Skin Doctor
Prediction of skin sensitization potential

Set a decision threshold (optional)

Decision threshold for the random forest model (classifier)
Decision threshold for the support vector machine (classifier)

Provide input molecules:

Enter SMILES

Example: CCOC(=O)N1CCN(CC1)C2=C(C(=O)C2=O)N3CCN(CC3)C4=CC=C(C=C4)OC

or upload a file with a list of SMILES
or upload an sdf file
or draw your own molecule

Skin Doctor

Skin Doctor features two machine learning models for classifying compounds of interest into skin sensitizers and non-sensitizers. The classifiers are trained on more than 1400 compounds annotated with local lymph node assay (LLNA) data. One of the models is a random forest model based on MACCS fingerprints ("RF_MACCS"); the other is a support vector machine based on PaDEL descriptors ("SVM_PaDEL"). Users may adjust the decision thresholds applied to the predictions. For example, lowering the decision thresholds will increase the sensitivity of the models, which can be useful in scenarios where any skin sensitization should be ruled out. The result page reports detailed information on class probabilities, the applicability domain of the models and additional indicators of the reliability of individual predictions.

Skin Doctor has been developed in collaboration with Beiersdorf AG. The web service is free for non-commercial and academic research purposes only.

How to cite

If you are using Skin Doctor for your research, please cite:

Wilm, A.; Stork, C.; Bauer, C.; Schepky, A.; K├╝hnl, J.; Kirchmair, J. Skin Doctor: Machine Learning Models for Skin Sensitization Prediction that Provide Estimates and Indicators of Prediction Reliability. Int. J. Mol. Sci. 2019.
doi:10.3390/ijms20194833

Selecting decision thresholds

Skin Doctor allows users to adjust the threshold above which a compound is predicted to be a skin sensitizer. Higher decision thresholds increase the probability for a compound being classified as a non-sensitizer and vice versa. The default values for the decision threshold are 0.5 for the random forest model and 0.0 for the support vector machine model. We recommend using these thresholds for any standard use cases.