Prospective Validation of a Comprehensive In silico hERG Model and its Applications to Commercial Compound and Drug Databases

Abstract

Ligand-based in silico hERG models were generated for 2 644 compounds using linear discriminant analysis (LDA) and support vector machines (SVM). As a result, the dataset used for the model generation is the largest publicly available (see Supporting Information). Extended connectivity fingerprints (ECFPs) and functional class fingerprints (FCFPs) were used to describe chemical space. All models showed area under curve (AUC) values ranging from 0.89 to 0.94 in a fivefold cross-validation, indicating high model consistency. Models correctly predicted 80 % of an additional, external test set; Y-scrambling was also performed to rule out chance correlation. Additionally models based on patch clamp data and radioligand binding data were generated separately to analyze their predictive ability when compared to combined models. To experimentally validate the models, 50 of the predicted hERG blockers from the Chembridge database and ten of the predicted non-hERG blockers from an in-house compound library were selected for biological evaluation. Out of those 50 predicted hERG blockers, tested at a concentration of 10 microM, 18 compounds showed more than 50 % displacement of [(3)H]astemizole binding to cell membranes expressing the hERG channel. K(i) values of four of the selected binders were determined to be in the micromolar and high nanomolar range (K(i) (VH01)=2.0 microM, K(i) (VH06)=0.15 microM, K(i) (VH19)=1.1 microM and K(i) (VH47)=18 microM). Of these four compounds, VH01 and VH47 showed also a second, even higher affinity binding site with K(i) values of 7.4 nM and 36 nM, respectively. In the case of non-hERG blockers, all ten compounds tested were found to be inactive, showing less than 50 % displacement of [(3)H]astemizole binding at 10 microM. These experimentally validated models were then used to virtually screen commercial compound databases to evaluate whether they contain hERG blockers. 109 784 (23 %) of Chembridge, 133 175 (38 %) of Chemdiv, 111 737 (31 %) of Asinex and 11 116 (18 %) of the Maybridge database were predicted to be hERG blockers by at least two of the models, a prediction which could, for example, be used as a pre-filtering tool for compounds with potential hERG liabilities.

ISIDA Property‐Labelled Fragment Descriptors

“…Their best model was able to achieve a TP = 0.57, a TF = 0.75 and a BA = 0.60. In another recent study, [45] hERG models were fitted on hand of 2644 compounds using ECPF of Daylight atomic invariants rule and functional pharmacophore role with linear discriminant analysis and support vector machines. The external validation on the PubChem data was only able to identify 78 blockers out of 193 corresponding to a TP rate of 0.40.…”

Novel Bayesian classification models for predicting compounds blocking hERG potassium channels

“…The work reported here used the hERG blockage data from Doddareddy and co-workers [12] , which contained 2644 compounds, including 1479 literature compounds tested for effects on the hERG channel and 1165 FDA-approved drugs. The dataset was divided into training and test sets as described [12] .…”

“…The dataset was divided into training and test sets as described [12] . Activity data were expressed as an IC 50 value, measured by either patch clamp or radioligand binding assays, which were conducted in human HEK-293 and CHO cell lines.…”

“…Extended-connectivity fingerprints were used to describe chemical space. The best SVM-ECFP_6 model showed 88% accuracy for the external test set, which contained 255 compounds [12] . In 2013, Wang et al reviewed recent developments in computational prediction of hERG blockage, and they proposed that more reliable experimental data and a consensus modeling strategy are required to improve the performance of current computational models [13] .…”

Investigation of miscellaneous hERG inhibition in large diverse compound collection using automated patch-clamp assay

“…This permits applications of automated electrophysiology in primary screening. In addition to experimental approaches, to predict the hERG liability of lead candidates early in drug discovery, numerical in silico methods have been developed [12][13][14][15][16][17][18] . Since experimental data from a large collection of structurally diverse compounds were yet available, models was usually based on the existing data of limited number of hERG inhibitors generated using different methods.…”