Hiromasa Kaneko

Development of a New Regression Analysis Method Using Independent Component Analysis

In this paper, independent component analysis (ICA) and regression analysis are combined to extract significant components. ICA is a method that extracts mutually independent components from explanatory variables. A relationship between the independent components and an objective variable is constructed by the least-squares method. This method is named ICA-MLR (MLR = multiple linear regression). We verified the superiority of ICA-MLR over partial least squares (PLS) with simulation data and tried to apply this method to a quantitative structure-property relationship analysis of aqueous solubility. We constructed models between aqueous solubility and 173 molecular descriptors. PLS and genetic algorithm PLS models were constructed for a comparison of ICA-MLR. R2, Q2, and Rpred2 values of the PLS model are 0.836, 0.819, and 0.848, respectively. These values of the ICA-MLR model are 0.937, 0.868, and 0.894, respectively. ICA-MLR achieved higher predictive accuracy than PLS. ICA-MLR could extract effective components from explanatory variables and construct the regression model with high predictive accuracy. In addition, the information of regression coefficients bICA-MLR indicates the magnitude of contribution of each descriptor in the analysis of aqueous solubility.

Novel soft sensor method for detecting completion of transition in industrial polymer processes

Soft sensors are widely used to estimate process variables that are difficult to measure online. In polymer plants that produce various grades of polymers, the quality of products must be estimated using soft sensors in order to reduce the amount of off-grade material. However, during grade transition, the predictive accuracy deteriorates because the state in polymer reactors is unsteady, causing the values of process variables to differ from the steady-state values used to construct regression models. Therefore, we have proposed to construct models that detect the completion of transition to ensure that the polymer quality evaluated after transition conforms to the predicted one. By using these models and regression models constructed for each product grade, the polymer quality can be predicted with high accuracy, selecting a regression model appropriately. The proposed method was applied to industrial plant data and was found to exhibit higher predictive performance than traditional methods.

Adaptive soft sensor model using online support vector regression with time variable and discussion of appropriate hyperparameter settings and window size

Abstract Soft sensors have been widely used in chemical plants to estimate process variables that are difficult to measure online. One crucial difficulty of soft sensors is that predictive accuracy drops due to changes in state of chemical plants. The predictive accuracy of traditional soft sensor models decreases when sudden process changes occur. However, an online support vector regression (OSVR) model with the time variable can adapt to rapid changes among process variables. One crucial problem is finding appropriate hyperparameters and window size, which means the numbers of data for the model construction, and thus, we discussed three methods to select hyperparameters based on predictive accuracy and computation time. The window size of the proposed method was discussed through simulation data and real industrial data analyses and the proposed method achieved high predictive accuracy when time-varying changes in process characteristics occurred.

Applicability domain based on ensemble learning in classification and regression analyses.

We discuss applicability domains (ADs) based on ensemble learning in classification and regression analyses. In regression analysis, the AD can be appropriately set, although attention needs to be paid to the bias of the predicted values. However, because the AD set in classification analysis is too wide, we propose an AD based on ensemble learning and data density. First, we set a threshold for data density below which the prediction result of new data is not reliable. Then, only for new data with a data density higher than the threshold, we consider the reliability of the prediction result based on ensemble learning. By analyzing data from numerical simulations and quantitative structural relationships, we validate our discussion of ADs in classification and regression analyses and confirm that appropriate ADs can be set using the proposed method.

Inverse QSPR/QSAR Analysis for Chemical Structure Generation (from y to x)

Retrieving descriptor information (x information) from a value of an objective variable (y) is a fundamental problem in inverse quantitative structure-property relationship (inverse-QSPR) analysis but challenging because of the complexity of the preimage function. Herewith, we propose using a cluster-wise multiple linear regression (cMLR) model as a QSPR model for inverse-QSPR analysis. x information is acquired as a probability density function by combining cMLR and the prior distribution modeled with a mixture of Gaussians (GMMs). Three case studies were conducted to demonstrate various aspects of the potential of cMLR. It was found that the predictive power of cMLR was superior to that of MLR, especially for data with nonlinearity. Moreover, it turned out that the applicability domain could be considered since the posterior distribution inherits the prior distributions feature (i.e., training data feature) and represents the possibility of having the desired property. Finally, a series of inverse analyses with the GMMs/cMLR was demonstrated with the aim to generate de novo structures having specific aqueous solubility.

Criterion for Evaluating the Predictive Ability of Nonlinear Regression Models without Cross-Validation

We propose predictive performance criteria for nonlinear regression models without cross-validation. The proposed criteria are the determination coefficient and the root-mean-square error for the midpoints between k-nearest-neighbor data points. These criteria can be used to evaluate predictive ability after the regression models are updated, whereas cross-validation cannot be performed in such a situation. The proposed method is effective and helpful in handling big data when cross-validation cannot be applied. By analyzing data from numerical simulations and quantitative structural relationships, we confirm that the proposed criteria enable the predictive ability of the nonlinear regression models to be appropriately quantified.

Development of a New De Novo Design Algorithm for Exploring Chemical Space.

In the first stage of development of new drugs, various lead compounds with high activity are required. To design such compounds, we focus on chemical space defined by structural descriptors. New compounds close to areas where highly active compounds exist will show the same degree of activity. We have developed a new de novo design system to search a target area in chemical space. First, highly active compounds are manually selected as initial seeds. Then, the seeds are entered into our system, and structures slightly different from the seeds are generated and pooled. Next, seeds are selected from the new structure pool based on the distance from target coordinates on the map. To test the algorithm, we used two datasets of ligand binding affinity and showed that the proposed generator could produce diverse virtual compounds that had high activity in docking simulations.

Ring-System-Based Exhaustive Structure Generation for Inverse-QSPR/QSAR.

Inverse‐QSPR/QSAR aims to solve the inverse problem of chemical structure generation based on QSPR/QSAR models, once the properties or activities are specified. To efficiently solve this problem, an exhaustive ring‐system‐based structure generation methodology was developed. The concept of the applicability domain (AD) is automatically acknowledged within the proposed strategy. The local AD is considered by introducing the probability distribution of a given data set, and the universal AD is considered using ring‐system‐based fragments in the training data set. Structures with desired properties or activities are enumerated by assembling fragments, including atomic elements, in a tree‐like way. The usefulness of the proposed method is demonstrated through a case study of ligand design for the human alpha 2A adrenergic receptor (ADR2A_HUMAN). We succeeded in generating structures focusing only on a pre‐defined region in chemical space, resulting in structures whose desired activity has a high likelihood being efficiently generated. In addition, the limitations of our proposed method and future challenges are discussed.

Chemical-Space-Based de Novo Design Method To Generate Drug-Like Molecules

To discover drug compounds in chemical space containing an enormous number of compounds, a structure generator is required to produce virtual drug-like chemical structures. The de novo design algorithm for exploring chemical space (DAECS) visualizes the activity distribution on a two-dimensional plane corresponding to chemical space and generates structures in a target area on a plane selected by the user. In this study, we modify the DAECS to enable the user to select a target area to consider properties other than activity and improve the diversity of the generated structures by visualizing the drug-likeness distribution and the activity distribution, generating structures by substructure-based structural changes, including addition, deletion, and substitution of substructures, as well as the slight structural changes used in the DAECS. Through case studies using ligand data for the human adrenergic alpha2A receptor and the human histamine H1 receptor, the modified DAECS can generate high diversity drug-like structures, and the usefulness of the modification of the DAECS is verified.

Ring system-based chemical graph generation for de novo molecular design

Generating chemical graphs in silico by combining building blocks is important and fundamental in virtual combinatorial chemistry. A premise in this area is that generated structures should be irredundant as well as exhaustive. In this study, we develop structure generation algorithms regarding combining ring systems as well as atom fragments. The proposed algorithms consist of three parts. First, chemical structures are generated through a canonical construction path. During structure generation, ring systems can be treated as reduced graphs having fewer vertices than those in the original ones. Second, diversified structures are generated by a simple rule-based generation algorithm. Third, the number of structures to be generated can be estimated with adequate accuracy without actual exhaustive generation. The proposed algorithms were implemented in structure generator Molgilla. As a practical application, Molgilla generated chemical structures mimicking rosiglitazone in terms of a two dimensional pharmacophore pattern. The strength of the algorithms lies in simplicity and flexibility. Therefore, they may be applied to various computer programs regarding structure generation by combining building blocks.