You are here:
Multivariate models for skin sensitization hazard and potency
Citation:
Strickland, J., Q. Zang, M. Paris, D. Lehmann, N. Choksi, S. Bell, D. Allen, J. Matheson, A. Jacobs, W. Casey, AND N. Kleinstreueur. Multivariate models for skin sensitization hazard and potency. OpenTox, Durham, NC, July 12 - 13, 2017.
Impact/Purpose:
One of the top priorities being addressed by ICCVAM is the identification and validation of non-animal alternatives for skin sensitization testing. This work suggest that computational methods developed using in vitro and in silico data are promising tools to effectively classify potential skin sensitizers without animal testing.
Description:
One of the top priorities being addressed by ICCVAM is the identification and validation of non-animal alternatives for skin sensitization testing. Although skin sensitization is a complex process, the key biological events have been well characterized in an adverse outcome pathway (AOP) proposed by OECD. Accordingly, ICCVAM developed defined approaches to skin sensitization hazard or potency assessment based on the OECD AOP that use in vitro, in chemico, and in silico information. Data were collected for 120 substances tested in the murine local lymph node assay (LLNA), direct peptide reactivity assay, human cell line activation test, and KeratinoSens assay. Data for six physicochemical parameters (octanol:water partition coefficient, water solubility, vapor pressure, molecular weight, melting point, and boiling point) were also collected. In silico read-across predictions of skin sensitization hazard were generated using QSAR Toolbox. Data from substances with human skin sensitization hazard and potency information were used to develop models to predict human responses. The in vitro, in chemico, in silico, and physicochemical data were used in various combinations as inputs to multiple machine learning approaches to predict LLNA and human skin sensitization hazard or three categories of potency (GHS 1A, strong sensitizers; 1B, weak sensitizers; or nonsensitizers). Each model was trained on a training set of substances and then tested on an external validation set and via leave-one-out cross-validation. The support vector machine (SVM) machine learning approach had the highest performance for skin sensitization hazard and potency for both LLNA and human endpoints. The SVM models performed better than any in chemico, in vitro, or in silico method alone. SVM models performed better than the LLNA in predicting human outcomes. These results suggest that computational methods developed using in vitro and in silico data are promising tools to effectively classify potential skin sensitizers without animal testing. This project was funded with U.S. Federal funds from NIEHS/NIH/HHS under Contract HHSN27320140003C. This abstract does not represent U.S. EPA policy or the policy of any federal agency.