Computational Workflows for Toxicity Prediction

Open and flexible platforms to capture modelling processes are required to support data capture, storage and retrieval, links of chemistry to pathways through Adverse Outcome Pathways (AOPs) as well as transparent modelling to evaluate the safety of chemicals to humans.

KNIME is the modular integration platform for the methods developed within COSMOS. By means of graphical workflows, data is read from various data sources and subsequently transformed into suitable formats for model building and/or visual analysis. The KNIME technology integrates access to databases, data processing and analysis, as well as modelling approaches into flexible computational workflows. They will be adaptable and form a set of building blocks allowing users to incorporate their own data and search existing data compilations.

KNIME workflows have been developed to identify structural rules, fragments and properties associated with particular mechanisms of toxicity. They use mechanism-based rule bases coded as SMARTS patterns to define chemotypes, e.g. for protein binding (Enoch et al, 2011) and DNA binding (Enoch et al, 2012). A chemical can be searched for a particular structural rule, associated with a mechanism of action, and compounds with the same rules can be identified from a compound list, e.g. the COSMOS database for cosmetics ingredients, to form a category. For this category toxicological data can then be retrieved, which allows for read-across to predict toxicity. A key part of these workflows is to provide the link between mechanism of action (which may be defined through an AOP) and chemistry to allow relevant compounds to be grouped together.

Moreover, an algorithm has been developed as KNIME node and workflow that queries the COSMOS database to export oral toxicity studies with a set of qualified criteria acceptable to toxicologists and risk assessors.

The development of the chemotype concept has been supported within COSMOS by the development of new KNIME workflows and nodes such as a CORINA node for 3D structure generation and structure cleanup, a MOSES descriptor node for the calculation of molecular properties, a fragment library, a CSRML editor for the implementation of chemotypes.

Furthermore, an open source version of the cell growth and toxicity model developed within COSMOS is being implemented, embedded in a KNIME workflow. This model is based on the Dynamic Energy Budget (DEB) theory and simulates high throughput screening in laboratory toxicity experiments.

KNIME Workflows.


Enoch SJ, Ellison CM, Schultz TW, Cronin MTD (2011) A review of the electrophilic reaction chemistry involved in covalent protein binding relevant to toxicity. Crit. Rev. Toxicol. 41: 783-802

Enoch SJ, Cronin MTD (2012). Development of new structural alerts suitable for chemical category formation for assigning covalent and non-covalent mechanisms relevant to DNA binding. Mutat. Res. 743: 10-19


This is European Comission Project. Seventh Framework Programme. Cosmetics Europe, The Personal Care Association. COLIPA, The European trade association for the cosmetic, toiletry and perfumery industry.