Identi fication of Pockets on Protein Surface to Predict Protein–Ligand Binding Sites - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 33

to each other (distance threshold 5 Å), they are defined to share the same RBS.
Here we define that one RBS is predicted correctly if it is located at the identified
pocket sites, i.e. any atom of the ligand is within 4 Å to the mass center of this
pocket. We also define that a prediction is a hit if at least one RBS in the given
protein is detected correctly in a certain number of top predictions. The top 1 to top
3 identified pocket sites from metaPocket and other methods are evaluated separately
in this work. Thus, to compare the performance of different approaches quantitatively,
the Success Rate (SR) is calculated according to the following formula:

N

HIT

Success _ Rate =

(2.2)

N

P

Where N P is the total number of proteins in the dataset; N HIT is the total number
of hit prediction. The success rate is calculated for all the methods for the top 1, top
2 and top 3 predictions, respectively.

2.4.1 Test Datasets

In the evaluation step, different datasets are being used, including 48 bound/unbound
protein-ligand complexes, 210 bound protein-ligand complexes and 198 drug-target
complexes. These datasets are described in details in the relevant publications (Huang
2009 ; Zhang et al. 2011 ; Huang and Schroeder 2006 ). For the bound protein-ligand
complexes, first the ligands are removed and only the protein structures are input for
pocket identification. Then the ligands are put back for success rate calculation. For
the unbound prediction, the unbound protein structures are input for pocket
identification and then aligned to bound protein structures. The ligands bound in the
bound proteins are then used to calculate the success rate. The detailed description
and the PDB structures of these three data-sets can be freely downloaded from
metaPocket web-server http://projects.biotec.tu-dresden.de/metapocket .

2.5 Results

In this section, we will describe the prediction results of metaPocket, as well as the
results for those eight single individual methods.
Table 2.2 shows the success rates for metaPocket and the eight single methods for
these three datasets. Overall, metaPocket archived better result than each of the eight
single methods. In the top1 and top2 prediction for drug-target data-set, LIGSITE cs
performed best among the eight single methods and metaPocket increased the
success rate by 13%. In the top3 predictions, Q-SiteFinder is the best method and
metaPocket also has 12% improvements. The reason why metaPocket improves
the success rate is that it takes the overlapping prediction results from different