logD7.4 modeling using Bayesian Regularized Neural Networks. Assessment and correction of the errors of prediction.

Bayesian Regularized Neural Networks (BRNNs) employing Automatic Relevance Determination (ARD) are used to construct a predictive model for the distribution coefficient logD7.4 from an in-house data set of 5000 compounds with experimental endpoints. A method for assessing the accuracy of prediction is established based upon a query compound's distance to the training set. logD7.4 predictions are also dynamically corrected with an associated library of compounds of continuously updated, experimentally measured logD7.4 values. A comparison of local models and associated libraries comprising separate ionization class subsets of compounds to compounds of a homogeneous ionization class reveals in this case that local models and libraries have no advantage over global models and libraries.

Classification of diverse organic compounds that induce chromosomal aberrations in Chinese hamster cells.

A data set of 297 diverse organic compounds that cause varying degrees of chromosomal aberrations in Chinese hamster lung cells is examined. Responses of an assay are categorized as clastogenic (>10% aberrant cells) and nonclastogenic (<5% aberrant cells). Each of the compounds is represented by calculated structural descriptors that encode topological, geometric, electronic, and polar surface features. A genetic algorithm (GA) employing a k-nearest neighbor (kNN) fitness evaluator is used to iteratively search a reduced descriptor space to find small, information-rich subsets of descriptors that maximize the classification rates for clastogenic and nonclastogenic responses. To further improve modeling, a similarity measure using atom-pair descriptors is employed to create more homogeneous data subsets. Three different data sets are examined. Results for a set of 297 compounds using the GA-kNN method were 86.5% and 80.0% correct classification in the training set and prediction set, respectively. Results for a subset of 279 compounds in model 2 are 85.7% and 85.7% for the training and prediction sets, respectively. Results for a subset of 182 compounds in model 3 are 91.5% and 94.4% for the training and prediction sets, respectively. Creating smaller, more topologically similar data sets result in improved classification rates.

QSAR and classification of murine and human soluble epoxide hydrolase inhibition by urea-like compounds.

A data set of 348 urea-like compounds that inhibit the soluble epoxide hydrolase enzyme in mice and humans is examined. Compounds having IC(50) values ranging from 0.06 to >500 microM (murine) and 0.10 to >500 microM (human) are categorized as active or inactive for classification, while quantitation is performed on smaller compound subsets ranging from 0.07 to 431 microM (murine) and 0.11 to 490 microM (human). Each compound is represented by calculated structural descriptors that encode topological, geometrical, electronic, and polar surface features. Multiple linear regression (MLR) and computational neural networks (CNNs) are employed for quantitative models. Three classification algorithms, k-nearest neighbor (kNN), linear discriminant analysis (LDA), and radial basis function neural networks (RBFNN), are used to categorize compounds as active or inactive based on selected data split points. Quantitative modeling of human enzyme inhibition results in a nonlinear, five-descriptor model with root-mean-square errors (log units of IC(50) [microM]) of 0.616 (r(2) = 0.66), 0.674 (r(2) = 0.61), and 0.914 (r(2) = 0.33) for training, cross-validation, and prediction sets, respectively. The best classification results for human and murine enzyme inhibition are found using kNN. Human classification rates using a seven-descriptor model for training and prediction sets are 89.1% and 91.4%, respectively. Murine classification rates using a five-descriptor model for training and prediction sets are 91.5% and 88.6%, respectively.