Kathrin S. Zeller, Andy Forreryd, Tim Lindberg, Ann-Sofie Albrekt, Aakash Chawade, Malin Lindstedt
Dept. of Immunotechnology, Lund University, Lund, Sweden; Swedish University of Agricultural Sciences, Alnarp, Sweden
The GARD assay is a cell-based transcriptional biomarker assay for the prediction of chemical ensitizers1 targeting key event 3, dendritic cell activation, of the skin sensitization AOP. Here, we present a modified assay based on Random Forest modelling, which is capable of predicting CLP
potency classes (1A – strong sensitizers, 1B – weak sensitizers, no category – non-sensitizers) as described by the European CLP regulation with an accuracy of 75 % (no cat), 75 % (1B) and 88 % (1A) based on a test set consisting of 18 chemicals previously unseen to the model.
We further can link the activation of distinct pathways to the chemical protein reactivity, showing that our transcriptomic approach can reveal information contributing to the understanding of underlying mechanisms in sensitization.
Results and Discussion
We here present a potency prediction approach based on a Random Forest model and 18 transcripts. 18 chemicals previously unseen to the model were classified as shown in Tables 1, 4 and Fig. 1. Interestingly, diethyl maleate, misclassified as 1A instead of 1B, is a human potency class 2 according to4, and iodopropynyl butylcarbamate, wrongly predicted as 1B instead of 1A, is classified as human potency class 44. Thus, the model seems to show more agreement with human data than CLP classifications (mainly derived from animal data) based on this limited dataset. Also Fig. 1C supports the hypothesis, that both data and model contain information allowing the prediction of human potency.
Furthermore, Key Pathway Advisor analysis reveals that these data can be used to investigate the cellular response in more detail (Table 3). In conclusion, we show that the modified GARD assay is capable of providing potency information, which is imperative for quantitative risk assessment of chemical sensitizers.