Qualitative structure-toxicity relationships (QSTR) on skin sensitization

Abstract: QSTR (Qualitative Structure-Toxicity Relationships) approach has been applied on skin sensitization research using 593 compounds in which 419 positive sensitizers and 174 negative sensitizers are included. Criteria of positive compounds were based on human epidemiological studies. Negative data were collected from a database developed by the Japanese Globally Harmonized System of Classification and Labeling of Chemicals (GHS) Inter-ministerial Committee. Perfect classification between positive and negative samples could not been achieved by conventional data analysis methods such as Linear learning machine (LLM), Neural networks (NN), Linear discriminant analysis by least squares algorithm, K-Nearest neighbor (K-NN) method, Support vector machine (SVM) and AdaBoost. By contrast, the KY (K-step Yard sampling) method could accomplish a perfect classification