You are here:
QSAR modeling for in vitro human NIS inhibition with blinded external validation and screening of 80,086 substances.
Citation:
Martins, A., T. Stoker, J. Wang, N. Nikolov, AND E. Wedebye. QSAR modeling for in vitro human NIS inhibition with blinded external validation and screening of 80,086 substances. QSAR2021 Modeling, QSAR to NAMS (virtual), June 01 - 07, 2021.
Impact/Purpose:
To develop global binary QSAR models using ToxCast HTP data (P1 and II and tested in E1K) that can be applied for screening purposes and single-compound identification of possible NIS antagonists.
Description:
Inhibition of the sodium/iodide symporter (NIS) can lead to learning and memory impairment in humans and rats (AOP 54 under development). The main target of this study was to develop global binary QSAR models that can be applied for screening purposes and single-compound identification of possible NIS antagonists. For this purpose, we processed the HTS assays results from U.S EPA’s ToxCast Program phases I and II for NIS inhibition to develop the first QSAR model for this endpoint adopting a new curation procedure including tautomer treatment and accounting for volatility and lipophilicity, resulting in a training set of 579 substances (64 actives and 515 inactives). Two models were developed and robustly cross-validated, one with high sensitivity and another with high overall accuracy. The models were subsequently subjected to external validation with ToxCast NIS inhibition results blinded to the QSAR developers for 7401 E1K substances. The external validation set underwent the same processing as the training set. Next, the training set was expanded with the E1K dataset and two final models were developed and cross-validated, applying the same methods as for the first versions. The final models were used to screen 80,086 REACH substances for NIS inhibition. These QSAR predictions will be published in the free online Danish (Q)SAR Database (https://qsar.food.dtu.dk). Furthermore, the final models will be published in the free online Danish (Q)SAR Models, accessible from the Danish (Q)SAR Database, for real-time prediction of user-submitted structures and download of detailed results in the QSAR Prediction Reporting Format.