Prediction of Protein-Protein Binding Interfaces - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 126

Table 6.3 Comparison of correct and incorrect solutions, indicating the validity of
various models
Lowest →

Best ↓ FOD HADDOCK RosettaDock ZDOCK

FOD 101.30 71.63 71.25

7.88 7.70 5.40

5.26 3.48 5.14

16.53 11.16 12.88

HADDOCK 67.75 46.63 39.34

4.90 5.47 2.59

2.29 2.26 2.39

12.40 8.28 9.00

RosettaDock 103.51
15.79
5.26
15.79
ZDOCK 109.49
5.64
8.70
19.38
The values given in each cell of the table are as follows: averaged number of non-bonding
contacts per residue, averaged number of H-bonds contacts per residue, averaged number
of hydrophobic residues engaged in protein-protein interaction per residue. The values
are calculated for proteins as shown in Table 6.2 . Values calculated according to the data
available in PDBSum database

the ligand is far less pronounced than the one caused by protein complexation. It
should be noted that other tools described in this chapter ran into serious problems
when trying to model the 1G8M complex, most likely due to the relatively large sur-
face area of its complexation interface.

6.5 Summary

The programs discussed in this chapter have been selected to showcase various
means of identifying protein-protein complexation sites. Such analysis cannot be
called “blind prediction” since it relies on user-picked starting structures and potential
zones of interest. The aim of our study was to assess the validity and accuracy of
each algorithm for a large set of sample proteins. A secondary goal was to divide
protein complexes into subgroups: since some proteins form rather peculiar com-
plexes, the “lock and key” abstraction is not always applicable.
Even when potential complexation interfaces are suggested by the user, the
presented tools do not always produce correct results. This is most likely due to
deficiencies in their conformation space search algorithms.