Best computational docking program.

In a recent study conducted at Johnson & Johnson Pharmaceutical Research and Development Lab, researchers set out to compare the accuracy of 5 docking programs currently on the market: DOCK, FlexX, Glide, GOLD and LigandFit.

The researchers chose 69 protein-ligand complexes from 14 receptor families with a wide spectrum of characteristics. The proteins varied in their crystallographic resolution and represented a range of active site topologies and water accessibilities. Some active sites contained metals in the binding pocket. In addition, the receptor-bound ligands varied significantly in their flexibility and activity ranging from low micromolar to nanomolar.

Program	Number of results
	High- accuracy	Medium- accuracy	Inaccurate
Glide	39	24	5
GOLD	47	13	8
LigandFit	24	35	10
FlexX	26	18	25
DOCK	4	19	45

The authors ranked the poses found by the docking programs in terms of accuracy in three broad categories. In the highest-accuracy category, there was at most minimal displacement of only one functional group per ligand. Poses of medium-accuracy had a few functional groups displaced per ligand. The lowest-accuracy, or inaccurate, poses resulted when docking programs were either unable to reproduce the binding mode or failed to position the ligand in the binding site completely.

In terms of docking accuracy, Glide and Gold were observed to outperform the other programs. Glide was seen to have the least number of inaccurate poses and 85% of Glide's binding modes had an rms deviation of 1.4?or less from the native co-crystallized structures.

However, even more important than the programs' ability to reproduce the correct binding modes is the ability of each program to correctly score and therefore rank the predicted poses. In order for computational docking programs to be of practical use in virtual screening, they must enrich the fraction of suitable lead candidates in a chemical database, thus contributing to enhanced productivity and reducing drug development costs. In this regard, Glide is seen to be well ahead of the competition, nearly outperforming the next-best, GOLD, by two-fold.

It was further observed that of Glide's 39 high-accuracy poses, 26 of them were ranked among the top six by their scores, a rate of 67%, leading the authors to conclude that Glide is the best scoring tool.

Program	No. of times the native binding mode was found as the top-scoring pose
Glide	17
GOLD	10
LigandFit	9
FlexX	7
DOCK	0

Glide was found to work well with both hydrophilic and lipophilic active sites, and exhibited less sensitivity than other programs to the complexity of the ligand, achieving a high success rate with molecules having 15 or more rotatable bonds. Glide's performance across all protein families led the researchers to conclude that Glide works well across a wide variety of systems.

The authors concluded that Glide's ability to accurately reproduce experimental binding modes and superior scoring make it an effective tool for lead optimization.

"Evaluation of Docking Performance: Comparative Data on Docking Algorithms," J. Med. Chem., Kontoyianni, M.; McClellan, L. M.; Sokol, G. S., 2004; 47(3); 558-565.