Summary of results for human liver microsomal models generated with very large training sets and different molecular descriptors
| HLM Model with CDK and SMARTS Keys | HLM Model with MOE2D and SMARTS Keys | |
|---|---|---|
| No. of descriptors | 578 | 818 |
| No. of training set compounds | 193,650 | 193,930 |
| Cross-validation results | 38,730 compounds | 38,786 compounds |
| Training R2 | 0.79 | 0.77 |
| 20% test set R2 | 0.69 | 0.69 |
| Blind data set (2310 compounds) | ||
| R2 | 0.53 | 0.53 |
| RMSE | 0.367 | 0.367 |
| Continuous and categorical | ||
| κ | 0.40 | 0.42 |
| Sensitivity | 0.16 | 0.24 |
| Specificity | 0.99 | 0.987 |
| Positive predictive value | 0.80 | 0.823 |
| Time (s/compound) | 0.252 | 0.303 |