“2 out of 3” weight of evidence Summary of defined approach

This method uses a weight of evidence approach by taking a majority of two congruent results of the first 3 key events (represented by the results of the DPRA, KeratinoSens™, and h-CLAT assays). This method returns a simple prediction of the compound being a sensitiser or non-sensitiser. Our implementation of this model performs an overall assessment using a 2 out of 3 majority voting approach, where if 2 out of the 3 assays are positive for a tested compound, then the substance is rated to be a skin sensitiser, and a substance is rated as a non-sensitiser if 2 out of the 3 tests are negative. The details of how each test is deemed positive or negative are described in the corresponding OECD guidelines for each of the 3 in vitro tests [1-3]. The accuracy of this model (based on the dataset of 213 compounds from the Bausch et al [4]) is 79% when compared to LLNA and 90% when compared to human data (see Table 1 below).

Table 1. Cooper statistics for the “2 out of 3” majority voting method compared to the LLNA or human reference data, respectively.

	Compared to LLNA data	Compared to human data
Accuracy	79%	90%
Sensitivity	82%	90%
Specificity	72%	90%
Positive predictive value	89%	96%
Negative predictive value	59%	79%

Experimental values used in SaferSkin™

Input parameters	Provided by user	Calculated by application
Experimental values
DPRA assay: DPRACys, DPRALys
KeratinoSens™ assay: KEC1.5
h-CLAT: CD54 and CD86 (positive/negative)