Test set: 100 complexes

To evaluate the strengths and limitations of GEMDOCK, we tested the program on a highly diverse dataset of 100 protein-ligand complexes (Tables 1 and 2) proposed by Jones et al. The overall accuracy of GEMDOCK in predicting the docked ligand conformations of 100 test complexes is shown in Table 2. The results generated by GEMDOCK are compared with those of other methods in Tables 3. All results are derived from 10 independent docking runs, and the docked lowest-energy structure was considered for each test case. On average, GEMDOCK took 305 seconds for a docking run on a Pentium 1.4 GHz personal computer with a single processor. The maximum time was 883 seconds for 1rne and the shortest time was 102 seconds for 2pk4.


Table 3: Comparing GEMDOCK with GOLD and FlexX on the dataset of 100 complexes
The success rate of GEMDOCK was based on solutions having the first rank.
a The success rate of GOLD, a steady-state genetic algorithm, was based on solutions having the first rank.
b The success rate of FlexX, a fragment-based approach, was based on the solutions having any rank on a dataset of 200 complexes extended from the GOLD data set.


[Go Back]