Results from the proteinprotein docking strategy

We are currently benchmarking the protein docking procedure on a set of 15 complexes with coordinates available for both components in their unbound state. The results will be placed on our Web page (www.bmm.icnet.uk). Preliminary inspection suggests the results of the new benchmark will be similar to our original benchmark. Here we report the results of the original benchmark. The approach used to obtain these results was (i) FTDOCK (as reported in Gabb et al. [12]), (ii) filtering on biological distance constraints [12], (iii) RPSCORE [20] and (iv) MULTIDOCK [25].

The benchmark consists of six enzyme/inhibitor and four antibody/ antigen complexes. With the exception of the coordinates of two antibodies (HyHEL5 and HyHEL10), all the coordinates were from the unbound state. Table 8.2 presents the results. In each study, 4000 solutions were generated by FTDOCK1. The total number remaining after the application of the biological filter is shown in column 2 of Table 8.2 and forms set 1. All the en-

Results of protein-protein docking. aCHYN-a-chymotrypsinogen, a-CHY-a-chymotrypsin, HPTI-human pancreatic trypsin inhibitor, BPTI-bovine pancreatic trypsin inhibitor, CHYI-chymotrypsin inhibitor, Subtilisin l-subtilisin inhibitor D1.3, D44.1, HyHEL5 and HyHELlOare monoclonal antibodies. For further details of coordinates see Gabb et al. [12]. In the table some degenerate identical complexes included our earlier studies have been excluded.

Results of protein-protein docking. aCHYN-a-chymotrypsinogen, a-CHY-a-chymotrypsin, HPTI-human pancreatic trypsin inhibitor, BPTI-bovine pancreatic trypsin inhibitor, CHYI-chymotrypsin inhibitor, Subtilisin l-subtilisin inhibitor D1.3, D44.1, HyHEL5 and HyHELlOare monoclonal antibodies. For further details of coordinates see Gabb et al. [12]. In the table some degenerate identical complexes included our earlier studies have been excluded.

System

Total no

N <2.5 A in

Rank

Rank

Rank MULTI-

Rank MULTI-

rmsd Ca

after

FTDOCK

FTDOCK

RPSCORE

DOCK in set

DOCK in set

atoms (A)

FTDOCK&

list

in set 1

in set 1

1

2 (i.e., after

FILTR (set 1)

(making set 2)

RPSCORE)

¬ęCHYN-HPTI

93

1

3

2

2

2

2.0

aCHY-ovomucoid

85

5

11

6

1

1

1.1

Kallikrein-BPTI

349

16

128

13

2

2

1.0

Subtilisin-CHY I

26

2

8

1

12

2

2.0

Subtilisin-subtilisin I

-

-

-

-

-

-

-

Trypsin-BPTI

205

7

12

14

23

3

1.7

D1.3-lysozyme

636

2

149

75

211

38

1.8

D44.1-lysozyme

539

4

34

24

101

21

2.5

HyHEL5-lysozyme

498

2

218

36

29

29

1.5

HyHELlO-lysozyme

700

4

48

6

9

2

1.1

zymes were serine proteases and the distance filter was that at least one residue in the inhibitor must be in contact with the one of the catalytic triad (i.e., His, Ser or Asp). For the antibodies, the constraint was that the antigen must contact either the third complementarity determining region of the light or the heavy chain (CDR-L3 or CDR-H3). In this list there are at least one, but often several, complexes that are considered a good prediction. In this study a good prediction is defined as within 2.5 A of the correct structure superposing Ca all atoms of the predicted complex generated from unbound coordinates onto the X-ray bound complex. If just FTDOCK is used, the rank of the first good solution is given in column 4. In four out of the six enzyme/inhibitor complexes a good solution is obtained within 16 solutions. However no suitable complex is obtained for subtilisin docking to its inhibitor. The results for docking antibodies with antigens are less successful.

The results from FTDOCK can then be re-ranked using empirical scores (RPSCORE) to yield set 2. In general the rank at which one finds a good solution is decreased. A good solution is always found in the top 12% of the list of structures generated by FTDOCK (apart for subtilisin with its inhibitor). Set 2 can be further re-ranked using MULTIDOCK. The strategy was to take the top 25% of structures generated by RPSCORE, which is over double the fraction required in this benchmark. These are solutions are then ranked using MULTIDOCK in vacuo (see column 7). For the five serine proteases (i.e., excluding subtilisin and its inhibitor), a good solution is at rank 3 or better. The results for the antibody-antigen complexes are poorer with ranks as large as 38 that must be examined to ensure a good solution. Possibly the lower affinity for antibody-antigen complex formation compared to the enzyme-inhibitors studied [32] is reflected in the poorer discrimination in modelling their docking. Table 8.2 also presents the results of re-ranking the results of FTDOCK directly with MULTIDOCK (column 6). The approach of first using RPSCORE and then MULTIDOCK is always as good as, and generally superior, to using just MULTIDOCK for screening. Table 8.2 also gives the rmsd of the first ranked good solution. For about half of the simulations, the predicted structure is better than 2.0 A rmsd of the bound coordinates.

Figure 8.7 illustrates some of the predicted complexes. Figures 8.7a and b show the Ca trace of the predicted complex generated by FTDOCK superposed on the X-ray complex for trypsin-BPTI (bovine pancreatic trypsin inhibitor) and Fab D1.3-hen lysozyme. Figure 8.7c illustrates the results of side-chain refinement using MULTIDOCK. Three side-chain in BPTI that mediate the interaction with trypsin are shown. After MULTIDOCK two of the three move closer to the correct position than their original location following rigid-body docking using FTDOCK. In this benchmark, the agreement between predicted and native structure is sufficient to suggest further experiments, such as mutagenesis, to probe the interaction. Furthermore, the identification of the loop and key side-chains in an inhibitor

Comparison of native and modelled protein-protein complexes. In the figures the black line denotes the X-ray structure of the complex and the grey lines the predicted complex. Fig.

a. Superposition of the predicted and native complexes for bovine pancreatic trypsin inhibitor (BPTI) docking to trypsin. A Ca trace is shown. The predicted structure is generated by FTDOCK1 and the rmsd of the superposition is 1.7 A. Fig.

b. Superposition of the predicted and native complexes of Fab D1.3 and hen lysozyme. A Ca trace is shown. The predicted structure is generated by FTDOCK1 and the rmsd of the superposition is 1.8 A.

Fig. 8.7 (continued) Fig. c. Refinement of the docking of BPTI to trypsin. Only BPTI is shown for clarity. The figure shows a Ca trace with three side-chain that are central fore the interaction of BPTI with trypsin. The grey lines are the initial results from FTDOCK1, the light lines, the prediction after refinement by

MULTIDOCK. The black lines are the X-ray structure of the complex.

Fig. 8.7 (continued) Fig. c. Refinement of the docking of BPTI to trypsin. Only BPTI is shown for clarity. The figure shows a Ca trace with three side-chain that are central fore the interaction of BPTI with trypsin. The grey lines are the initial results from FTDOCK1, the light lines, the prediction after refinement by

MULTIDOCK. The black lines are the X-ray structure of the complex.

could be a valuable guide for the design of novel compounds of pharmaceutical use.

Continue reading here: Modelling proteinDNA complexes

Was this article helpful?

0 0