Predicting the Skin Sensitization Potential of Small Molecules with Machine Learning Models Trained on Biologically Meaningful Descriptors

Abstract

In recent years, a number of machine learning models for the prediction of the skin sensitization potential of small organic molecules have been reported and become available. These models generally perform well within their applicability domains but, as a result of the use of molecular fingerprints and other non-intuitive descriptors, the interpretability of the existing models is limited. The aim of this work is to develop a strategy to replace the non-intuitive features by predicted outcomes of bioassays. We show that such replacement is indeed possible and that as few as ten interpretable, predicted bioactivities are sufficient to reach competitive performance. On a holdout data set of 257 compounds, the best model ("Skin Doctor CP:Bio") obtained an efficiency of 0.82 and an MCC of 0.52 (at the significance level of 0.20). Skin Doctor CP:Bio is available free of charge for academic research. The modeling strategies explored in this work are easily transferable and could be adopted for the development of more interpretable machine learning models for the prediction of the bioactivity and toxicity of small organic compounds.

Keywords: bioactivity descriptors; conformal prediction; in silico prediction; machine learning; random forest; skin sensitization; toxicity prediction.

Figure 1

Schematic representation of the workflow for feature selection.

Figure 2

Mean performance of 10-fold CV as a function of the number of bioactivity descriptors selected for model building at the significance level (A) $\epsilon =0.05$, (B) $\epsilon =0.10$, (C) $\epsilon =0.20$, (D) $\epsilon =0.30$. The horizontal, dashed line indicates the validity expected from the selected significance level ε; the vertical, dashed line marks the performance of models trained on 10 descriptors.

Figure 3

PCA quantifying the coverage of the LLNA data by the reference sets of (A) cosmetics, (B) approved drugs, and (C) agrochemicals in the feature space of the 10 selected bioactivity descriptors. The percentages in parentheses report the variance explained by the respective principal component (PC).

Figure 4

LLNA data set analyzed by PCA in the feature space of the ten selected bioactivity descriptors. (A) Scatter plot colored by the binary skin sensitization potential; (B) loadings plot of the ten descriptors. The percentages in parentheses report the variance explained by the respective principal component (PC). Note that the axis sections differ for panels (A,B).

Figure 5

Architecture of the consensus model.

Figure 6

Relationship between MCC and coverage for the individual and the combined models.

Figure 7

Performance of the RF classifier (n_estimators = 500; all other parameters default) underlying the CP model as a function of the number of instances the model was trained on.