Download the data (CSV and SD files) which were used for training of the models
The data on chemical structures and quantitative values of endpoints (the percentage of inhibition (inhibition %); 50% inhibitory concentration, IC50; inhibition constant, Ki) of compounds interacting with hERG channels were extracted from the ChEMBL database (version 24). A median value was calculated for the compounds having multiple values for appropriate endpoints.
The results of the quantitative prediction for the pIC50 and pKi calculated by 5-fold CV.
End-point | Number of compounds | Interval of values | Mean | R2 | RMSE | AD, % |
---|---|---|---|---|---|---|
pIC50 | 4987 | [2.2:9.9] | 5.4 | 0.551 | 0.602 | 97.9 |
pKi | 1180 | [3.9:9.6] | 5.9 | 0.574 | 0.608 | 99.4 |
The results of SAR prediction at 5-fold CV for classification exact and inexact data of IC50, Ki and Inhibition % values.
End-point | Number of active/inactive compounds | Sensitivity | Specificity | BA | AD, % |
---|---|---|---|---|---|
IC50 , nM | 1246/6257 | 0.812 | 0.820 | 0.816 | 99.9 |
Ki , nM | 511/969 | 0.798 | 0.777 | 0.787 | 100.0 |
Inhibition, % | 1976/791 | 0.765 | 0.777 | 0.771 | 100.0 |