the RCSB Protein Data Bank. These biomolecular data are categorized as transmembrane proteins, glycoproteins, and others in the data book. The various structural features of molecules, such as hydropathy index of amino acid residue, can be figured by checking a radio button or a check box.

View full abstract

We have developed an user interface system for high speed gene network simulator named starpack. Starpack simulates chemical reactions and gene networks in high speed, typically between ten thousand and one million times faster than software based simulators. The user interface system consists of (1) Substance list widow, (2) DNA microarray window, (3) Network visualization window, and (4) time-course window. The system is implemented on a pen-display to provide direct usability. The substance list window shows the substances under simulation, and simulation progress can be verified in three other windows. The DNA microarray window shows the quantities of substances using same color mapping as the actual DNA microarray, between green and red. The network visualization window shows genes as nodes and draws links among genes that are functionally related or participate in same function. Nodes are colored with same color used in DNA microarray window, and links show reaction speed by flying dots on edges. The time-course window show the quantity variation of selected genes on a quantity vs time graph. Evaluation indicates that the network visualization window is particularly valuable, and the pen display is better than conventional display and mouse system.

View full abstract