This presentation concerns Generative Topographic Maps (GTM) [1] which is an universal tool to visualize the chemical space, to develop classification models, to estimate the applicability domain of models and to compare databases of chemical compounds [2].As a tool for visualization GTM overcomes most of drawbacks of such popular approaches as Principal Component Analysis (PCA), Sammon Mapping (SM) and Self-Organizing Maps (SOM). The probabilistic character of the GTM approach appears to offer additional advantages, which directly stem from the rigorous probabilistic character of the GTM approach. Thus, GTM can also be used to assess an overlap between the datasets, to develop classification models if information about activities of compounds is available and to assess the applicability domain of models.In this study, we present a supervised modification of GTM approach and its application to in silico assessment of chemical liabilities. Chemical liabilities, such as adverse effects and toxicity, play a significant role in drug discovery process, in silicoprediction of chemical liabilities is an important step aimed to reduce costs and animal testing by complementing or replacing in vitro and in vivo experiments. In supervised extension the class information is integrated in GTM training procedure to enrich description of the cluster structure provided by the model. Using proposed approach allowed one to increase the predictive classification performance of developed models up to 0.15 in balanced accuracy as well as data visualization performance for six data sets of varying size and difficulty (carcinogenicity, mutagenicity, hERG, acute toxicity, hepatotoxicity and phospholipidosis).References:[1] C.M. Bishop et al, Neural Computation 10 (1998) 215-234.[2] N. Kireeva et al. Mol. Informatics 31 (2012) 301-312Acknowledgements:Authors thank Russian Foundation for Basic Research (projects no. 11-03-00161 and 12-03-33086) and the French Embassy in Russia for the support. NK acknowledge Prof. Alexandre Varnek and Dr. Igor Baskin for their help and advice..

Supervised generative topographic mapping for in silico assessment of chemical liabilities

246th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.