The goal of the project 'Strategies to develop chemical categories in the context of REACH', funded by the Federal Ministry of Education and Research (BMBF), was to find groups of compounds that share a similar structural and toxicological profile.
A dataset including 1022 studies of 899 compounds that effected 28 different rat toxicity endpoints was analyzed. Challenges in this project were the high experimental error as well as the high number of missing experimental endpoint values. We applied Predictive Clustering Trees, a combined clustering and multi-label classification approach, to provide a prediction model as well as clusters with overlapping of structural and toxicological similarity.
Prediction web-page: http://mlc-reach.informatik.uni-mainz.de