Application of the random forest method in studies of local lymph node assay based skin sensitization data.

The random forest and classification tree modeling methods are used to build predictive models of the skin sensitization activity of a chemical. A new two-stage backward elimination algorithm for descriptor selection in the random forest method is introduced. The predictive performance of the random forest model was maximized by tuning voting thresholds to reflect the unbalanced size of classification groups in available data. Our results show that random forest with a proposed backward elimination procedure outperforms a single classification tree and the standard random forest method in predicting Local Lymph Node Assay based skin sensitization activity. The proximity measure obtained from the random forest is a natural similarity measure that can be used for clustering of chemicals. Based on this measure, the clustering analysis partitioned the chemicals into several groups sharing similar molecular patterns. The improved random forest method demonstrates the potential for future QSAR studies based on a large number of descriptors or when the number of available data points is limited.

Structure-activity models for contact sensitization.

Allergic contact dermatitis (ACD) is a widespread cause of workers' disabilities. Although some substances found in the workplace are rigorously tested, the potential of the vast majority of chemicals to cause skin sensitization remains unknown. At the same time, exhaustive testing of all chemicals in workplaces is costly and raises ethical concerns. New approaches to developing information for risk assessment based on computational (quantitative) structure-activity relationship [(Q)SAR] methods may be complementary to and reduce the need for animal testing. Virtually any number of existing, de novo, and even preconceived compounds can be screened in silico at a fraction of the cost of animal testing. This work investigates the utility of ACD (Q)SAR modeling from the occupational health perspective using two leading software products, DEREK for Windows and TOPKAT, and an original method based on logistic regression methodology. It is found that the correct classification of (Q)SAR predictions for guinea pig data achieves values of 73.3, 82.9, and 87.6% for TOPKAT, DEREK for Windows, and the logistic regression model, respectively. The correct classification using LLNA data equals 73.0 and 83.2% for DEREK for Windows and the logistic regression model, respectively.