As cricket fans well know, one of the toughest deliveries to face in cricket is a "googly", where the ball's movement is discontinuous and difficult to predict, posing a challenge for even the most experienced batsmen.
The fraction of entries that have correctly predicted poses within the top-10 ranked poses was 92%.
This is frequently encountered in docking calculations, where this step has to be done multiple times to evaluate predicted poses.
In order to grant a thorough assessment of hydrogen bonds, thereby, the predicted poses are forwarded to an efficient post-optimization of the hydrogen-bond network.
With predicted poses from image features, we map the problem from image feature space to pose space, where a Bag of Poses (BOP) model is learned for the final goal of HAR.
When using Omega generated low-energy conformers as docking input the fraction of entries with correctly predicted poses at the top-1 position increased significantly from 10.9% without MC optimization to 39.1% with MC optimization.
Predictions had a root-mean-square error of 1.95 kcal/mol to the experimental affinities, and predicted poses had an average RMSD of 1.7 Å to the crystallographic poses.
When the native conformer was used as docking input, the fraction of entries that have correctly predicted poses (RMSD ≤ 2 Å to the native binding pose) at the top-1 position was nearly doubled using MC optimization compared to pharmacophore matching only (Figure 6A).
Thus, eHiTS predicted poses for the ADP and pyridone 6 with root-mean-square deviations relative to their experimental poses in 2CN5 and 2J90 of 1.23 and 0.38 Å, respectively.
Ranked by the docking score, the top five predicted poses are almost identical.
