Supplementary Material
Defining the limits of homology modelling in information-driven
protein docking
J.P.G.L.M. Rodrigues1, A.S.J. Melquiond1, E. Karaca1, M. Trellet1,2, M. van Dijk1,
G.C.P. van Zundert1, C. Schmitz1, S.J. de Vries1,3, A. Bordogna4, L. Bonati4, P.L.
Kastritis1 and Alexandre M.J.J. Bonvin1*
1. Bijvoet Center for Biomolecular Research, Faculty of Science/Chemistry,
Utrecht University, Utrecht, 3584CH, The Netherlands.
2. Current address: Groupe VENISE, CNRS LIMSI, 91403 Orsay CEDEX,
France
3. Current address: Physik-Department (T38), Technische Universität München,
James-Franck-Str. 1, 85748 Garching, Germany
4. Università degli Studi di Milano-Bicocca, Piazza della Scienza, 1 - 20126
Milano, Italy
* To whom correspondence should be addressed.
Alexandre M.J.J. Bonvin
Bijvoet Center for Biomolecular Research, Faculty of Science, Utrecht University,
Padualaan 8, 3584 CH Utrecht, The Netherlands.
Tel. +31 30 2533859
a.m.j.j.bonvin@uu.nl
Content:
Homology modelling of interacting partners ................................................................................................................. 3
HADDOCK protocol and energetics details .................................................................................................................. 3
Restraints used in the docking predictions of CAPRI rounds 22-27 ..................................................................... 4
Supplementary Figure S1. .................................................................................................................................................... 6
Supplementary Figure S2. .................................................................................................................................................... 7
Supplementary Table S1. ...................................................................................................................................................... 8
Supplementary Table S2. ...................................................................................................................................................... 9
Supplementary Table S3. ................................................................................................................................................... 11
References ............................................................................................................................................................................... 12
2
Homology modelling of interacting partners
Each interacting partner of the protein-protein complexes used in this study was
modelled using the following protocol:
1. A sequence similarity search was performed on a database of sequences of
proteins with known structure (‘PDB’) using the PSI-BLAST algorithm1 with
default parameters.
2. All hits with low statistical significance (Expect (E) value greater than 10-3) or
with an alignment overlap shorter than 85% were discarded.
3. Sequences with identities to the target sequence in the range of 20-80% were
binned in steps of 5%, and one representative sequence from each ‘bin’ was
selected.
4. Each sequence was re-aligned to the target sequence using three different
algorithms: DALIlite2, T-Coffee3, Praline4 (all with default parameters).
5. Based on the previously derived alignments, MODELLER 9v85 was used to
build homology models.
6. The 10 best models were selected according to their DOPE score6.
HADDOCK protocol and energetics details
HADDOCK7, 8 incorporates biochemical and/or biophysical information as restraints
to drive the modelling (EAIR). The docking protocol consists of three steps: i) rigidbody energy minimization stage, ii) semi-flexible refinement consisting of a simulated
annealing optimization in torsion angle space with side-chain and backbone flexibility
at the interface, and iii) final refinement in explicit solvent (e.g. TIP3P water). Nonbonded interactions were calculated with the OPLS force field9 using a 8.5Å cut-off.
The electrostatic energy (Eelec) was calculated using a shifted Coulomb potential,
while the van der Waals energies (Evdw) were calculated with a Lennard-Jones
potential, with a switching function between 6.5Å and 8.5Å.
In HADDOCK, experimental or prediction information is incorporated in the form of
Ambiguous Interaction Restraints (AIRs), a notion similar to that of ambiguous
distance restraints based on NOE data in structure calculation of NMR structures (see
e.g. Linge et al.10). HADDOCK creates AIRs based on active and passive residues
defined by the user. For every active residue, HADDOCK creates one single AIR
restraint between that residue and all active and passive residues on the partner
molecules. These restraints are incorporated in the energy function through a softsquare harmonic potential EAIR) that depends on an effective distance. This latter is
calculated using the following formula:
in which A and B are molecules, i iterates over all distance restraints, Natoms indicates
all atoms of a given residue and Nres the sum of active and passive residues for a given
protein. For two single atoms, the effective distance is equal to the Cartesian distance,
3
but as the number of atoms grows, the effective distance becomes shorter and shorter
due to the larger number of distances taken into consideration. By default,
HADDOCK enforces an upper limit to the effective distance of 2Å. In case this
distance is exceeded, the AIR energy becomes positive and the active residue
experiences an attractive force towards the active and passive residues it is restrained
to in the partner molecule. Otherwise, the restraint is satisfied and the AIR energy,
and consequently the attractive force, is zero. The exact relation between effective
distance and energy is described in Nilges et al. 11. Given the many atom-atom
distances contributing inversely to the effective distance, a typical AIR restraint is
satisfied, depending on the level of ambiguity, if the residue comes within 3-4 Å of
any active or passive residues of the partner molecule. As such, (putative) interface
residues are driven into (a surface region on) the partner protein, but not to any
specific partner residue.
Restraints used in the docking predictions of CAPRI rounds 22-27
The specific restraint files and the PDB files matching them (given likely differences
in residue numbering, chain naming, etc.) are provided upon request to the authors.
Target 46
Unambiguous interaction restraints, derived from inter-domain contacts on a distant
structural homologue (1P9112). The N6 Adenine Specific Dna Methylase partner in
T46 (chain B) structurally aligns to the fragment 67-185 of 1P91, while the last betastrand of Trm112 Activator Protein (chain A in T46) – residues 115-118 aligns with a
small beta-strand in 1P91 (residues 33-40). The restraints were implemented between
Carbon alpha atoms with distances taken from 1P91 with a lower bound of 2Å.
Additional unambiguous restraints were defined to keep geometry of the SAH
heterogroup and the zinc coordination sites in Mtq2.
Target 47
Unambiguous interaction restraints, derived from the interface residues of structure
provided by the CAPRI committee. The restraints were implemented between Carbon
alpha atoms with lower and upper bounds of 0.25Å.
Targets 48 & 49
The interaction restraints are a combination of CPORT28 predictions
(haddocking.org/services/CPORT) and homology-derived restraints using the PDB
entry 2YVJ (NADH-dependent ferredoxin reductase and Rieske-type [2Fe-2S]
ferredoxin)13. We also included unambiguous distances to maintain the geometry of
ferrodoxin iron coordination site.
Target 50
The interaction restraints were defined based on a statistical analysis of the most
contacted residues on hemagglutinin (ab initio docking run with 10000 models for the
rigid-body stage), together with the assumption that the designed protein HB36.3 was
binding to a less mutation-prone surface. On the HB36.3 molecule, the interaction
surface was defined based on biophysical properties (aromatic and hydrophobic
patches).
4
Target 51
We only defined connectivity restraints between the different monomers of the
complex, together with center of mass14 restraints to keep the solutions compact.
Target 53 & 54
CPORT predictions (haddock.science.uu.nl/services/CPORT) were used to drive the
docking.
Target 57
We defined ambiguous interactions restraints between each sulfate group on the
heparin molecule and all arginine and lysine residues (guanidinium and amino groups
only) of Bt4661.
5
Supplementary Figure S1.
Influence of the quality of the homology model and of the data on the scoring
performance. The quality of the model is measured by sequence identity between
target and template. The impact of the quality of the data is assessed by performing
the docking using CAPRI restraints (CI) (right panel) and True interface restraints
(TI) (left panel), respectively.
6
Supplementary Figure S2.
Changes in backbone interface RMSD of the predicted complexes upon flexible
refinement for all acceptable models (iRMSD < 4Å) for all targets (21564 models).
The top panels (A & B) show the distribution of changes as measured by the
difference in the backbone interface RMSD of the modelled complex between the
rigid-body stage and after final refinement. A negative value means the docked model
is closer to the native bound conformation. The bottom panels (C & D) show the
relation of sequence identity between target/template and change in backbone
interface RMSD after flexible refinement with respect to the bound conformation.
The models colored according to their CAPRI target: red - T12, blue – T18, green –
T26, yellow – T27, purple – T40, orange – T41.
7
Supplementary Table S1.
Z-DOPE
Cα RMSD (Model vs Template)
Backbone i-RMSD (Model vs Template)
Verify3D
Molprobity
Cα RMSD (Model vs Crystal)
GDT-TS (Model vs Crystal)
Backbone i-RMSD (Model vs Crystal)
Sidechain i-RMSD (Model vs Crystal)
0.86
0.75
0.79
0.80
0.52
0.74
0.68
0.66
0.72
0.81
0.51
0.67
0.78
0.71
0.84
0.86
0.62
0.73
0.70
0.65
0.67
0.74
0.59
0.74
0.77
0.72
0.64
0.64
0.11
0.53
0.39
0.43
0.55
0.68
0.44
0.64
0.95
0.75
0.72
0.71
0.43
0.72
0.63
0.58
0.59
0.73
0.41
0.62
0.51 0.64 0.58 0.53
0.52 0.74 0.68 0.66
0.62 0.73 0.70 0.65
0.11 0.53 0.39 0.43
0.43 0.72 0.63 0.58
Predictive Indices
0.63
0.72
0.67
0.55
0.59
0.73
0.81
0.74
0.68
0.73
0.54
0.51
0.59
0.44
0.41
0.72
0.67
0.74
0.64
0.62
0.75 0.83 0.61 0.74
1.00 0.89 0.73 0.85
0.89 1.00 0.67 0.77
0.73 0.67 1.00 0.78
0.85 0.77 0.78 1.00
Calculated Indices
TVSMod_RMSD
0.62
0.64
0.75
0.74
0.38
0.55
0.52
0.55
0.68
0.64
0.39
1.00
Qmean
0.37
0.33
0.52
0.55
0.54
0.67
0.61
0.46
0.30
0.40
1.00
0.39
Interface Sequence Identity
0.73
0.63
0.80
0.81
0.37
0.60
0.48
0.53
0.84
1.00
0.40
0.64
Global Sequence Similarity
0.56
0.48
0.73
0.73
0.37
0.44
0.41
0.46
1.00
0.84
0.30
0.68
Global Sequence Identity
0.61
0.59
0.63
0.59
0.37
0.83
0.84
1.00
0.46
0.53
0.46
0.55
i-RMSD from Native Structure (CI)
0.63
0.55
0.59
0.58
0.53
0.94
1.00
0.84
0.41
0.48
0.61
0.52
i-RMSD from Native Structure (TI)
TVSMod_Over
Cross correlation table of all indices. The values are colour coded from green (high correlation) to red (low correlation).
i-RMSD from Native Structure (TI)
i-RMSD from Native Structure (CI)
Global Sequence Identity
Global Sequence Similarity
Qmean
TVSMod_RMSD
TVSMod_Over
Z-DOPE
Cα RMSD (Model vs Template)
Backbone i-RMSD (Model vs Template)
Verify3D
Molprobity
1.00
0.84
0.70
0.69
0.39
0.73
0.63
0.61
0.56
0.73
0.37
0.62
0.84
1.00
0.59
0.57
0.34
0.65
0.55
0.59
0.48
0.63
0.33
0.64
0.70
0.59
1.00
0.99
0.44
0.66
0.59
0.63
0.73
0.80
0.52
0.75
0.69
0.57
0.99
1.00
0.46
0.65
0.58
0.59
0.73
0.81
0.55
0.74
0.71
0.57
0.92
0.93
0.51
0.64
0.58
0.53
0.63
0.73
0.54
0.72
0.39
0.34
0.44
0.46
1.00
0.53
0.53
0.37
0.37
0.37
0.54
0.38
0.73
0.65
0.66
0.65
0.53
1.00
0.94
0.83
0.44
0.60
0.67
0.55
Interface Sequence Identity
Cα RMSD (Model vs Crystal)
GDT-TS (Model vs Crystal)
Backbone i-RMSD (Model vs Crystal)
Sidechain i-RMSD (Model vs Crystal)
0.71
0.86
0.78
0.77
0.95
0.57
0.75
0.71
0.72
0.75
0.92
0.79
0.84
0.64
0.72
0.93
0.80
0.86
0.64
0.71
1.00
0.75
0.83
0.61
0.74
8
Supplementary Table S2.
Details of the methods used for the predictive and measured indices.
Index Name
Sequence Identity
Sequence Similarity
Sequence Identity
(interface only)
CαRMSD (model –
template)
Backbone interface
RMSD (model –
template)
Index
Type
Predictive
Predictive
http://www.ebi.ac.uk/Tools/psa/emboss_needle/
http://www.ebi.ac.uk/Tools/psa/emboss_needle/
N/A
N/A
Measured
http://www.ebi.ac.uk/Tools/psa/emboss_needle/
N/A
Predictive
ProFit V3.1 (//www.bioinf.org.uk/software/profit/)
Used alignment from homology modelling as ZONEs for fitting.
Predictive
ProFit V3.1 (//www.bioinf.org.uk/software/profit/)
Used alignment from homology modelling as ZONEs for fitting.
Calculation Method
Notes
The Molprobity score is calculated by a log-weighted combination of the clash
score, the percentage of Ramachandran Plot not favoured, and the percentage
of bad rotamers. It reflects the crystallographic resolution at which these
values would be found. Therefore, the lower the score the better.
Score calculated from the atomic coordinates of a model that assesses the
compatibility of that model to its own amino acid sequence, as measured by a
3D profile.
Atomic distance-dependent statistical potential calculated from a sample of
native structures.
SVM regression model parameterized on a dataset containing 5.790.889
models that predicts the Cα RMSD between the model and the native
structure.
SVM regression model parameterized on a dataset containing 5.790.889
models that predicts the fraction of Cα atoms within 3.5Å of their correct
positions in the native structure.
Linear combination of six structural descriptors: two distance-dependent
interaction potentials of mean force based on Cβ atoms and on all atom types
to assess long-range interactions; a torsion angle potential; a solvation
potential; agreement of secondary structure prediction and calculation;
agreement of predicted buried surface area and calculation.
MolProbity15
Predictive
http:// psvs-1_4-dev.nesg.org
Verify3D16
Predictive
http:// psvs-1_4-dev.nesg.org
Z-DOPE6
Predictive
http://modbase.compbio.ucsf.edu/evaluation/
TVSMod_RMSD17
Predictive
http://modbase.compbio.ucsf.edu/evaluation/
TVSMod_Over17
Predictive
http://modbase.compbio.ucsf.edu/evaluation/
Qmean18
Predictive
http://swissmodel.expasy.org/qmean/
CαRMSD (model –
reference)
Backbone Interface
RMSD (model –
Measured
Measured
ProFit V3.1 (//www.bioinf.org.uk/software/profit/)
N/A
Measured
ProFit V3.1 (//www.bioinf.org.uk/software/profit/)
N/A
9
reference)
Sidechain Interface
RMSD (model –
reference)
GDT-TS19 (model –
reference)
Measured
ProFit V3.1 (//www.bioinf.org.uk/software/profit/)
Calculated on all atoms except for the backbone, excluding hydrogens.
Measured
http://as2ts.proteinmodel.org/AS2TS/LGA_list/lga_
pdblist.html
Parameters for LGA calculation: -3 –o1 –d:4.0 -gdc
10
Supplementary Table S3.
Performance of the HADDOCK in the latest CAPRI rounds. Detail of the quality of
the best submitted model: interface RMSD, ligand RMSD, Fraction of Native
Contacts (FNAT), and CAPRI Star Quality.
Manual Submission
Target Name
Complex Name
Target Type
T46
T47
T48
T49
T50
T51
T53
T54
T57
T58
Methyl transferase Mtq2/Trm112
colicin E2 DNase-Im2
T4moC /T4moH mono-oxygenase
T4moC /T4moH mono-oxygenase
HB36.3 designed protein / flu hemagglutinin
Xylanase Xyn10B
Designed Rep4/Rep2 a-repeat
Designed neocarzinostatin/Rep16 a-repeat
BT4661/heparin complex
PliG /SalG lysozyme
HH
UU
UU
UU
HU
UUHU(U)
UH
UH
UU
UU
Best Model
i-RMSD (Å) l-RMSD (Å)
3.76
10.1
0.98
1.52
3.42
9.08
3.55
14.0
1.63
6.69
2.08
6.03
1.73
4.86
--2.04
4.68
2.61
6.91
Fnat Quality
0.49
*
0.86
***
0.23
*
0.26
*
0.67
**
0.41
*
0.50
**
--0.88
**
0.29
*
Best Model
i-RMSD (Å) l-RMSD (Å)
3.83
11.9
0.80
1.81
--3.07
10.7
--------2.11
4.88
---
Fnat Quality
0.57
*
0.86
***
--0.11
*
--------0.77
**
---
Server Submission
Target Name
Complex Name
T46
T47
T48
T49
T50
T51
T53
T54
T57
T58
Methyl transferase Mtq2/Trm112
colicin E2 DNase-Im2
T4moC /T4moH mono-oxygenase
T4moC /T4moH mono-oxygenase
HB36.3 designed protein / flu hemagglutinin
Xylanase Xyn10B
Designed Rep4/Rep2 a-repeat
Designed neocarzinostatin/Rep16 a-repeat
BT4661/heparin complex
PliG /SalG lysozyme
11
HH
UU
UU
UU
HU
UUHU(U)
UH
UH
UU
UU
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
Boratyn GM, Camacho C, Cooper PS, Coulouris G, Fong A, Ma N, Madden TL, Matten WT,
McGinnis SD, Merezhuk Y, Raytselis Y, Sayers EW, Tao T, Ye J, Zaretskaya I. BLAST: a
more efficient report with usability improvements. Nucleic Acids Res 2013.
Holm L, Kääriäinen S, Rosenström P, Schenkel A. Searching protein structure databases with
DaliLite v.3. Bioinformatics 2008;24(23):2780–2781.
Taly J-F, Magis C, Bussotti G, Chang J-M, Di Tommaso P, Erb I, Espinosa-Carrasco J,
Kemena C, Notredame C. Using the T-Coffee package to build multiple sequence alignments
of protein, RNA, DNA sequences and 3D structures. Nat Protoc 2011;6(11):1669–1682.
Simossis VA, Heringa J. PRALINE: a multiple sequence alignment toolbox that integrates
homology-extended and secondary structure information. Nucleic Acids Res 2005;33(Web
Server issue):W289–W294.
Sali A, Blundell TL. Comparative protein modelling by satisfaction of spatial restraints. J Mol
Biol 1993;234(3):779–815.
Shen M-Y, Sali A. Statistical potential for assessment and prediction of protein structures.
Protein Sci 2006;15(11):2507–2524.
Dominguez C, Boelens R, Bonvin AMJJ. HADDOCK: a protein-protein docking approach
based on biochemical or biophysical information. J Am Chem Soc 2003;125(7):1731–1737.
de Vries SJ, van Dijk M, Bonvin AMJJ. The HADDOCK web server for data-driven
biomolecular docking. Nat Protoc 2010;5(5):883–897.
Jorgensen W, Tirado-Rives J. The OPLS potential functions for proteins. Energy minimizations
for crystals of cyclic peptides and crambin. J Am Chem Soc 1988;110(6):1657–1666.
Linge JP, Habeck M, Rieping W, Nilges M. ARIA: automated NOE assignment and NMR
structure calculation. Bioinformatics 2003;19(2):315–316.
Nilges M, Gronenborn AM, Brünger AT, Clore GM. Determination of three-dimensional
structures of proteins by simulated annealing with interproton distance restraints. Application
to crambin, potato carboxypeptidase inhibitor and barley serine proteinase inhibitor 2. Protein
Eng 1988;2(1):27–38.
Das K, Acton T, Chiang Y, Shih L, Arnold E, Montelione GT. Crystal structure of RlmAI:
implications for understanding the 23S rRNA G745/G748-methylation at the macrolide
antibiotic-binding site. Proc Natl Acad Sci USA 2004;101(12):4041–4046.
Senda M, Kishigami S, Kimura S, Fukuda M, Ishida T, Senda T. Molecular mechanism of the
redox-dependent interaction between NADH-dependent ferredoxin reductase and Rieske-type
[2Fe-2S] ferredoxin. J Mol Biol 2007;373(2):382–400.
de Vries SJ, Bonvin AMJJ. CPORT: a consensus interface predictor and its performance in
prediction-driven docking with HADDOCK. PLoS ONE 2011;6(3):e17695.
Chen VB, Arendall WB, Headd JJ, Keedy DA, Immormino RM, Kapral GJ, Murray LW,
Richardson JS, Richardson DC. MolProbity: all-atom structure validation for macromolecular
crystallography. Acta Crystallogr D Biol Crystallogr 2010;66(Pt 1):12–21.
Eisenberg D, Lüthy R, Bowie JU. VERIFY3D: assessment of protein models with threedimensional profiles. Meth Enzymol 1997;277:396–404.
Eramian D, Eswar N, Shen M-Y, Sali A. How well can the accuracy of comparative protein
structure models be predicted? Protein Sci 2008;17(11):1881–1893.
Benkert P, Biasini M, Schwede T. Toward the estimation of the absolute quality of individual
protein structure models. Bioinformatics 2011;27(3):343–350.
Zemla A. LGA: A method for finding 3D similarities in protein structures. Nucleic Acids Res
2003;31(13):3370–3374.
12