Evaluation metric of docking experiments

The docking performance is evaluated mainly by two metrics. The first is the root-mean-square deviation (RMSD) of the predicted ligand conformation relative to the native structure:

where xip,yip,zip and xie,yie,zie are, respectively, the coordinates of ith heavy atom in the predicted model and experimental structure of the ligand. The second metric is the distance between the geometric centers of the predicted and experimental structures of the ligand.

Both RMSD and center distance can be directly computed from the coordinates of the ligands, if the receptor is from the native structure and is kept unchanged during simulations. In case that the receptor structure is created from protein structure prediction (or from the native but with its orientation changed in the final output), the receptor model will be first superimposed on the target receptor structure by TM-score [26]. The RMSD and center distance are then calculated after the ligand model is superimposed onto the receptor structures based on the same TM-score rotation matrix.

Here, it is noted that Eq. (6) based on the default atom order can result in artificially high RMSD values for ligands with symmetric structures (such as a benzene ring). We use the DockRMSD program, which was designed to identify the minimum RMSD values by a quick graph isomorphism searching algorithm [27], to evaluate the symmetry-corrected RMSD between the docking pose and native ligand conformation.