Exploring in silico tools to predict estrogen receptor activity of chemicals for the assessment of endocrine disruption

In silico software and tools are increasingly being employed as an alternative to in vivo animal testing to predict toxicity of chemicals. One particular application of the underlying in silico models for hazard assessment has been to predict the potential endocrine disrupting activity of chemicals, which is one of the three fundamental elements of an endocrine disrupting chemical (EDC). In this study, 11 in silico tools based on methods ranging from Quantitative Structure-Activity Relationship (QSAR) to docking were selected and tested for their predictivity of estrogen receptor (ER) activity using a set of 80 chemicals of known ER activity potential. The accuracy in prediction, as determined by Matthew’s correlation coefficient (MCC), among the 11 individual tools tested ranged from 0.16 to 0.54 (min–max). However, when combining various tools and applying rules set for a conservative approach in assessing the prediction outcomes, the MCC increased as high as 0.68, demonstrating the higher probability of generating a correct prediction when multiple in silico tools are employed. This study presents the strengths and weaknesses of the individual tools/models tested and provides insights on how in silico predictions could supplement the weight-of-evidence approach in determining endocrine activity potential of chemicals.