Development and experimental validation of 3D QSAR models for the screening of thyroid peroxidase inhibitors using integrated methods of computational chemistry

The intricate network of glands and organs that makes up the endocrine system. Hormones are used to regulate and synchronize the nervous and physiological systems. The agents which perturbate an endocrine system are called endocrine disruptors and they can eventually affect cellular proliferation and differentiation in target tissues. A subclass of endocrine disruptors known as thyroid disruptors (TDs) or thyroid disrupting chemicals (TDCs) influence the hypothalamo-pituitary-thyroid axis or directly interfere with thyroid function by binding to thyroid hormone receptors. Thyroid hormone levels in circulation are now included in more test guidelines (OECD TG 441, 407, 408, 414, 421/422, 443/416). Although these might be adequate to recognize thyroid adversity, they are unable to explain the underlying mechanism of action. Thyroid peroxidase (TPO) and sodium iodide symporter (NIS), two proteins essential in the biosynthesis of thyroid hormones, are well-accepted molecular targets for inhibition. The screening of a large number of molecules using high throughput screening (HTS) requires a minimum quantity of sample, cost, and time consuming. Whereas 3-dimensional quantitative structure-activity relationship (3D-QSAR) analysis can screen the TDCs before synthesizing a compound. In the present study, the human TPO (hTPO) and NIS (hNIS) structures were modelled using homology modeling and the quality of the structures was validated satisfactorily using MD simulation for 100ns. Further, 190 human TPO inhibitors with IC50 were curated from Comptox and docked with the modelled structure of TPO using D238, H239 and D240 centric grid. The binding conformation of a molecule with low binding energy was used as a reference and the rest other molecules were aligned after generating the possible conformers. The activity-stratified partition was performed for aligned molecules and training set (139), test set (51) were defined. The machine learning models such as k Nearest Neighbor (kNN) and Random Forest (RF) models were built and validated using external experimental dataset containing 10 molecules. Among the 10 molecules, all 10 molecules were identified as TPO inhibitors and demonstrated 100 % accuracy qualitatively. To confirm the selective TPO inhibition all 10 molecules were docked with the modelled structure of hNIS and the results have demonstrated the selective TPO inhibition.