Martin Gütlein, David Vorgrimmler
Developing Prediction Models and Visualization Tools for Predictive Toxicology
We present recent developments of our group in the field of predictive toxicology and their application in current research projects. We introduce lazar (lazy structure-activity relationships), a modular framework for predictive toxicology. Similar to the read across procedure in toxicological risk assessment, lazar creates local QSAR (quantitative structure-activity relationship) models for each compound to be predicted. Model developers can choose from a large variety of algorithms for descriptor calculation and selection, chemical similarity indices, and model building. Our 3D viewer CheS-Mapper can be used to analyze the relationship between the structure of chemical compounds and their biological or chemical activity. It arranges compounds in virtual 3D space, such that compounds with similar physico-chemical or structural properties are close to each other. The software further supports a range of descriptor calculation methods, cluster algorithms and 3D embedding techniques. We describe how these tools are applied in two ongoing BMBF projects for descriptor calculation, model building, visulization and the prediction of compounds. The first project ExITox (Explain Inhalation Toxicity) aims at developing an integrated testing strategy (ITS) for human health risk assessment of repeated-dose toxicity after inhalation exposure. The second project developes chemical categories of structurally similar compounds with a common toxicology profile in the context of REACH.