Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Development of a QSAR model for respiratory irritation

: Wehr, M.; Karwath, A.; Sarang, S.S.; Rooseboom, M.; Boogaard, P.J.; Escher, S.E.

Naunyn-Schmiedebergs archives of pharmacology 392 (2019), Supplement 1, pp.S67
ISSN: 0028-1298
ISSN: 1432-1912
German Society for Experimental and Clinical Pharmacology and Toxicology (DGPT Annual Meeting) <85, 2019, Stuttgart>
Association of the Clinical Pharmacology Germany (VKliPha Annual Meeting) <21, 2019, Stuttgart>
Fraunhofer ITEM ()

The RespiraTox project, funded by NC3R CrackIT, develops a QSAR model, which aims to predict the potential of individual compounds to cause irritation in the respiratory tract. We distinguished two mode of actions, i) "sensory irritation", characterized by a decrease in breathing rates, and ii) "tissue irritation" characterized by primarily histopathological findings. If possible, the QSAR model will lead to a reduction of de novo animal testing.
We based the classification "irritating to respiratory tract" on several in vivo studies with inhalation exposure. In a tiered approach, we considered information from studies with acute exposure from the ECHA CHEM database; the Hazardous Substance database; the harmonized classification and labelling inventory from ECHA, and repeated dose studies from the Fraunhofer RepDose database. Prior to model development, the CAS numbers and compound structures were quality controlled and corrected. Multiple features were generated from these structures: a) structural descriptors (ECFPS), and b) physico-chemical properties. We explored several machine learning algorithms including Logistic Regression, Random Forests, and Gradient Boosted Decision Trees to derive a classification model. The internal validation procedure employed stratified k-fold cross-validation (k=5).
The final dataset includes about 2500 irritating and 800 non-irritating individual compounds. The criteria for success of a given model is the measured Area Under ROC-Curve (AUC). The AUC for Logistic Regression is 0.71 using the combined feature set and 0.78 for both Random Forest and Gradient Boosted Decision Trees. The applicability domain is determined by features with highest impact on the final model. We additionally investigated five read-across groups within the training set to better understand the performance of the model. Each read-across group represents a set of structurally similar compounds, with a common toxicological mode of action. Within these groups, predicted and experimental values are compared.
The developed QSAR model predicts respiratory irritation with reasonable accuracy. To promote its use and acceptance, the final model can be accessed online via an user-friendly interface. The tool provides, a prediction for untested compound and in addition shows nearest neighbours with their experimental data from the training set. Our current approach will be further refined and improved e.g. by differentiating sensory and tissue irritation.