The Receptor Modeling Challenges
Over 38,000 Structure files in PDB (~25/1 new structures a w.day).
A need to extend crystallography with structure prediction:
• low gene and domain coverage of the proteome : "the dark
matter" of the structural proteome (predict by homology & ab initio)
• imperfect interpretations of incomplete electron density:
predict H, Asn, Gin, His, protonation, errors, missing side chains, loops
• predict conf. of alternative functional states, induced fit
• predict protein or domain association
• predict mutations, SNPs, post-translational modifications
• predict ligand docking for virtual screening and design
Revolution in Virtual Chemistry and Pharmacology
		jN>-
-©-	Qr>xy	9
Millions of easily available vendor compounds
10 NIH centers: public bio-screening data
PubChem
Predicting compound properties: LogP,l_ogS,
CNS, hERG, PgP, CYPs: 3A4,2D6,2C9..
Predicting Molecular Properties					
					E-ADMET. LogP, LogS, PSA, Ubiquitous binding, Drug-Likeness, hERG, PgP, CYPs: 3A4,2D6,2C9
		0 AS	(   \	o	
|     !     .     L 1   II H	'				Half-life Metabolites
A critical bottle-neck for e-ADMET predictions: Available e-ADMET data (prediction algorithms are easier) Goal for public e-ADMET initiatives: generate e-ADMET databases					
Preparing receptor coordinates
•  PDB coordinates: imperfect interpretation of incomplete electron density.
•  Build a complete model (missing side-chains, loops etc.)
•  Predict correct Asn, Gin, His orientations, protons, detect errors.
Preparing a pdb-structure for docking
A.  Search for a pdb with the closest sequence to your protein of interest
B.  Choose the most suitable entry (or several entries)
C Find, build and edit the pocket composition and geometry.
•    X-ray with up to 2.5-2.8A resolution is preferable over NMR
•    NMR or homology models are only dockable by skillful operators
•    Forget electron microscopy
•    X-ray Resolution < 2.2 A is preferable.
(Structures with resolution > 2.3 A may have up to 30% peptide flips, the maps are not self-refinable)
•    Analyze symmetry if the pocket might be at the interface
•    Analyze relative b-factors. B > 100. are not credible
•    Pay attention to occupancies (in many cases pocket geometries of ligand conformations/presence are pure fantasies of the authors!).
•    Analyze alternative positions
•    Check orientations of His, Asn, Gin
•    Check protonation states of Glu, Asp, His
•    Analyze stongly bound water molecules, ions and co-factors .
Preparations: symmetry
Example: Cycloldextrin glycosyltransferase Entry: 1cdg, Res. 2.0A (Docking Rmsd without symmetry: 9.76) More examples: transthyretin 1f41 (thyroid hormone binds at the dimer interface)
Problem: the true pocket is formed by chains which are not explicitly present in a pdb entry.
Goal: Find all molecules/subunits or chains involved in the interaction with the ligand.
Warning signs: ICM pocket finder does not show pocket density; Binding site is obviously exposed
Recovery: generate symmetry related subunits (View/Cryst.Cell)_______________
Preparations: occupancies, b-factors and alternatives        ^iffiwjix
Glossary:
B-factor (or temperature factor):
mean-square displacement of atom from its position in the model. Bi = 79*<u2> (B of 80 means 1A dev.) Normal range: 5. - 50. A2.
Occupancy:
A fraction of atomic density at a given center. It there are two equally occupied conformers, both will have occupancies of 0.5 Normal value: 1. Range: 0.-1.
Alternatives:
If two or more alternative conformations for the same atom or group are discernable in the density, several alternative sets of coordinates are deposited.
Occupancies <= 0.5 are shown in magenta High b-factors are colored red
Problem: sometimes, when electron density is poor and/or ambiguous, crystallographers make things up (or just deposit an arbitrary conformation from a refinement program)
Goal: Identify fantasy atoms/groups
Warning signs: occupancies less than 0.5, b-factors
larger than 60-80 A2.
Tool: Color/label pocket atoms by occupancies/b-factors.
Recovery: Choose another entry, or refine with a ligand, or perform restrained minimization. Choose one of alternatives, or create alternative models
Preparations: occupancies, b-factors and alternatives. Example.
This is a very high resolution structure. For some key residues two alternative conformations are provided.
Recovery:
Choose one alternative or generate several separate docking models____________
Alternative positions for Thr and Val32
Entry: 1hmt.
Res. 1.4
Fatty Acid Binding Protein with stearic acid
Preparations: fixing histidines
That is hew histidine
Orientation at the heavy atom level
We need to discriminate between These two conformations
Often the xi2 angle needs to be Corrected by 180 d
Uncertainly at the protonation level
You need to decide which of the three conformations is correct for each important location.   The charged conformation is
Problem: orientations and protonation states of histidines are frequently wrong on pdb entries and need to be fixed to ensure correct docking results. Placement principe: maximization of hydrogen bonds and other hteractions with the rest of the protein and/or with the ligand. Recovery: ICM procedure optimizeHisProAsnGIn finds the best orientation and protonation state
Preparations: determining orientations of Gin, Asn, side chains
&
Orientation at the heavy atom level
The two conformations shown give similar electron density. We need to discriminate between these two conformations of the Asn side chains. The same ambiguity needs to be resolved for the xi3 angle of Gin
Background: xi2 in asparagines and xi3 in glutamines are frequently wrong or undefined and need to be corrected ensure correct docking.
Placement principle: maximization of hydrogen bonds and other interactions with the rest of the protein and/or with the ligand.
Recovery: ICM optimizeHisProAsnGIn procedure.
Preparations: do I need to uncharge Asp, Glu, Lys and Arg?
Definitions: DERK is Asp (D), Glu (E), Arg ® or Lys (K) Facts: pKs:  His 6.0, Cys 8.3 Glu 4.2, Asp 3.9
General recommendation: keep the DERK residues charged.
Problem: while in most cases DERKs are charged, in some special cases ED need to be uncharged or His needs to be charged.
Warning signs:
a DERK is buried and NOT involved in a salt bridge; Several DERKs of the same kind/charge are pointii the same space.
Example: HIV protease. 1ida. Asp 25 and 25' are protonated.
Recovery: Modify them to the uncharged forms.
Preparations: which waters to keep?
%
Example: leyedihydropteroate
synthase, anti-mycobacterial/TB target.
It binds to the buried Asp177 and            I    7 ,WM
improves electrostatic desolvation by       |
~10 units.
i*¥
Definition: crystallographic water: an oxygen
placed by a crystallographer or a refinement
to a blob of electron density. General recommendation: get rid of all water molecules, Keep only water molecules with three or four hydrogen bonds with the protein or Iigand atoms.
Reason: keeping inappropriate water(s) will prevect correct docking, while dropping good waters is usually tolerated.
However some tightly bound water molecules help docking and scoring and prevet from erroneous placement of H-bond-rich Iigand groups in water sites.
Recovery: Find interface waters with 3 or more protein/iigand neighbors and include them into your model.
Preparations: cofactors and metals?
Problem: metals may be required to dock a charged native Iigand (e.g. ATP is charged and requires 2 Mn++ ions.)
However, to the metals are not necessary for docking of neutral drugs.
Example: a kinase domain. 1atp
Local quality: Energy Strain
		;,-,.......
e  il		.
é '11- _n V "  -,		i^i
°" ^M^i^		
i		
■  More sensitive that geometrical and clash criteria
■  Based on fast ICM calculation of residue energies (recently accelerated 1000 fold)
■  Energy = Vacuum Energy + Solvation + Entropy
Deriving energy distributions for each amino-acid type
■ All high resolution PDB structures (<1.5A) collected
■  Distributions of residue energies calculated
■  Energy Distribution for each amino acid derived
■  Normalized energies derived
Calculating normalized residue energies for a model
■  Calculate Z-score (normalized energy) for each residue
■  Residues with EnorTn > 5 are probably wrong
Maiorov, Abagyan, 1998 Proteins,    The 1000-times faster version: 2004
1? ?w!rafoi!]?
Detecting Small Molecule Pockets from Structure
The problem
•   We do not know the nature of the ligand
•    Find the location and the extent / envelope of a pocket
•    The Lennart Jones potential is short-range and does not predict the location of the small molecule site.
A Physical Idea:
•    The CUMULATIVE potential integrated over a typical size of a ligand may predict the site location and extent
Detecting Binding	Pockets
Challenge: Predicting Ligand Binding Sites Without Knowing the Ligand Method: 1 .Calculate this potential	
a  r          r a      ag               ag	
\p(r)=\e-{Cp-hlÄÝP°Cp)	£ESbr
/I = 2.6 A	
2. Contour the potential 3. Filter out small blobs	
Pocket Database	
17,000 pockets                  Example: Biotin-binding protein (2izi)	
Benchmarking the Pocket Prediction Algorithm
•    95% of 11535 pockets in apo structures overlap >50% with a predicted pocket.   (96.8% out of 5656 complexed entries)
•    In 82.3% of apo-cases the predicted pocket covers > 80% of the ligand contact atoms!
Binding Site Prediction: Conclusions
Pockets can be used to : • Identify allosteric sites and alternative druggable pockets
De-orphanize (pre-docking): Identification of ligand binding potential and site location for orphan receptors
• Evaluate druggability of protein-protein interaction inhibition by applying the icmPocketFinder to separated protein subunits and evaluating the "pocket" strength
4<£ß-catenin
M2-P53
a,-Antitrypsin deficiency and pathological aggregation
Collaboration with David Lomas, Cambridge
J
• 342Glu to Lys M^Z mutant
•  1:1700 of North European Caucasians
• Risk of death from liver disease during childhood is 2-3%
• Low plasma arantitrypsin level (10-15, 85% retained in liver %), emphysema and higher risk of lung cancer
Z (^-antitrypsin: finding a polymerization inhibitor
Collaboration with David Lomas, Cambridge
^-antitrypsin is retained in ER and forms polymers in vivo
•■SK ■
Lomas et al, Nature 1992; 357: 605-607
Lomas et al, J.Biol.Chem. 1993; 268: 15333-15335
] jj i ü r i íj^í ^ J i» í; 3J'i j o jj
üJj£jÜ JJJ 3/J5   £5'Í3JÍ3
~Lij*pijíijj  1 jj i:3 f'IZlS3 =3 jjj 3 jjj Jj fz± JJ3 j jj j; >=! W3J ä ü
The problem of predicting transient interfaces
•    Proteins do not have open hydrophobic surfaces
•    Previous efforts that looked are residue frequences were not sufficiently predictive
•    We do not know the partner to look for complementarity
A physical idea: Desolvation & entropy
The transient interaction patch may have lower desolvation energy and lower entropy loss upon association.
Both terms can evaluated via atomic surface areas
(Eisenberg&McLachlan, 1986, Abagyan, Totrov, 1994)
Optimal Docking Area: A New Method for Predicting Protein-Protein Interaction Sites
JuanFerna-ndez-Recio,1 MaxTotrov^ťonstantin Skorodumov,^ and Ruben Abagyan1*       Proteins, 2004
* ODA identifies contiguous patches
• Atomic desolvation energy
£« = -1^4
• Optimized on 66 complexes using      ^ protein docking results to include other physical components, like entropy
• Located 80% of the interfaces
• Larger study with 1568 complexes
Bordner, Abagyan, Protdns (2005) 60, 353
Predicting transient protein-membrane interfaces
Irina Kufareva, Collaboration with the Overduin Group
Polar heads
MEMBRANE
Polar heads
Not electrostatic Not hydrophobic
#
*

*.
%
-b lc;;Prg		m lď/p pne; i-* » m m	I
*			I  m m
* ■■-.> s •			m
Transient Protein Interactors Like the Membrane
Conformational Searching for the Global Free Energy Minimum in Internal Coordinates (ICM)
Energy optimization. Internal Coordinate Mechanics
ICM
Goal: Find the global minimum of AGfree=AEvacuum+AGsolv,S
Representation, Energy, Derivatives, Eqs. of Motion
Charmm,CFF,.. {X,J,Z}             1970 s, 1981
ECEPP               {(/)} ,                  1975,1983
ICM                 {0,<Š>,a,b}     1985,1989,1994
Vacuum Energy    (   .       „     „.     ^          (   .,     „ A
™\d,j   d,    edtj J >-»»*{dv   dtJ) ~
Electrostatic Solvation and Entropy
Poisson(/Boltzmann)
J-J entropy Main ICM References: Abagycm, Mazur (1989) JBSD Abagycm et al (1994) "ICM- a n Abagyan, and Totrov,  (1994). "B,
-RT ^S;"ap/a;
residuep
ICM Stochastic Global Optimization
•  Full atom, selected internal coordinates for the area of interest
•  Gradient local minimization after random moves
• Optimally biased, designed, continuous group moves:
•  Double energy scheme           prggroup x _   í pobs m\
•  Reactive history mechanism, stack
Not simulated annealing (T=const), Not Monte Carlo (RHM, no local balance)
The 16-residue a -peptide	
(Scholtz e	tul., 1991}
^	
§**	"ř^A
Abagyan, Totrov, JMB1994, JCP, 1999 Zhou,Abagyan,   JCP, 1999
The 23-residue ßß (Strutters ei a!., 3		a-peptide 996}
	CSE	1
£	rJ	
		
Collective moves for ligand optimization, protein structure prediction and docking in ICM
Automated Homology Modeling for Docking with ICM		
•    Find close template(s) •    Align sequence to the template •    Copy the aligned backbone •    Predict side chains •    Predict loops -   Best Db fragment -   Explicit ICM-local SGO -   Grid simulation •    Refine by ICM SGO •    Predict local reliability (B,) •    Validate by docking a known ligand if possible The only input for ICM-homology builder: a sequence and a template structure	Marsden, Abagyan, submitted, 2003	
íífflJWÍÍ ?Ä]
* Fast docking: atomic ligand to the grid potentials of the receptor
* Method: stochastic global optimization in internal coordinates
*  ICM performance from multiple benchmarks: 60-90% poses are correct
* Speed: takes 20 seconds per compound per processor.
Ou Evaluating Molecular-Docking Methods for Pose Prediction and Enrichment Fac
Hongining Chcn.:':' P;v.l D. Lynr.: Fabrizio GiordanetroJ Timothy Lovell.*T- and Jin LiT
J Chem Inf Model. 2006 Feb;46(1):401-15.
• 1 bhx - alpha thrombin •Rmsd 1.18(0.71), rank 1
ß     • 1 dmp - HIV protease Rmsd 0.90, rank 1
• 1bji-neuraminidase Rmsd 0.89, rank 1
ICM Binding Score
A COMPROMISE between physics and errors
Coordinate errors due to induced fit, charge errors, docking errors, etc.
Sbindins = AEvWint + ^UgStrain + TASlor + «1 ^HBond +
a2AbHBDesol + Cf3Ah,SolE1 + Q^ArAjjp^ + Qř5Qsiz(.
(X !_5 were optimized on a benchmark
- Van der Waals truncated at 4kcal/mole
-   Hbonds calculation is based on lone pairs
-   Penalty for desolvated hydrogen bonding donors/acceptors
-   Electrostatics by Poisson equation (boundary element)
Preparing pdb compounds for docking
Problem"!: compounds/ligands in PDB are not suitable for automated conversion. They lack bond types, formal charges and chiralŕty flags.
Problem 2: compound databases contain only 2D drawings. They need to be converted to 3D.
To foe a PDB ligand follow these steps: ■ Assign correct bond order manually
• Assign correct formal charges manually
• Assign               if necessary (less validated)
• Save is as a mol file or Run the conversion tool
The conversion tool performs these steps:
• Adds hydrogen according for elements, bond orders and formal charges
•  Runs ICM MMFF atom type assignment routine
• Assigns partial electrostatic charges
• Assigns rotatable torsions
• Creates a 3D model by full MMFF94 energy optimization
Preparing compound database for screening
Background:
Preparation of the compound database depends on software used. Some software requires rigid conformations pro-generated. Some will generate 3D structures of ligands and sample them on the fly.
Typically, some kind of index is required to speed up access to the compounds in a very large compound file.
ICM just needs a mol/sdf file with correct drawings
Each molecule from a database will be converted on ttie fly and flexibly docked into a pocket. If the score is lower than a predefined threshold, it will be retained in the "answers'' file.
Things to decide:
1)   To keep (or not) the carboxyls neutral
2)   To charge or not the amino/imidazole groups
3)   Filters (rotatable bonds, donors, acceptors, mass, etc.)
Choosing a grid box and a probe for intial _____________placement_____________
Background:
The docking procedure needs the force field
precalculated as grid potentials.
Also, one need to define the initial placement
of the ligand.
Both decisions can be semi-automated.
Initial orientations will be based on thi probe. If sampling is sufficient, the answer does not depend on the initial position. A good position cai make the search shorter.
Energy maps (or grid potentials) will be calculated inside this box.
jNrpiiBtarthi
in SiliCO with the 'native' coordinates
Q: given an empty pocket and the metabolome, can we identify the native
substrate in-silico?
De-orphanizing a GPCR by docking
Rhodopsin Enrichment =
Cavasotto, Orry et ah, Proteins, 2003
i/pon ligand binding
Receptor flexibility statistics
•    1132 PDB complexes of 65 receptors with > 5 different ligands each analyzed                                    j||       !■ =\
Sidechains                                            j| ,||||,fj
•   A ligand contacts with ~ 10 side chains       ' ^^:~:l::
•   ~75% ligand contact atoms are s.c. (vs 50% in protein core)
•   3 s.c. in 85% of receptors will move by > 1.5A
•    But only 14% severe clashes with 1s.c. and 3% with > 1 s.c.
Backbone
•   ~ 30% receptors had substantial backbone movements: >1A backbone deviations leading to ligand clashes
•   8 elastic deformations, 8 loop, 1 secondary structure
Evaluating side-chain flexibility
•   Identify the side-chains of interest
•   Perform an I CM simulation (~15min)
•  Cluster and space-filter (retain best Ei)
•   Evaluate Boltzmann-weighted RMSD for each sidechain atom
<D2*exp(-AE/kT)>
•   ICM Flexibility tool
Representing receptor by multiple static conformations
Nuclear Receptors: Predicting Specificity
Schapira, Abagyan, Totrov. J. Med. Chetn. 2003
Androgen receptor
Rhodopsin: pocket-flexible docking
fore prediction: Ligand: ~3A SCl>2.A
........r
Ligand: 049A SC:0.19A
to, Orry et al, Proteins, 2003
Predicting Larger Backbone Deformations: Normal Modes and Hinge areas
Coc-Coc spring strengths:
^       iro\6 ,
Derive
u" = normal modes
con = vibrational frequencies
Deformability and hinge regions:
5U = sym(Vu
Idea & Protocol
• Soft, low-res and smooth harmonic model of residue interactions (atomic model does not work)
• Find normal modes U
• Derive deformability
divu/
du = \\Su\\
nt)
Kovacs, Chacon, Abagyan, Proteins, 2004
Choosing Relevant Normal Modes
Cii]ii|ioiuul	1FMO	Receptor A
Adenosine	0.4                  0.6	
B;ikmol	8.5                   I.I	
.Sl;ninis|)oiiiu:	9.8                 7.0	
117	0.8	0.8
lis	0.8	1.1
HK'i	9.2	>]
• Very few normal modes are needed for docking (< 10 )
• These modes are NOT the lowest frequency modes !
• The small number of relevant modes can be combined
Cavasotto, Kovacs, Abagyan, JPC, 2006
Mutants and Mutations
We are all different at 0.1% level (almost every protein has one amino acid different)
8% of liveborns will suffer from a genetically based disorder by age 25
Spontaneous mutations occur continuously (smoking, tanning, eating, age)
Portrait of a Girl Covered in Hair" By Lavinia Fontana (1552-1614)
		Geometry, stability and
		functional effects of
	w^k «	single point mutations
^^		Growing volume of SNP and
-	k 1      JT   wm	Pharmacogenetics data Predicting the effect on
		•   geometry and dynamics
	EUbShI	•    stability changes •    bio-function and binding
	iflL^H^^H^:	•   drug binding "The Sistine Madonna"
	V      tiSfc     ^	by Rafael (1513)
	l*&£r	Look at Pope Sixtus IV
Predicting energy and geometry of mutants
<££" ^fe
■ The largest database of 2141 ordered pairs of structures with a mutation compiled
■ A filtered training set of 1816/2 mutants including "sma/Z-to-big/'witU Protherm AAGs.
■  Cross-Prediction accuracy on the second half ~ 1.1 kcal/mole (correlations.66)
■  Regression from a subset of 317 mutants gives the same prediction accuracy.
■ 20 unfolded state energies derived for each residue
■ Terms: van der Waals, electrostatics, hydrogen bonds, solvation, entropy, residue type constant.
Bordner, Abagyan, Proteins, 2004
kEx^Y = £w,A£: at->7 + (E°x - E\)
Stability prediction without structure
■    Fit simple energy function AAG=EX,-EX for the mutation X-X' to the entire data set without outliers (1768 values).
■    Buried residues: r=0.71 (std=1.21 kc/m); surface res.: r=0.55 (std=1.14 kc/m);
■    Only includes residue energies: useful when no structure is available.
■    Residues with small side chains (glycine, serine, and alanine) most destabilizing
■    Most stabilizing residues are tyrosine, isoleucine and leucine. Agrees with their high occurrence frequency in ß sheets.
■    Also separately fit parameters for buried and surface residues
■    Mutation from Lys to Arg stabilize protein by 0.5-1 kcal/mole
ACDEFGH I KLMNPQR S TVWV
ü'Jp PiBŮÍziíui]
QuickTime™ and a YUV420 codec decompressor are needed to see this picture.
Predicting and redesigning the 15 residues of the triosephosphate isomerase backbone to 8-res. loop
Collaboration with the Wierenga group Structure, PNAS, Prot. Eng. 1993-2002
Predicting Short Loops (benchmark)
Loop Benchmarks: Fiser & Sali, 2003 for 4,8 and 12-residue loops. Friesner et al.
• Challenge: prediction methods break down at 8-12 residue loops.
• 10 years of CASP did not result in X-ray quality loop prediction (NIH)
5a: loop is a separate chain with loose closure conditions.
• Randomized starting conformation
• Run from 5 starts to convergence
• A homology loop benchmark was also compiled and tested
Convergence and Freedom
■ Convergence is a necessary condition of a search
Start
Non-convergent   .
Convergent
■ Set them free.. Departing from a strict loop closure search
	12-residue loops predicted by the ICM optimization after convergence			
	In most cases the prediction is virtually identical to the crystal structure! An, Totrov, Abagyan, 2006			
			iMH	
				
More 12 residue Loops Predictions		
1 1G2Q:A/161:173                                                   1 1 1IXH: /84:96   >                                                  1 f^ it J 1     -^ ~      Hi	1 1HFU: A/410:422               ^J_ 1           -^     2L1	
		*"        1 i»        l
		ť~
	1 lMTP:A/282:294                         7"                        1 \W\ 1   >r>y>r)           1	
States within 10 kc/m from the lowest E
How unique is the crystallographic state?
We observed a large diversity of types of low-energy conformational ensembles.
:Native Yellow: The Prediction     >thers:within 10kcal/mol
Predicting peptide protein association		
		
i i	Docking flexible phosphoryiated peptide to a repector (pYLRVA to V-SRC SH2)	
	WXĚM ^fm^^J-JSI^-^m^:	
	mm w -                         i	
End-guided docking of 27 peptides (8-9) to the HLA receptors, including homology models. Bordner, Abagyan, 2006	Ab initio docking to a receptor of 24 peptides to SH2 and PTB domains Zhou et al. 1998, Folding&Design, 3, 513	
EM-guided Atomic Models
•  Full atom global energy + densityFit optimization. Flexible backbones
• Sampling strategy combines systematic grid and overlapping stochastic searches
• Solvation models with specific geometry built through solvation maps.
•  Benchmark reconstitutions for KcsA tetramer and MscL pentamer show about 1 to 2A RMSD for the contact residues.
Julio Kovacs, Mark Yeager
Protein D< Procedure: ■ Both receptor and ligand are presented by atomic models ■  Convergent Multistart ICM Stochastic Energy	tekiňo	
optimization with pseudo-Brownian moves (JMB, JCC, 1994) and side-chain minization ■ Explicit simulaneous global optimization side-chain and 6 positional variables of candidate solutions Benchmarks GCN4 ab initio helix docking (jcc, 1994) Lysozyme-Antibody {Nature SB, 1994)	äEkt,'. ľ" X.Z.\~>    ".',./' *    f' "   i^m	
Competitions. Docking challenge (Nature sb 1995,90) CAPRI Rounds 1:5		
Local
lization
Nature, SB, 1994
Detailed ab initio prediction of lysozyme-antibody complex with 1.6 A accuracy
Maxim Totrov and Ruben Abagya
s surprisingly cl ition of 1.S7 Ä f. • energy (by 20 I
CAPRI Round 1-2 results
• 3 out of 7 targets —\   predicted correctly
• Refinement of Target 6 Dramatically improves the near-native solution
Proteins, Mendez, ..Wodak, 52, 51-53, 2003
hn|.-!i .h -II it;
■>.;, ■■.■.-■ ..■.■.;.■ ■
■T.--K-. -■
06
• Decent (but not ideal) models for 8 out of 9 targets
•  64-71 % of native contacts                                 .  0ne faMure: T9 with
•  0.4-1A interface RMSD for the best cases              |arge hinge-bending
•  ForT14, Rmsd 0.6A, Rank 1 by energy                 movements
•   Successfully used new scoring function for T14, T18 & T19
•  T19: antibody - prion. Used no CDR bias + NMR model for prion. Fernandez-Recio, Totrov (2005) Proteins, 60, 308
Summary
•  Accurate cross-docking to receptors represented by 'static' grid potentials works in most cases.
•   Receptor flexibility can be predicted in advance
•  A combination of ligand based methods with receptor structure methods can help to de-orphanize receptors.
•   Stochastic global optimization in internal coordinates is a powerful and general method for modeling membrane proteins.
Acknowledgements
Scrípps Group Members
• Julio Kovacs   (normal modes.membrane proteins)
• Irina Kufareva (orphan protein interfaces)
• Adrian Saldanha (antitrypsin)
• William Bisson, Anton Cheltsov (AR inhibitors)
• Giovanni Bottegoni (receptor flexible docking)
Molsoft (www.molsoft.com)
•  Maxim Totrov       (ICM.Ligand Docking)
• Andrew Bordner (peptide docking, Mutations)
• Claudio Cavasotto (kinase docking, flexibility)
• Andrew Orry    (GPCRs)
Former Group Members
• Juan Fernandez-Recio (Barcelona, protein docking)
• Matthieu Schapira (Lyon, TR, NR)
• Jianghong An       (Vancouver, pockets, loops)
Collaborators
AAT: David Lomas, Meera Mallya and the team, Cambridge
Mark Yeager, Scripps
1Ä: Patrick Sexton, Melbourne, Xiaokun Zhang, Burnham
Membrane: Michael Overduin Group
Funding
NIH grant on protein docking
24