Table 1. Comparison of
protein-protein docking programs/methods.
Name Author Publication Web page |
ZDOCK (in C) Chen & Weng Proteins. 47:281-294 (2002). |
MolSoft 2000 ICMAbagyanProtein science (2002), 11:280-291. http://www.scripps.edu/~jfrecio/ICMprotdock/ |
MolFit + Electrostatics
Heifetz & Eisenstein (Israel) Protein science (2002), 11:571-587. |
BiGGERPalma & Moura Proteins 39:372-384 (2000) |
GRAMM
Vakser PNAS
96:8477-8482(2000) More papers on http://reco3.ams.sunysb.edu/gramm |
HEX3.0
Ritchie & Kemp Proteins 39:178-194 (2000) http://www.biochem.abdn.ac.uk/hex/ |
Search algorithm Shape score |
Systematic search with FFT grid spacing =1.2Å (100 grids for a typical complex) rotational interval=15°. Shape score = correlation
(receptor1, ρi, 0,
ligand1, ρi, 0) |
Pseudo-Brownian Monte Carlo sampling. 5 grid probes: two vdw
probes H and C atoms. 120 starting orientation by systematically rotating the ligand. Each starting orientation
explored by pseudo-Brownian Monte-Carlo sampling. new conformations selected
by Metropolis criterion at 300K, 5000K. |
Systematic search with
FFT. Grid spacing= 1.0-1.2 Å Rotation interval: 12°. Shape score is similar to
that in ZDOCK. |
Systematic search with
FFT. Grid spacing = 1.0 Å. Rotational interval = 15°. |
Systematic search with
FFT. The first paper
introducing FFT to shape correlation in protein-protein docking. PNAS
89:2195-2199 (1992). Basically, it is a
shape-based docking program (hydrophobic contacts can also be considered). Resolution can be adjusted. |
Systematic search using
spherical polar Fourier correlations. No need for grid. The search space is
represented by 5 Euler rotation angles and one inter-mol distance. Fast and less memory. Shape and electrostatics
are represented using series expansions of orthonormal spherical polar basis
functions. |
Name Author Publication Web page |
ZDOCK (in C) Chen & Weng Proteins. 47:281-294 (2002). |
MolSoft 2000 ICMAbagyanProtein science (2002), 11:280-291. http://www.scripps.edu/~jfrecio/ICMprotdock/ |
MolFit + Electrostatics
Heifetz & Eisenstein (Israel) Protein science (2002), 11:571-587. |
BiGGERPalma & Moura Proteins 39:372-384 (2000) |
GRAMM
Vakser PNAS
96:8477-8482(2000) More papers on |
HEX3.0Ritchie & Kemp Proteins 39:178-194 (2000) http://www.biochem.abdn.ac.uk/hex/ |
Energy function ( Electr. Vdw. Desolvation ) |
1. Electr:
Coulombic with CHARMM19 charges. 2. Vdw: no 3. Desolvation:
Atomic Contact Energy (ACE, emperical, 18 atom types, cutoff = 6Å ) Desolvation score =
correlation (receptorACE, ligandACE) Score=0.01Shape+ Desol.
+0.06Elect. And other combinations. |
1. Electr + desol
Eele/solv: coulombic with e=4r + atomic solvent accessible surface
term. 2. Vdw (EHvw
and ECvw): smoother 6-12 potential 3. Hbond Ehb:
spherical Gaussian 4. Hydrophobic Ehp:
30 cal/mol.Å2´ buried
hydrophobic surface area. Score = EHvw +
Ecvw + Eele/solv + Ehb + Ehp optimal scrore = EHvw
+ Ecvw + 2.16Eele/solv + 2.53Ehb + 0.20Ehp
+ 0.20Esolv |
1.
Electr: PB by Delphi; grid spacing = 0.5 Å. The potential of Delphi
grid is transferred to each MolFit grid by constructing a potential spheres.
Delphi potential is calculated only once. But the potential of each MolFit
grid will change with the translation/rotation of the molecules during the
search. Charges: PARSE partial
charges (better) or formal charges on Arg, Lys, Asp, Glu and His. Score = Shape + wElectr. Optimal w = 0.35 |
1. Electr:
coulombic with rij = rij+c. c is the minimal contact distance = 1.5 Å.
Amber4.1 force field. 2. solvation:
solvent-accessible surface area. |
Only shape complementarity is considered in GRAMM. But energy score functions can be used in the post-GRAMM stage to
rank the results. See table 2. |
1. electr.: Poisson’s equations.
For HyHel-5-lysozyme complex, KR = 8 KH= 0.8 KJ/mol/Å3. |
Name Author Publication Web page |
ZDOCK (in C) Chen & Weng Proteins. 47:281-294 (2002). |
MolSoft 2000 ICMAbagyanProtein science (2002), 11:280-291. http://www.scripps.edu/~jfrecio/ICMprotdock/ |
MolFit + Electrostatics
Heifetz & Eisenstein (Israel) Protein science (2002), 11:571-587. |
BiGGERPalma & Moura Proteins 39:372-384 (2000) |
GRAMM
Vakser PNAS
96:8477-8482(2000) More papers on |
HEX3.0
Ritchie & Kemp Proteins 39:178-194 (2000) http://www.biochem.abdn.ac.uk/hex/ |
Flexibility Side chain or backbone |
Arg and Lys: side chain collapse allowed. |
Biased probability Monte Carlo minimization on side chains torsion
angles of surface residues of ligand. |
No consideration. |
Side chain flex: Arg, Lys, Asp, Glu and Met. |
No consideration. |
No consideration. |
timing |
Computer: R10000 origin 2000 10 h: 100 grids (single cpu, typical) 19 h: 128 grids |
667MHz Alpha: 2-7h for rigid docking and 7-20 min for refinement. (FTDOCK: 6h on 8 R10000) |
R10000 SCI octone: 9h for 128 grids. |
2-8 h on 450MHz pII. |
Convex C-220: 7.5h with 1100 atoms. If grid=128, orientations = 3 ´ 107 |
2h: SCI R5000 for 5.4 ´ 108
orientations. |
Name Author Publication Web page |
ZDOCK (in C) Chen & Weng Proteins. 47:281-294 (2002). |
MolSoft 2000 ICMAbagyanProtein science (2002), 11:280-291. http://www.scripps.edu/~jfrecio/ICMprotdock/ |
MolFit + Electrostatics
Heifetz & Eisenstein (Israel) Protein science (2002), 11:571-587. |
BiGGERPalma & Moura Proteins 39:372-384 (2000) |
GRAMM
Vakser PNAS
96:8477-8482(2000) More papers on |
HEX3.0
Ritchie & Kemp Proteins 39:178-194 (2000) http://www.biochem.abdn.ac.uk/hex/ |
Test proteins. ( cases showing conformational changes ) |
27 in total. 5 homodimers, 13 unbound-unbound, including 1brs (Barnase has conformational change
) and 1fss (Fasciculin II has conformational change) interface = 10Å near-native = interface Cα < 2.5 Å. Shape helped 1fss, Electr helped 1brs, desol did not help 1fss or
1brs. 24/27 complexes found near native in top 2000. On the web page more test cases are shown, including 1bth (unbound
thrombin/ unbound BPTI, thrombin has large conformational change) best
rmsd=3.67 |
24 in total. 20 enzyme/inhibitor (motion:
1fss/FasII, 1bgs/bn, 1ay7/Sa, 1acb/ eglin C), Bound-bound redocking and unbound-unbound docking. Docked: ligand interface Cα < 4Å. Rank is <20 in 85% complexes with no major backbone motion. 1fss:1.7Å; 1bgs: 4.2Å; 1ay7: 6.2Å; 1acb:eglin C deformed (interface
bone rmsd>1.8Å ?) Side chain refinement helped 1fss, not 1bgs or 1ay7. |
17 in total. 11 enzyme/inhibitor (8 unbound-unbound
docking), Rmsd: all interface Cα< 3Å. Electrostatics helped 1brs: 1.82 Å rank 1 (0.77 crystal), 1fss: 0.88
Å rank 3 (0.91 crystal) , 1bth: 3.87 Å rank 559 (2.52 crsytal) |
25 in total. Near native: all Cα rmsd < 4Å. Near native were found for 20 pairs, 14 of which were in top 20
ranks. 1acb: Bound-bound docking
rmsd = 0.61 Å. 1fss: unbound-unbound docking rmsd = 3.2 Å. 1brs: unbound-unbound docking rmsd = 1.89 Å. |
475 complexes: low resolution docking (7 Å) 52% showed low resolution recognition. |
30 complexes: 20 antibody-antigen, 8 enzyme-inhibitor, 2 dimers. 1bgs: 0.88 Å Cα rmsd. Correct conformation frequently identified for re-docking. Unbound docking: 11 out of 18
within top 20. |
|
Table 2. Two ranking methods and Treedock |
|
|
AuthorPublication Web page |
Ranking after Shape-based docking Camacho & Vajda Protein 40:525-537 (2000) |
Ranking after shape-based docking Camacho & Vajda PNAS 98:10636-10641 (2001) |
Treedock (in C) Fahmy & WagnerJACS On web 01/25/2002 |
Score functions |
Encounter complexes from
GRAMM and DOT. Score = electr. (CHARMM19+4r+polar
H) + desol. (ACE) + VDW This is the first time to
combine the empirical energy functions for protein-protein docking ranking. |
Encounter complexes from
DOT: 7-14 Å RMSD (not clear which atoms). Princeple: vdw becomes
sensitive only when two proteins get close. Multi-steps: 1.
rank by ΔGS : electr. (charmm19+4r +polar H) + desol
(ACE). 2.
construct new pair from two highly ranked. 3.
minimize the new and better pairs (fixed backbone) until vdw
converges. Here it is not clear if vdw is calculated in the minimization. 4.
rank by ΔGS + γ vdw. γ<= 1. |
Systematic search over
contact points on molecular surfaces. No need grid. Search space: drastically
reduced by using anchor atoms specified by users. (a pair of anchor atom are
two atoms on each molecular surface, which are supposed to be in contact upon
binding.) Search steps: 1. specify anchor atoms.
If both binding sites are known, only one pair anchor atoms is used.
Therefore, the docking accuracy depends on the select of anchor atoms. 3. generate a number of contact points (tangent contact) on each
anchor atom. Score: only Lennard-Jones potential. |
AuthorPublication Web page |
Ranking after Shape-based docking Camacho & Vajda Protein 40:525-537 (2000) |
Ranking after shape-based docking Camacho & Vajda PNAS 98:10636-10641 (2001) |
Treedock (in C) Fahmy & WagnerJACS On web 01/25/2002 |
Test cases |
Correlation R (score, rmsd): 5 complexes. 0.25-0.69. |
8 complexes, including 1brs and 1fss. All backbone: < 2 Å in 4 complexes (1brs =2.52Å, 1fss = 1.59Å). Interface Cα (within 10 Å) < 2 Å in 6 complexes (1brs=2.58Å,
1fss = 1.78Å). |
3 complexes of immunoglobulin superfamily domains. 1 phosphatase-small inhibitor More efficient for small molecules to dock to a binding site of a
protein. |
timing |
No data |
24 h on RISC 10000 SGI |
1-30 min on R10000SGI for one
pair of anchor atoms. 10 h for 18×13 anchor atoms. Faster if anchor atoms are less solvent accessible. |
Comments: In terms of search algorithms, the spherical polar Fourier method in
HEX3.0 is faster than the commonly used FFT method. In terms of accuracy,
MOLFIT and HEX3.0 are good because they use PB for electrostatics (good results
for 1brs and 1fss). Treedock is efficient only when at least a pair of anchor
atoms are known.
Questions: is the grid spaceing equal to the translation step size in
FFT? The search space = translation steps ´ rotation steps. If the
molecules is large, the search space will be very large. In spherical polar
Fourier, the search space is represented by 5 Euler rotation angles and one
inter-mol distance.