The Center for Alternatives to Animal Testing is an academic center affiliated with the Division of Toxicological Sciences in the Department of Environmental Health Sciences of the Johns Hopkins University Bloomberg School of Public Health.

 

Johns Hopkins School of Public Health

Commemorative Booklet for CAAT 20th Anniversary Symposium

Predicting Allergic Contact Dermatitis (ACD) BIOSAR I & BIOSAR II QSAR Models

Jurij J. Hostynek
Euroamerican Technology Resources, Inc., Lafayette, California

Our goal was to develop a mathematical model for the prediction of the skin sensitization potential of chemicals. Our BIOSAR models combine features of both SAR and QSAR. The models thus link chemical structure elements (indicator variables such as functional groups, ring systems or multiple bonds) and measured or calculated physico-chemical properties (e.g., partition coefficient logP; molar refraction MR) with biological properties.

Initially we analyzed sensitization data both from human and from several animal models. When the limitations of a mixed-species or guinea pig-based model became apparent, further model development focused exclusively on human data, leading to development of BIOSAR H. Thus, comprehensive algorithms were derived and improved stepwise through multivariate or two-class multiple regression analysis. Class I identifies allergens, class 0 non-allergens, and multiple regression is performed against these variables.

The classes incorporate indicator variables for structural features relating to barrier transport, protein reactivity and metabolism, all prerequisites for the factual description of eventual tissue reactions. The processes of membrane diffusion and binding are determined by mechanistic descriptors of size, lipophilicity and hydrogen bonding, expressed as continuous variables that also become part of the computation process. The dichotomous or indicator descriptors account for reactivity and metabolism, an aspect that as yet is not yet adequately accounted for in current expert systems. In each step of the study, through the process of backward selection, the least significant variables were deleted until only the statistically relevant ones remained.

The BIOSAR I model, developed using two-value regression analysis, also was validated in vivo in a double-blind study by matching its prediction against sensitization data obtained in mouse studies.

In the classification of 50 common fragrance chemicals analyzed as a test set of human data exclusively, BIOSAR II, a refinement of BIOSAR I, correctly predicted 44, an 88% concordance. All 6 incorrect predictions were false negatives, rated from human experience to be allergens. In this case, the BIOSAR II model performs better in the identification of non-allergens, with a specificity (predicted negatives/true negatives) of 100%, than it does on allergens, where sensitivity (predicted positives/true positives) is 83%.