Unforeseen toxicity comprises one of the main reasons for failures in the drug development process. In order to assess the safety risk as early as possible in the development pipeline, in silico methods play an increasingly important role. However, prospective assessment of toxicity also requires experimental data, which often serve as basis for toxicological read across. In this approach, toxicologists query databases and the available literature for compounds which are structurally similar to their development candidate in order to retrieve information on potential toxicity endpoints. <br/>The result of a read across of course heavily depends on the similarity metrics used. Example queries using a set of selected drugs and drug classes such as benzodiazepines clearly demonstrate the high dependency of the search results on the descriptor set used. Using KNIME to derive compound-target interaction fingerprints and expanding them towards consensus fingerprints retrieved for similar compounds provides another possibility for similarity measures. To further expand the compound-target interaction we both implemented the machine learning based models provided by ChEMBL as well as a KNIME workflow which retrieves data from the public domain and our internal Jupyter notebook which automatically builds classification models for all targets where sufficient data are available. The combined similarity searching/machine learning based prediction is implemented in ToxPHACTS v2 as an integrated way for toxicological read across.

Integrated toxicological read across – combining similarity searching with machine learning

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

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