Supervised extensions of chemography approaches: case studies of chemical liabilities assessment.

Chemical liabilities, such as adverse effects and toxicity, play a significant role in modern drug discovery process. In silico assessment of chemical liabilities is an important step aimed to reduce costs and animal testing by complementing or replacing in vitro and in vivo experiments. Herein, we propose an approach combining several classification and chemography methods to be able to predict chemical liabilities and to interpret obtained results in the context of impact of structural changes of compounds on their pharmacological profile. To our knowledge for the first time, the supervised extension of Generative Topographic Mapping is proposed as an effective new chemography method. New approach for mapping new data using supervised Isomap without re-building models from the scratch has been proposed. Two approaches for estimation of model's applicability domain are used in our study to our knowledge for the first time in chemoinformatics. The structural alerts responsible for the negative characteristics of pharmacological profile of chemical compounds has been found as a result of model interpretation.

Nonlinear dimensionality reduction for visualizing toxicity data: distance-based versus topology-based approaches.

Over the years, a number of dimensionality reduction techniques have been proposed and used in chemoinformatics to perform nonlinear mappings. In this study, four representatives of nonlinear dimensionality reduction methods related to two different families were analyzed: distance-based approaches (Isomap and Diffusion Maps) and topology-based approaches (Generative Topographic Mapping (GTM) and Laplacian Eigenmaps). The considered methods were applied for the visualization of three toxicity datasets by using four sets of descriptors. Two methods, GTM and Diffusion Maps, were identified as the best approaches, which thus made it impossible to prioritize a single family of the considered dimensionality reduction methods. The intrinsic dimensionality assessment of data was performed by using the Maximum Likelihood Estimation. It was observed that descriptor sets with a higher intrinsic dimensionality contributed maps of lower quality. A new statistical coefficient, which combines two previously known ones, was proposed to automatically rank the maps. Instead of relying on one of the best methods, we propose to automatically generate maps with different parameter values for different descriptor sets. By following this procedure, the maps with the highest values of the introduced statistical coefficient can be automatically selected and used as a starting point for visual inspection by the user.

Impact of distance-based metric learning on classification and visualization model performance and structure-activity landscapes.

This study concerns large margin nearest neighbors classifier and its multi-metric extension as the efficient approaches for metric learning which aimed to learn an appropriate distance/similarity function for considered case studies. In recent years, many studies in data mining and pattern recognition have demonstrated that a learned metric can significantly improve the performance in classification, clustering and retrieval tasks. The paper describes application of the metric learning approach to in silico assessment of chemical liabilities. Chemical liabilities, such as adverse effects and toxicity, play a significant role in drug discovery process, in silico assessment of chemical liabilities is an important step aimed to reduce costs and animal testing by complementing or replacing in vitro and in vivo experiments. Here, to our knowledge for the first time, a distance-based metric learning procedures have been applied for in silico assessment of chemical liabilities, the impact of metric learning on structure-activity landscapes and predictive performance of developed models has been analyzed, the learned metric was used in support vector machines. The metric learning results have been illustrated using linear and non-linear data visualization techniques in order to indicate how the change of metrics affected nearest neighbors relations and descriptor space.