Structure activity relationship (SAR) toxicological assessments: The role of expert judgment

Structure activity relationships (SAR) and read-across are widely used animal alternative approaches for filling toxicological data gaps. A framework describing the use of expert judgment in evaluating analogs for SAR has been published and widely cited, however, reliance on expert judgment can introduce inconsistent results across experts and hinder transparency. Here we explore the use of a quantitative similarity score between an analog and a Structure of Interest (SOI) to see if these scores correlate with the expert judgement-based suitability rankings. We find these global similarity scores representing a whole-molecule view of similarity to be insensitive to differences in local structure which may be important for toxicity, and, therefore, cannot be substituted for expert judgement-based similarity rankings. In this paper, we suggest that the next step in the progression of SAR approaches retains the insights from expert judgment, but facilitates consistency and transparency through the development of rating rules. This report outlines and defines analog rating rules for several compound categories. While not comprehensive, the exercises demonstrate the development of rules for categories with a large spread in molecular weight and alkyl chain length and explains the advantages that we see in this approach compared to relying solely on a computational approach or an unstructured expert judgement approach. These rules may be incorporated into analog searching work flows to define boundaries for analogs suitable for read-across.