hERG (the human <i>Ether-à-go-go</i>-Related Gene) codes for the alpha subunit of one of the ion channels that conducts potassium out of the muscle cells of the heart. The current generated by the channel is crucial for cell membrane repolarization during the cardiac action potential. Inhibition of hERG channels leads to prolongation of the action potential and the QT interval, inducing heart arrhythmias and can result in the potentially fatal condition torsades de pointes. A number of marketed drugs inhibit the hERG channel, and it is an important safety pharmacology endpoint to evaluate in preclinical development. Experimental evaluation of the binding affinity of compounds to hERG channels is laborious and time consuming. Computational methods like QSAR allow one to rapidly and accurately identify potential inhibitors of the hERG channels, but most of the developed models are based on heterogeneous data coming from different biological assays. We have analyzed the level of data uncertainty produced by different hERG assays reported in ChEMBL and generated in house. Mix-and-match analysis of different assays allows us to pool data to ensure high quality modeling results. Correlation and balanced accuracy values between different assays data are ranging 0.35-1 and 0.43-1, respectively. In this study, we have developed and rigorously validated QSAR models for the prediction of hERG inhibition. Different strategies to efficiently mix-and-match assay data were investigated and compared for building predictive QSAR models. As result, we have generated a variety of QSAR models using different statistical methods: Random Forest, Generalized Boosted Machine, Deep Learning; and descriptor sets: QNA (Quantitative Neighbourhoods of Atoms) descriptors, RDkit descriptors, Dragon descriptors. The obtained models were compared with publicly available ones using external validation sets. The most accurate models were used to predict hERG inhibitors from NCATS screening libraries and obtained results were experimentally validated. We also report on our efforts to incorporate our most predictive model as well as experimental results in the publicly available NCATS web services (http://tripod.nih.gov/adme).

QSAR modeling of hERG inhibitors using a mix and match approach

252nd National Meeting (2016)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.