Flexible and transparent computational workflows for the prediction of target organ toxicity

Abstract

In silico modeling of target organ toxicity has been held back in part by an inability to capture all relevant information into a meaningful reductionist approach. It has also been considered at times too simplistic, using data of often variable quality and seldom allowing the user to assess the relevance to the intended use. The purpose of this study was to develop a novel computational toxicology workflow system, to allow the users greater control and understanding of the target organ toxicity prediction. The workflows were built on the KNIME open-access platform which allows pipelining via a graphical user interface. Various building blocks, known as nodes, were incorporated, to access chemical inventories and/or databases, to profile structures and calculate properties and to report prediction results. The “basic” user sees a web-interface, whilst a “trained” user can go behind this to interrogate the nodes and, if required, link to additional data sources or investigate and update the models. The workflow was developed to address in particular the prediction of target organ toxicity of cosmetic ingredients. It comprises an inventory of over 4,400 unique chemical structures (cosmetic ingredients and related substances). The database contains repeat dose toxicity data for over 1,100 compounds including NOEL values. Thus, a user is able to search for similar compounds in the inventory file or database. The compound is then profiled using relevant structural alerts and chemotypes, currently comprising 108 alerts for protein reactivity, 85 for DNA binding, 32 for phospholipidosis and 16 for other liver toxicity endpoints. The workflows are flexible and transparent, they are successful in guiding a user through the process of making a prediction of target organ toxicity. Supported by the EU FP7 COSMOS Project.