We have developed the total system of molecular design. This system can align molecules by HNN (Hopfield Neural Network), analyze molecules by CoMFA (Comparative Molecular Field Analysis), produce candidate structures of drugs by LigConstructor and etc. In molecular alignment method by HNN, each molecule is represented by four kinds of chemical properties (hydrophobic, hydrogen-bonding donor, hydrogen-bonding acceptor, hydrogen-bonding donor/acceptor) and same kind of properties are corresponded by HNN. Then molecules are aligned based on the correspondences. The CoMFA method is one of typical 3D-QSAR techniques. This method makes a model by using 3D-strcutural features and molecular activities. And it is able to predict molecular activity and find important part of structure. The LigConstructor brings up molecular structures with framework of EC (Evolutionary Computation). And this uses CoMFA model as criterion to select high activity molecules. By these procedures, drug structures with high activity are automatically created. This system has integrated graphical user interface and user can use methods of molecular designs mentioned above on the integrated framework.

View full abstract

The identification of appropriate reaction models is very helpful for developing chemical vapor deposition (CVD) processes. In our previous work, we developed a system, which analyzes experimental data of CVD processes and identifies reaction models automatically using genetic algorithms (GA). Cost for developing and maintaining the system increases, as the scale of the system increases. Therefore we proposed a new structure of the modeling system using software agents, that is, a multi agent system, in order to decrease the cost. In the system, we developed an interface agent, which deliver information of the reaction models to partners (i.e. users and another agents), using J2SE (Java2 platform, Standard Edition). The interface agent delivers the information in the most suitable form to the partners. For examples, the agent sends it in a bit string form to an inference agent, in a figure form to the users. In additions, the agent provides environment for users to create and edit the reaction model figures as the reaction model information using a mouse with windows like drawing software.

View full abstract

The gas separation membrane module which consists of three engine was designed. The engine of creating a separation modular structure generates the arrangement coordinates data of a separation unit by GUI, and generates the input file for hydrogen recovery rate calculation. Since it can respond to a cylinder or square pillar type module outward form and can respond to unit arrangement of a triangular lattice or a square lattice, the output of the unit arrangement coordinates of general multi-pipe set structures (separation, extraction, heat exchanger, etc.) is possible for it. The engine of performing hydrogen recovery rate calculation applies a commercial computer fluid dynamics (CFD) package, receives an input data from a modular structure creation portion, and outputs a convergence calculation result to a file. The engine of optimizing a modular structure has managed the input data and the result for the recovery rate calculation stored in the data base, proposes the parameter for structure optimization using a genetic algorithm, and determines the data set for performing the next recovery calculation. It was suggested about the unit size and arrangement form in a separation modular structure that a genetic algorithm can optimize efficiently.

View full abstract

A quantitative structure and property relationships(QSPR) analysis on gas chromatograph relative retention times(GCRT) of dioxin related compounds(Dioxins), such as polychlorinated dibenzo-p-dioxin and polychlorinated dibenzofuran, and their molecular geometrical and topological parameters was carried out. It was found that there was a different change in GCRT depending on solvent accessible surface area(SAS) between the congeners and the isomers. To elucidate the retention characteristics of the isomers, the GCRT were classified in terms of the number of the pair of proximitive hydrogen substituents bonding to the dibenzo-p-dioxin skeleton(Nad-H). The correlation coefficients were markedly improved between SAS and GCRT ranging in the same Nad-H category. Next an analysis on the basis of the connectivity indices, which had been derived by Randic, Kier, and Hall, was performed to clarify the topological characteristics for the proximitive positions. It was found that there was a close correlation between the third-order connectivity descriptor, ³χ, and GCRT for the isomers. On the other hand, from analyses on the correlation between ³χ and some electrical descriptors, it was revealed that there was a significant correlation between ³χ and sum of net charge at hydrogen substituents.

View full abstract

The relationships between the bioconcentration factor (BCF) and the octanol/water partition coefficient (Pow) of about 400 chemicals, these were conducted the bioaccumulation study based on the Japanese Chemical Substances Control Low, were analyzed. The measured logPow data were collected from our laboratory database, Hansch or KowWIN database. The calculated logPow were computed by three logPow calculation softwares, ClogP, KowWIN and Pallas, which calculate logPow from chemical structure, respectively. The correlation with the measured logPow and the calculated logPow was low in ionic substances, but it was high in nonionic substances. In nonionic substances, the correlation of the measured logPow and logBCF was better as compared to those of the calculated logPow and logBCF. When the BCF was estimated from the measured logPow, there was no False-Negative (FN) that the observed logBCF is higher than a predicted logBCF, and is over the confidence limit of 95% of the predicted logBCF. However, when the BCF was estimated from the calculated logPow, a few FN existed. Thus, it is difficult to use a predicted logBCF from the calculated logPow as alternative data of screening test. The evaluation of BCF from the calculated logPow will appropriate to use as preliminary prediction of screening test.

View full abstract

In genomes and corresponding polypeptides, simple tandem-type sequence repetitions which are not significant as genes or proteins might frequently be found and called low complexity regions(LCR). LCR could disturb correct homology searches. BLAST includes a routine which estimates LCR using Shannon entropy. We focused a Tsallis entropy which does not satisfy additivity. Tsallis entropy is the one parameter(we call q) generalization of Shannon entropy. We applied the two parameters(q,q') entropy to BLAST in place of Shannon entropy, and expected more efficient homology searches of proteins. In homology searches, the gaps are often seen in scores and e-values. Some proteins such as the cancer control proteins which were derived by p53 genes were analyzed. Assuming the group of high e-value sequences form a protein family, we computed the hit ratio between the number of sequences in a family and the number of all searched sequences. In some cases, the searched number of sequences is larger than the number obtained by default Shannon entropy BLAST method. For almost all cases, the hit ratio was higher in the region satisfing the condition q' = 1/q. In another cases, the number of sequences which form the family have higher values around q' = 1/q.

View full abstract

It is important problem in a protein science to detect active sites on protein structures. Recently some automatic detection methods were presented; finding a similar structure to known protein active sites, finding pockets and/or cavities on protein surfaces, and so on. We already presented a method to detect amino acid residues, which construct a function site, using the similarity in 3-D structure of same function proteins. For the improvement of the finding method now we proposed to compare the side chain angles. We applied the new method to detect the function site of electron transfer proteins and that of acid proteases. The obtaining function site amino acid residues were compared to that from the previous method, which compare only the distance of alpha carbon atoms. The results showed that this improved method could more clearly find the function site.

View full abstract