Classification of a Diverse Set of Tetrahymena pyriformis Toxicity Chemical Compounds from Molecular Descriptors by Statistical Learning Methods

Abstract

Toxicity of various compounds has been measured in many studies by their toxic effects against Tetrahymena pyriformis. Efforts have also been made to use computational quantitative structure-activity relationship (QSAR) and statistical learning methods (SLMs) for predicting Tetrahymena pyriformis toxicity (TPT) at impressive accuracies. Because of the diversity of compounds and toxicity mechanisms, it is desirable to explore additional methods and to examine if these methods are applicable to more diverse sets of compounds. We tested several SLMs (logistic regression, C4.5 decision tree, k-nearest neighbor, probabilistic neural network, support vector machines) for their capability in predicting TPT by using 1129 compounds (841 TPT and 288 non-TPT agents) which are more diverse than those in other studies. A feature selection method was used for improving prediction performance and selecting molecular descriptors responsible for distinguishing TPT and non-TPT agents. The prediction accuracies are 86.9% approximately 94.2% for TPT and 71.2% approximately 87.5% for non-TPT agents based on 5-fold cross-validation studies, which are comparable to some of earlier studies despite the use of more diverse sets of compounds. The selected molecular descriptors are consistent with those used in other studies and experimental findings. These suggest that SLMs are useful for predicting TPT potential of diverse sets of compounds and for characterizing the molecular descriptors associated with TPT.

Prediction of antibacterial compounds by machine learning approaches

“…We further separated these compounds into training set and testing set by two different methods, 5-fold cross-validation and independent evaluation set, which have been widely used in the published ML compound prediction studies. [16][17][18][19][20]26,27 For 5-fold cross-validation, these compounds were randomly divided into five subsets with approximately equal size. Four subsets were arbitrarily chosen to form the training set to train the SVC model, and then the fifth subset was used as the testing set to examine the performance of the SVC model.…”

“…[16][17][18][19][20]26,27,[30][31][32][33][34] In this work, a group of 198 molecular descriptors was manually selected as the possible candidates from more than 1000 descriptors commonly described in the literature by roughly excluding those descriptors that are obviously redundant or irrelevant to the pharmacological and biological properties. These descriptors as described in the earlier studies 16,20 are given in Table 1, including 18 descriptors in the class of simple molecular properties, 27 descriptors in the class of molecular connectivity and shape, 97 descriptors in the class of electrotopological state, 31 descriptors in the class of quantum chemical properties, and 25 descriptors in the class of geometrical properties. The values of these descriptors were computed from the 3D structure of each compound by using our own developed molecular descriptor computing program.…”

“…11,13,39,40 SVC was originally developed by Vapnik and coworkers 11 and it has shown promising capability for solving a number of biological classification problems. 14,[16][17][18][19][20][23][24][25][26][27][28][29] SVC is based on the structure risk minimization (SRM) principle from statistical learning theory. 11 For linearly separable cases, SVC performs classification tasks by constructing a hyperplane in a multidimensional space to separate two different classes of feature vectors with a maximum margin.…”

“…3 Therefore, it is highly desirable to explore methods that facilitate the identification of antibacterial compounds in the early drug design phase without requiring the knowledge about their mechanisms, the intrinsic relationships between activities and molecular properties, and the structures of targeted proteins and other macromolecules and molecular assemblies. [11][12][13][14][15][16][17][18][19][20] Quantitative structure activity relationship (QSAR) has been successfully applied in the prediction of antibacterial compounds. For instance, Cronin et al 21 have developed the linear discriminant analysis (LDA) and binary logistic regression (BLR) models on the basis of two-dimensional (2D) structural and physicochemical descriptors with accuracies of 87.4 and 92.0% for the prediction of antibacterial compounds, and 97.3 and 95.7% for nonantibacterial compounds, respectively.…”

Graph Wavelet Alignment Kernels for Drug Virtual Screening

“…Many machine learning and data mining algorithms have been applied to study the structure-activity relationship of chemicals. For example, Xue et al reported promising results of applying five different machine learning algorithms: logistic regression, C4.5 decision tree, k-nearest neighbor, probabilistic neural network, and support vector machines to predicting the toxicity of chemicals against an organism of Tetrahymena pyriformis 21 .…”