A review of quantitative structure-activity relationship modelling approaches to predict the toxicity of mixtures

Abstract

Exposure to chemicals generally occurs in the form of mixtures. However, the great majority of the toxicity data, upon which chemical safety decisions are based, relate only to single compounds. It is currently unfeasible to test a fully representative proportion of mixtures for potential harmful effects and, as such, in silico modelling provides a practical solution to inform safety assessment. Traditional methodologies for deriving estimations of mixture effects, exemplified by principles such as concentration addition (CA) and independent action (IA), are limited as regards the scope of chemical combinations to which they can reliably be applied. Development of appropriate quantitative structure-activity relationships (QSARs) has been put forward as a solution to the shortcomings present within these techniques – allowing for the potential formulation of versatile predictive tools capable of capturing the activities of a full contingent of possible mixtures. This review addresses the current state-of-the-art as regards application of QSAR towards mixture toxicity, discussing the challenges inherent in the task, whilst considering the strengths and limitations of existing approaches. Forty studies are examined within – through reference to several characteristic elements including the nature of the chemicals and endpoints modelled, the form of descriptors adopted, and the principles behind the statistical techniques employed. Recommendations are in turn provided for practices which may assist in further advancing the field, most notably with regards to ensuring confidence in the acquired predictions.