In silico models
have been considered as a promising
tool that can be used to screen and identify potential endocrine disrupting
chemicals (EDCs) effectively and assist finally realizing the sustainable
assessments and management of EDCs. However, most of the current modeling
efforts are focused on constructing predictive models for the end
points related to nuclear receptors. Considerable actions are needed
to model the interfering effects of compounds on the nonreceptor mediated
targets, e.g., hormone transporters. Here, a tiered screening approach
was proposed and derived based on classification models and quantitative
structure–activity relationship (QSAR) models for identifying
potential disruptors of human transthyretin (hTTR), a critical thyroid
hormone transporter. Regarding classification models, the sensitivity,
specificity, and predictive accuracy of all five optimum models for
corresponding training sets and validation sets were >0.800, implying
that they had good classification performance. Concerning quantitative
prediction, k-nearest neighbor and multiple linear regression QSAR
models were developed. The statistical parameters of all constructed
QSAR models met the acceptable criteria (e.g., leave-one out cross
validation Q
2 (Q
2
LOO) and bootstrapping coefficient (Q
2
BOOT) > 0.600, determination coefficient
(R
2) > 0.700, externally explained
variance (Q
2
EXT), concordance
correlation coefficient
(CCC) > 0.850), indicating that those QSAR models had good robustness,
goodness-of-fit, and external prediction ability. Finally, an extra
data set with 38 data was used to evaluate the predictive performance
of the tiered approach. The assessment results showed that more than
65% of compounds can be classified correctly. Thus, the tiered approach
proposed here can be employed to distinguish whether a given compound
within the applicability domain of corresponding models is a potential
hTTR disruptor or not.