Multivariate regression methods such as a partial least squares (PLS) method and support vector regression (SVR) are powerful tools for handling several problems in chemoinformatics. Attempts to construct models having high predictive accuracy have been made by using those methods, and then significant results have been produced. On the other hand, because predictability of constructed models differ among query samples predicted with these models, it is important to estimate prediction errors of these samples. Therefore, in this study, we tried to quantify relationships of applicability domains (AD) of regression models and prediction errors. The larger distances to models (DM) are, the lower the accuracy of prediction would be estimated. We used Euclidean distances to an average of training data and ones to the nearest sample in training data as DM, and PLS and SVR as methods constructing regression model. The proposed method were applied to quantitative structural-property relationships (QSPR) and soft sensor analyses. Estimate accuracy of prediction errors increased in QSPR analysis. In soft sensor analysis, higher fault detection ability was achieved than a traditional method.

View full abstract

Regression analysis is one of the most basic and important methods when we need to predict a response and evaluate the contributions of factors to the response. In pharmaceutical area, the regression analysis is applied to a number of datasets in various studies such as QSAR, PK/PD, pharmacoepidemiology and so on. Because the relationships between the response and factors are often complicated in this area, there are a number of cases in which nonparametric regression analyses which enable us to build a nonlinear model flexibly are applied. But it is difficult to evaluate a contribution of each factor explicitly in some nonparametric procedures. So, in the previous study, we developed Simple Additive NOnparametric Regression Analysis (SANORA) which is a nonparametric regression procedure with a simple model and algorithm and enables us to evaluate the contributions of factors. In this study, when we applied SANORA to the dataset in which the interaction between variables had been reported, the appropriate model for the prediction by using the interaction term was obtained. In addition, in the model which enables us to evaluate the contribution of each variable, the nonlinear relationship between the response variable and the interaction term was indicated.

View full abstract

QSAR (Quantitative Structure Activity Relationships) is a method for building models to predict activity of compounds and a protein by using statistic methods. When we build a QSAR model, we use only information of compounds, not proteins. Recently, new QSAR methods using not only compound information but also protein information have been researched. These methods are considered to have some merits of good prediction power and so on. One of these methods is Support Vector Machine-Target Ligand Kernel (SVR-TLK), which use product of ligand kernel and protein kernel as a kernel of SVR. In this study, we used SVR-TLK to activity data set of Dopamine type 2 and type 3. Descriptors of compounds are constitutional descriptors and topological descriptors and so on, and descriptors of proteins are frequency of a pair of 20 amino acids. The SVR-TLK model shows better Q2 value than normal QSAR models. Therefore, usefulness of SVR-TLK has been proved.

View full abstract

In the field of pharmaceutical industry and other areas, QSPR/QSAR method is widely used for efficient design of functional molecules having desired combination of properties. In this study, we propose a novel methodology for helping molecular design using inverse-QSPR/QSAR method. This method enables us to exhaustively generate chemical structures which have desired properties or activities. The method is composed of three parts: construction of QSPR/QSAR models, analysis of the model inversely, generation of structures exhaustively. QSPR/QSAR models are constructed with MLR/OLS for the following inverse analysis. A mixture of Gaussian distributions is used to characterize the probability distribution of data set in descriptor space. Combining these probability models, inverse analysis is conducted using the framework of linear Gaussian models. Exhaustive chemical structures are generated by using specific kind of descriptors inside the pre-determined region in which the probability density values are high. It is also possible to determine the region considering applicability domain of QSPR/QSAR model. By using proposed method, efficient structure generation is expected.

View full abstract

As candidate compounds for medicine, they have to show good balance between high inhibitory activities and desirable ADMET properties. Several ADMET end points are measured routinely and then prediction for their values in advance is crucial for efficient lead optimization. We have built ADMET local models using MOE descriptors and statistical package R. Furthermore, we have develpoed Web GUI for chemists, who can make designs of any intersing molecules in desktop PC.

View full abstract

SYNSUP is a synthetic route design system, which proposes possible routes for a given target molecule non-interactively. In our initial approach called Exhaustive Search the more complex the molecule is the thicker the branch of synthetic tree evolves. This combinatorial explosion prevents getting synthetic routes for complex molecules that require 6 steps or more. Experienced synthetic chemists, on the other hand, can find synthetically important bonds in the molecule, and try to modify the molecule so that some reactions can be applied to make those bonds. Inspired by this mental process we developed new concepts, center of complexity and key-bond. We also classify reactions in two categories, constructing and facilitating. The latter is functional group exchange and deprotection reaction, and the priorities for them are low in principle. However, we deliberately promote the priorities when there is no way to apply construction reaction to the key-bonds. We call this approach Selective Search. We present the method to implement the algorithm to SYNSUP. Test runs were carried out with some natural bioactive compounds, and the results showed Selective Search is effective for dealing with complex molecules over Exhaustive Search.

View full abstract

We have been investigating to develop synthetic routes of efficient organic compounds by fusing theoretical calculations and synthetic route design systems (SRDS). It is necessary to select effective synthetic routes from many routes SRDS creates. In the present study, we adopted an amide derivative with the physiological activity to confirm that the theoretical computations are useful to choose possible synthesis routes created by SRDS. We examined synthetic routes such as amidation, Piner-pyrimidine synthesis, Beckmann rearrangement reaction, and Suzuki-Miyaura cross coupling reaction. It was calculated that the target compound can be synthesized using the four synthetic routes. It was reported to obtain the target in 61.2% yield using the amidation reaction. The result is consistent with the calculated one. It was confirmed that PH3 instead of PPh3 should not be used to investigate the mechanism of Suzuki-Miyaura reaction.

View full abstract

Amino acid tautomerizes from neutral form to zwitterionic form. It is known that zwitterionic form is more stable than neutral form in aqueous solution, althogh zwitterionic form is very unstable in gas phase. This difference originates from solvation effect. We put 101 water molecules around the solute molecule explicitly. The double perturbation approach was adopted. In this approach, 20200000 solvent configurations were generated at the MM/MM level of theory in Monte Carlo simulation, and 2020 configurations were selected randomly among the random walk. Selected configurations were calculated by means of QM/MM level of theory, so that we obtained ensemble average of hydration structure. Free energy perturbation theory was performed to calculate free energy change. The agreement between our result (7.2 kcal/mol) and experimental result (7.3 kcal/mol) is fairly good. Our method is effective to calculate solvation effect.

View full abstract

We developed the QM/MC/FEP program as a method to theoretically estimate the solvent effect for chemical reactions. The method was applied to Diels-Alder reactions, Cope and Kemp reactions. As the method uses Monte Carlo simulation, the method can be useful to analyze reactions in supercritical fluids. In the present study, we investigated hydrolysis of 4-((2-hydroxyethoxy) carbonyl) benzoic acid as a model of /poly/-ethylene terephthalate. The result is consistent with the experimental fact that a reaction is accelerated in supercritical water. The hydrolysis for a dimer of e- aminohexanoic acid as a model of 6-nylon is also investigated in supercritical water.

View full abstract

In water cluster, the hydrogen bond (H-bond) network plays an important role in the stability of the cluster. Many theoretical studies have focused on the local minimum structures which are located in bottoms of the potential energy surface. At finite temperatures, however, the accessible configuration space of water cluster becomes larger because of the weakness of H-bond. Then, the local minimum structures are no longer the representative of the configurations. In this work, we study water clusters at finite temperature based on the H-bond pattern, i.e., H-bond network of the cluster. H-bond pattern describes H-bond network of the water cluster as directed graph, i.e., digraph, where water molecules and H-bonds are represented by vertices and arcs, respectively. One billion of configurations in NVT ensemble formed by Monte Carlo simulation are divided into the corresponding H-bond patterns. The result suggests that the H-bond pattern can classify the water cluster configurations and characterize the dipole moments of water clusters and the constituting molecules.

View full abstract

Peroxisome prolierator-activated receptor γ(PPARγ) is one of the nuclear receptors which control the transcription depending on ligands.The agonists of PPARγ have effects on the therapy for adiposity, insulin resistance and hyperlipidemia.Various kinds of molecules are known as the ligands of PPARγ, among which 15-deoxy-Δ12,14-prostaglandin J2 (15d-PGJ2) covalently binds to a cysteine residue in the PPARγ-ligand binding pocket.The details of the reaction mechanism are not known.There are a lot of possible structures for the side chains of the amino acid residues as well as its ligand.It is necessary to take these possible structures into consideration to assess the reactivity of ligand to the protein.In this study, we show reaction paths between 15d-PGJ2 and a cysteine residue taking account of the direction, the orientation and the position of addition of a cysteine residue to ligand and conformations of ligand by means of ab initio MO method.

View full abstract

The Palladium-catalyzed Mizorogi-Heck reaction between aryl halides and nucleophiles has been intensively development for their important organic synthetic applications. In the past theoretical report using chelating diphosphine-paradium dihalide complex to catalyst, the chloride ion of Pd comlex is dissociated and reaction mechanisms of the catalytic cycle for the Mizorogi-Heck reaction are investigated using chelating diphosphine-Pd(0) neutral and chelating diphosphine-Pd(II) halide cation complexes. Although these molecular structures are not exactly known, coordinative unsaturated Pd complexes such as PdL2 and PdL2Cl+ are assumed to be the active species. However, the bond energy for the chloride ion of Pd complex is about 110 kcal/mol with the DFT method and very strong. Therefore, it is considered that the dissociation of the Pd-Cl bond doesn't easily occur. In this work, we theoretically investigated the Pd-catalyzed Mizorogi-Heck reaction between bromobenzene and ethylene with the DFT method. Here, we focus on the prediction of the active species in the PdCl2(dppe) (dppe = Ph2PCH2CH2PPh2) catalyzed Mizorogi-Heck reaction. Our purpose here is to clarify the active species. Also, we examined the formation process of the active species without the dissociation of the chloride ion from PdCl2(dppe) complex.

View full abstract

It is well known that molecular properties, including biological functions, are closely related not only to the atomic connectivity but also to the 3D spatial arrangements of the atoms. In the present work, to describe the distribution of particular atoms in the 3D space, we have proposed two types of structure feature representations. For a given molecule, principal component analysis was carried out using 3D coordinate data for each atom. We have referred a set of three eigen values and its contribution ratios as PCA pattern. CGD pattern is also created by discretizing all distances between a center of gravity and each atom. We have prepared a target data set that contains 2,950 molecular structures taken from CSD, and represented by two patterns mentioned above. Using PCA pattern, the dataset was classified linear chains, plain rings, and spheres. On the other hand, by using CGD pattern, the structures which have free space around the center of gravity were identified. Structural symmetry such as 2-poloid and 3-poloid is also taken into the database search. As a result, we successfully found the molecules that have the common structural feature based on PCA and CGD pattern representations.

View full abstract

It is well known that the three-dimensional (3D) structure of a protein is closely related to its function. Some proteins such as hemoglobin possess a quaternary structure, which is resulted from the interaction of more than one protein chain. These structures are also important to understand more protein complex functions. In our laboratory, the quaternary structure classification database have constructed based on the combination of chain and domain patterns. In the present work, the geometrical relations between subunits are taken into account. To describe such structural feature of proteins, the authors have defined the number of neighborhood residues. That is the number of residue located in the area of radius r from the residue considered. More detailed information about the neighborhood residues is used to identify the interaction region of subunits. We have prepared a target data set that contains 397 quaternary structures which are consisted of four chains and belonged to four all-α domains. Using our representation, the dataset was classified into four structure classes based on the pattern of the interaction region. These results show the potential applicability of this method for quaternary structural classification classification of proteins.

View full abstract

2,2':6',2"-terpyridine undergoes on/off fluorescence switching via heat-mode reversible interconversion between two different crystal structure (typ-N and tpy-P). The switching mechanism in molecular level is not clearly resolved because the two interconvertible molecular structures tpy-P and tpy-N show quite small structural and spectroscopic differences. In this computational study, we calculated the ratio of non-radiative decay rate based on Fermi Golden Rule, where electronic coupling and Franck-Condon (FC) term were independently examined in detail. The slight difference in the molecular structure of tpy-N and tpy-P indicates the differences in the electronic coupling term is too small to play a definitive role in their drastic fluorescent behavior. We thefore focused on the FC-term and found the delicate balance between inner- and outer- reorganization energies and the driving force (the difference of Gibbs free energy between initial- and final states), which decisively affect the non-radiative decay rate.

View full abstract

In order to evaluate pi-conjugation effects on intramolecular resonance-assisted hydrogen bonds, DFT calculations were carried out on 2-hydroxycyclohexadienones fused position-isomerically with 3-, 5-, or 7-membered pi-conjugated rings. The values for the aromaticity index HOMA on the basis of the bond alternation in the outer pi-conjugated rings for the compounds with the hydrogen-bond form are always larger than the values for those with the corresponding non-hydrogen-bond form. The HOMA index depends noticeably on the positional isomerism, and is sensitive to the number of the pi-conjugation in the fused rings. The aromaticity index NICS on the basis of the NMR chemical shifts for the fused rings does not seem to depend on the positional isomerism. It, however, increases negatively for the compounds in the hydrogen-bond form with the 3- and 7-membered fused rings compared to those in the corresponding non-hydrogen-bond form, indicating an enhancement of the aromaticity, but increases positively for those with the 5-membered fused ring compared to those in the corresponding non-hydrogen-bond form, displaying an enhancement of the anti-aromaticity.

View full abstract

The study of intermolecular interactions is a useful step toward gaining a basic understanding of the dynamic behaviour of molecules in various environments. In the present study, high-level ab initio calculations [MP2] using correlation consistent basis sets of Dunning [aug-cc-pVTZ] were carried out to study the potential energy surfaces of CH3OCH3/Ar, CH3OCF3/Ar and CF3OCF3/Ar Van der Waals dimers and to clarify the stable structures of these clusters. For the calculation of interaction energies, the basis set superposition errors (BSSE) were estimated by counterpoise method. In the present model, we assumed Ar atom exists in the COC plane of CH3OCH3, CH3OCF3 and CF3OCF3. The potential energy surfaces of CH3OCH3/Ar and CH3OCH3/Ar exibits three minima and five minima, respectively. The potential energy surface of CH3OCF3/Ar partially resemble to that of CH3OCH3/Ar and that of CF3OCF3/Ar. The most stable structures of CH3OCH3/Ar, CH3OCF3/Ar and CF3OCF3/Ar are predicted to those in which Ar atoms is on the bisectors of COC plane and is opposite to oxygen atoms. The details of the potential energy surfaces are discussed.

View full abstract

The palladium-catalyzed C-C coupling between aryl halides or vinyl halides and activated alkenes in the presence of a base is referred as the "Heck Reaction". Previous papers used a cation complex detached Cl^- from [Pd(P-N)Cl_2 ] as an active species, where P-N indicate a ligand with two coordination sites. However, the cation complex should not be a reactant of the reaction since it is very unstable in comparison with its neutral complex. It is considered that the neutral Pd complex reacts with the alkenes and aryl halides in the reaction mixture to form a real reactant for the reaction. In the present study, we tried to find the reactant using a model complex as well as ethylene, PhBr and concluded that [Pd(P-N)PhBr] is the starting complex for Heck reaction.

View full abstract

In order to analyze and predict chemical reactions of biomolecules in water from the viewpoint of thermodynamics, an appropriate molecular modeling is needed for calculating quantitatively the changes in enthalpy and entropy of the chemical reactions. In this study, we adopted a cluster model in which a solute was placed into a water cluster with cavity to investigate the structure and stability of three short oligopeptides containing 3, 5, and 7 alanine residues, in a helix and extended form, in water. The hydration shells around the fixed oligopeptides were equilibrated through a 1-ns molecular dynamics (MD) simulation. and the sampling of cluster structures was carried out every 100-ps MD simulation after the equilibration. The obtained cluster structures were further fully optimized by the semiempirical molecular orbital calculations using the PM6 hamiltonian. Entropy of the hydrated oligopeptides was also obtained by the FORCE calculations. From Gibbs energy of the oligopeptides, we found that, in vacuo, oligopeptides in an extended form was more stable than those in α helix, while the propensity of these oligopeptides was reversed in water.

View full abstract

Studies of icosahedrally symmetric virus coat proteins (capsids) constitute one of the developing fields in recent computational chemistry. For easier and more efficient modelling of icosahedrally symmetric capsids, we developed a new version of the molecular graphics software, RasMol, named RasMol-v by implementing new functions concerning icosahedral symmetry. The most important ones of the newly implemented commands are "copymol", "addico", and "packmol". By typing "copymol", the entire capsid structure is built using a protomer structure stored in the Protein Data Bank. This command can also build any selected protomers by writing their protomer numbers as command parameters. By typing "addico", a cage of asymmetric unit of icosahedral symmetry is shown. This command can also show all the 60 cages of asymmetric unit that enclose the entire capsid structure. By typing "packmol", the amino acid residues that are placed outside the asymmetric unit are moved into the unit by rotation with the matrices of icosahedral symmetry. The newly implemented commands can be also executed through the menu bar.

View full abstract

Beckmann rearrangement has been using reaction to synthesize N-substituted amides from keto-oxime. Industrially, the reaction is widely used in the synthesis of caprolactam as monomers for nylon. Strong acid such as sulfuric acid as well as high-temperature are required to proceed the reaction with a large amount of by-products. Ishihara reported the reaction using cyanuric chloride as catalyst. In this case, Beckmann rearrangement progresses under mild conditions. However, a detailed reaction mechanism is not clear. In the present study, the catalytic behavior of cyanuric chloride was investigated for Beckmann rearrangement of cyclohexanone oxime as well as cyclododecanone oxime. It was confirmed that HCl yielded during the reaction play a very important role in decreasing the activation energy of the rearrangement reaction.

View full abstract

Development of faster computational methods for more quantitative evaluation of solvation free energy becomes increasingly important for performing a larger scale molecular simulation for a more complicated molecular state in solution. Hydration shell model is one of the fastest approaches for calculating the solvation free energy, but it does not seem to be suitable for the exact evaluations. In this study, we have tried to improve the continuum model by the usage of the solvent excluding (Se) surface compiled by Michel Sannar's molecular surface (MSMS) method, and compared with the general model by using the solvent accessible surface. Two assignment rules for assigning partial region of MSMS-Se surface to each atom were also tested. In comparison with the experimental solvation energies of some alkanes, the calculated values by adopting the rule that a partial region of the Se surface is assigned to the atom having nearest adjacent van der Waals surface for the solute surface, can represent the experimental values more qualitatively than those by another assignment rule based on the atomic position of closest approach to the solute surface.

View full abstract

Irregular chemical vibrations have been observed in several chemical reactions such as Belousov-Zhabotinsky reactions, a certain kind of redox reactions, interfacial chemical reactions, and enzyme reactions. It is known that these reactions form chaos and the chaotic dynamics is embedded in a strange attractor. The strange attractor is represented mathematically by the branched manifold also called the template or the knotholder. A new classification method of the chemical reactions is expected by examining the structure of template, and further there is a hope which develops into the classification of metabolisms and physiologies. S.Smale had found the unstable asymptotic periodic orbits are buried in the template of the horseshoe map. The structure of the branched manifold uniquely defines the intertwining or the braid between all orbits. Inversely, the braid is obtained from the relative rotation rates which can be calculated from the experimentally constructed strange attractor. In the present report, the possibility of the topological analysis was verified in the case of Roessler differential equation system which is one of the typical nonlinear chemical reaction model.

View full abstract

Estimation of prediction accuracy is a significant task to improve quality of several design problems such as material design and manufacturing process design. We have developed a new method to assess reliability of predictions by quantitative models. The method is based on the idea that the accuracy level of estimations strongly affected by the density of the data used for training quantitative models. Thus we make use of one-class support vector machines (OCSVM), which is typically used for determination of high density region. We have modified the algorithm and outputs of OCSVM to estimate reliability of predictions quantitatively by incorporating several properties of prediction models into an optimization problem. For validation, we have prepared two datasets of organic chemical structures with thousands of experimental values of aqueous solubility and melting points. Both linear and non-linear quantitative structure-property relationships (QSPR) models have been built and we have tested their prediction reliabilities on each data point using proposed method.

View full abstract

We proposed a new production methods of chitin from prawn husk or crab shell with low environmental load. This method is characterized by using a converge mill and a special enzyme for saccharification of prawn husk or crab shell. Concretely, we fabricate powders of prawn husk or crab shell and with the mill, firstly. Next, we use the enzyme reaction. In this case, we would like to know the crystallization rate and the powders particle diameter distribution because we should expect the influence of the crystallization rate and the powders particle diameter distribution toward the saccharification rate. X-ray diffraction (XRD) is usually used for the estimation of the powdered chitin. However, we utilize the near infrared spectrometer, more conveniently if we could estimate the crystallization rate and the powders particle diameter distribution. We report the result and have some discussion in this poster.

View full abstract

We have developed a system that automatically identifies the reaction models involved in the deposition of films on substrates with nanometer- or micrometer-sized trenches by Chemical Vapor Deposition (CVD) processes. The reaction models, which comprise various deposition species and surface reactions with different values of sticking coefficient, were determined both quantitatively and qualitatively on the basis of chemical kinetics. Automatic modeling is achieved by the system through the analysis of the experimental data of the cross-sectional shapes of the deposited films. The system consists of three components, that is, a user interface, an inference engine, and a process simulator based on Simplified Monte Carlo (SMC) method that reproduce film shapes. The inference engine proposes a reaction model, examines its validity by comparing the experimental results with the predicted ones obtained by using the simulator, and then corrects it using both simple Genetic Algorithms (GA) and Real-Coded GA (RCGA). By the use of the system, users can determine the appropriate reaction model for a given set of experimental data by simply inputting the experimental data of film shapes and conditions to the system. We demonstrated that the system could successfully model the reaction mechanisms by synthetic experimental data.

View full abstract

Science partnership program, class A, was done by Mikio Kaihara, invited presenters, Haruo Hosoya and Shinich Nakayama for high school students. In this seminar, we introduced two idea processing methods, the mindmap by Tony Buzan and the expansion of thinking by Yhotaro Hatamura. Both have quite similar characteristic, similar to the KJ method by Jiro Kawakita. At first stage, both methods adopted making much idea as possible as we can. Next, grouping and labeling are adopted. However, the mindmap is essentially open end type of idea processing. Oppositely, the expansion of thinking is converged to one result, solving problems. Therefore, we proposed a new method, which brought both merits, as the expansion of thinking with much colorful image. This looks like a picture scrolls, and by using them, we look forward to understanding the contents of basic chemistry, more easily. After participants tried such methods for expressing their interpretation on oxidation-reduction or thermochemical equation and so on,we heard their opinion towards such methods. And we took relatively good reaction from them.

View full abstract

In the field of the drug discovery, methodology to construct high-quality virtual library has been desired for efficient lead discovery. To construct useful virtual library, synthetic feasibility of chemical structures is an important factor as well as drug-likeness, diversity, and so on. However, there are only a few studies of synthetic feasibility to construct virtual library. Thus we have developed the system to construct virtual library considering synthetic feasibility. The proposed system is based on the concept of computer-assisted chemical reaction prediction using "transform". Transform is information that abstracts reaction rule based on known reaction and is able to make virtual reaction. If it is applied to chemical structures of reactants, chemical structures of products will be predicted with reaction information, such as a reaction name, reaction condition, etc. This information shows that chemical structures of products have synthetic feasibility. This system enables us to construct virtual library which consists of chemical structures that can be synthesized.

View full abstract

Healthcare professionals such as physicians, pharmacists and nurses are obliged to report adverse effects of drugs to the Ministry of Health, Labour and Welfare. The reported information is stored into a database and provided on the website of Pharmaceuticals and Medical Devices Agency in order to promote their proper use. So, it is informative for not only healthcare professionals but also scientists that the important and available information about adverse effects of drugs is analyzed and used effectively.Therefore, we tried to develop efficient procedures for extracting more valuable information from this database with the aim of building the system which enables us to obtain analyzed results on adverse effects such as the prediction of the event probabilities. In this study, we introduce an example of the application of our procedures to the analysis of adverse effects on the central nervous system when antibacterial and antiviral drugs are used, since they are often used and reported as the drugs having adverse effects on the central nervous system which is one of adverse effects with the worst prognosis.

View full abstract

The synthetic yield is considered to be one of the most important factors in determining which synthesis route is better. We have been developed a method to predict yields of synthetic reactions using experimental data in papers. In a previous study, we tried to make an equation predicting the yield of Prilezhaev reaction with a little success. It is because we used experimental data performed under the different experimental conditions. In order to obtain a better result for the expectation, we have to use experimental results with the unified conditions. In the present study, we planned to do a series of experiments of microreactors for Prilezhaev reaction using styrene and its derivatives for that purpose. Theoretical calculations at the B3LYP/6-31G* level of theory have done to obtain activation energies required for making the equation for predicting experimental yields.

View full abstract