Docking and scoring for virtual ligand screening

1
Docking and scoring for virtual ligand
screening Uwe Koch IRBM
Context? Matching receptors and
ligands Scoring Evaluation of matched receptor
ligand couples Receptor flexibility
2
Context
Steps in drug discovery
Target identification validation
Pharmacology Synthesis Clinics
Screening 1000s of actives
Biochemical assays
Lead selection Statistics
Synthesis (Q)SAR
Receptor structure by X-Ray, NMR or modelling
Virtual Screening
Virtual screening
Structure based design Targeted library design

• What you need to know for docking
• the structure of the receptor
• whether this structure is relevant for what you
  are looking for
• what the ligands you are looking for do to the
  receptor, mechanism binding site etc.

If you are looking for drugs the MW lt 600 that is
rmax is often smaller than 10 Å a functional
group essential for binding could measure 3 Å.
How does that fit with the resolution of your
receptor structure.
3
Docking MethodologyOverview
Compounds libraries proprietary ACD CMC etc
Receptor
Ligands
Match receptor and ligand (Docking) Accuracy
speed Score the ligands
4
Docking Methodology Docking

• Docking computational methods for finding the
  best matching between two molecules, a receptor
  and a ligand.
• Sampling across entire range of positional,
  orientational and conformational possibilities
• Various methods have been developed
• Fast shape matching (Dock, Eudock)
• Incremental construction (FlexX, Hammerhead)
• Tabu Search (Pro_Leads, Sfdock)
• (Lamarckian) Genetic algorithm (Gold,
  Autodock3.0)
• Simulated annealing (Autodock2.4)
• Monte Carlo simulations (MCDock, QXP)
• Distance geometry (Dockit)
• Usually RMSD lt 2 Å is considered acceptable

5
Docking Methodology Docking
Incremental construction Modelling receptor
ligand interactions Receptor interaction surface
from crystallographic information
etc. Approximation by a finite set of
interaction centers Fragmentation of ligand into
base fragments Place ligands into active
site by matching interaction centers (first
triples than line matching for pairs) Reduction
of number of solutions by clash test and
clustering Link base fragments in compliance
with a torsional database or a forcefield
H
H
H
H
H.J. Böhm, J. Comput. Aided Molec. Des. 8,
623-632 (1994) M.D. Miller, R.P. Sheridan, S.K.
Kearsley, J. Med. Chem. 1999, 42, 1505-1514
6
Docking Methodology Docking
Match triangles of interaction sites onto
complementary ligand atoms (Flog, FlexX)
The binding site is filled with spheres and the
sphere centers are matched to ligand atoms (fast
shape matching, Dock)
o
o
o
O
O
O
o
o
o
o
o
o
O
O
O
Ewing TJ, Makino S, Skilman AG, Kuntz ID, J
Comput Aided Mol Des. 200115(5)411-28.M.D.
Miller, R.P. Sheridan, S.K. Kearsley, J. Med.
Chem. 1999, 42, 1505-1514
7
Docking Methodology Evaluation

• Evaluation of different methods
• Gold genetic algorithm
• FlexX incremental docking
• Dock fast shape matching
• Docking results for ligands of thymidine kinase
  were compared with the known complex
• structures (one receptor structure was used for
  all dockings)

Best results for Gold, which finds a solution for
all 10 ligands
8
Docking Methodology Scoring
Bound and associated H2O
Receptor
Ligand
Displaced H2O
Affinity DG DH -TDS

• Upon complex formation
• water molecules are released
• receptor and ligand loose degrees of freedom
• interactions between ligand and receptor
• complication mutual compensation of enthalpy and
  entropy

9
Docking Methodology Scoring

• Scoring functions are used
• during docking for optimization of ligand
  orientation and conformation
• for docked ligands to estimate affinity relative
  to other compounds
• Various criteria for the quality of a docking
  function
• ability to find the correct binding mode out of
  alternative docking solutions
• ability to rank related ligands with respect to
  their binding affinity
• ability to select (however weak) inhibitors from
  a large database of inactive compounds
• they should be fast and error tolerant

The inaccuracy of functions used to estimate the
affinity between receptor and ligand is
considered to be the major weakness of docking
programs
H.J. Böhm, J. Comput. Aided Molec. Des. 8,
623-632 (1994) M.D. Miller, R.P. Sheridan, S.K.
Kearsley, J. Med. Chem. 1999, 42, 1505-1514
10
Docking Methodology Scoring

• Scoring methods
• force field based methods (Dock, Gold)
• separate contributions from hydrogen bonds, ionic
  and lipophilic interactions,
• clashes and entropy (nr of rotatable bonds)
• potentials of mean force (PMF, Drug Score)
• description of observed interatomic distances
  and/or frequencies implying
• that these describe favorable/unfavorable
  interactions
• consensus scoring
• combination of multiple scoring functions
  increases hit rates by reducing the
• number of false positives

• two stage ranking
• first filter to limit the number of docked
  conformations,
• second filter to reject false positives

C. Bissantz, G. Folkers, D. Rognan, J. Med. Chem.
2000, 43, 4759-4767 PS Charifson, JJ Corkery, MA
Muecko, WP Walters, J. Med. Chem. 1999, 42,
5100-5109
11
Docking Methodology Scoring
Force field methods E pot E bond E angle E
torsion E oop (internal terms) E
electrostatics E vdW (external terms) E
constraint (special)
12
Docking Methodology Scoring
Potential of mean force or knowledge based
methods Convert experimentally derived
interparticle distances into pair potential
functions Calculate a normalized distance
dependent pair distribution function for atom
types i and j and a
distance dep. pair potential
MJ Sippl, J. Mol. Biol, 1990, 213, 859-883 I.
Muegge, YC Martin, J. Med. Chem., 1999, 42,
2498-2503
13
Evaluation of different docking/scoring
combinations
Three docking algorithms Gold, FlexX and
Dock Scoring algorithms Gold, FlexX and PMF Data
set 990 random compounds and 10 known
ligands Hit Rate age of known ligands among
top 5 scorers Thymidine Kinase Hit
rates do not strongly dependent on docking tool
used Hit rates significantly improved by
consensus scoring High number of false positives
C. Bissantz, G. Folkers, D. Rognan, J. Med. Chem.
2000, 43, 4759-4767
14
Evaluation of different docking/scoring
combinations
Three docking algorithms Gold, FlexX and
Dock Scoring algorithms Gold, FlexX and
PMF Prediction of absolute binding energies for
ligands of thymidine kinase r
correlation coefficient S standard deviation
kJ/Mol No scoring function predicts absolute
free binding energies If two sets of protein
coordinates are used with crystal water r
0.96 Binding of pyrimidines is water mediated
binding of purines is not
C. Bissantz, G. Folkers, D. Rognan, J. Med. Chem.
2000, 43, 4759-4767
15
Docking Receptor flexibility
Proteins, like their ligands, exist in a range of
conformational states and ligands will associate
with the most favorable conformers Small scale
motions Najmanovich et al compared a set of
complexed and uncomplexed protein structures
Generally small number of residues undergoes
conformational changes LysgtArg,Met,GlngtGlu
,Ile,LeugtAsn,Thr,Val,Tyr,Ser,His,AspgtCys,Trp,Phe L
arge scale motions Low energy transitions
which are often critical for docking (Hinge type
motions) Protein disorder Transition from
disordered to ordered on ligand binding (larger
capture radius of unstructured protein)
Structural stability of binding sites Binding
sites are often chracaterized by regions of very
high structural stability and by the presence of
regions with low stability, often loops. (Freire)
R Najmanovich, J Kuttner, V Sobolev, M Edelman,
Proteins, 2000, 39, 261-268 BA Shoemaker, JJ
Portman, PG Wolynes, PNAS, 2000, 97, 8868-8873 I
Luque, E. Freire, Proteins, 2000, Suppl. 4, 63-71
16
Receptor flexibility Computational approaches

• Computational approximations to protein
  flexibility
• Generate an ensemble of receptor conformations
  and dock to those (NMR, X-ray, MD)
• Composite grids from multiple X-ray or NMR
  structures of protein ligand complexes
• (Knegtel, Broughton)
• FlexE averages conserved coordinates rather
  than grids. It is able to mix rotamers thus
• creating new combinations but can not handle
  domain motions (Claussen)
• Other approximations
• Methods which allow hinge-bending motions of
  domains, subdomains or substructural parts with
• rotations around points have been developed
  (Sandak)
• Side-chain flexibility can be modelled
  according to rotamer libraries or approximatively
  
• by a surface belt of non-penalized
  penetration area (Palma)
• Knegtel RMA, Kuntz ID, Oshiro CM., J Mol Biol
  1997266424440.
• Broughton HB. J Mol Graph 200018247257.
• Claussen H, Buning C, Rarey M, Lengauer, T., J.
  Mol Biol 2001308377395.
• Sandak B, Nussinov R, Wolfson HJ., J Comput Biol
  19995631654.

17
Receptor flexibility Computational approaches
Composite grids from NMR, X-ray or
modelling Superposition of HIV-protease with 5
peptidic inhibitors
Different side-chain conformers
not every ligand can be docked

to every enzyme structure

due to steric clashes
18
Receptor flexibility Computational approaches
Composite grids average ensembles of
conformations
Geometry weighed average Superposition of
residues close to the binding site Calculation
of average positions and standard deviations for
individual atoms Generation of a single
structure For atoms with standard deviation
above a threshold the original positions are
kept For less flexible atoms the average
position is used Disordered atoms are considered
to be volumeless (no repulsion) The vdW and
coulomb part of the potential is calculated for
points outside the vdW volume Grid interpolation
Knegtel RMA, Kuntz ID, Oshiro CM, J. Mol. Biol.
1997, 266, 424-440
19
Receptor flexibility United protein description
Recombines elements from different ensemble
structures (conformations or point
mutations) Uses a graph search to find the
optimal combination of ensemble structures for
each ligand United protein description
Superposition of receptor structures. Merge
similar parts and treat dissimilar parts as
separate alternatives (instances)


20
Receptor flexibility United protein description
Recombines elements from different ensemble
structures (conformations or point
mutations) Uses a graph search to find the
optimal combination of ensemble structures for
each ligand United protein description
Superposition of receptor structures. Merge
similar parts and treat dissimilar parts as
separate alternatives (instances) Incompatibilit
y graph Two instances are incompatible if they
can not be realized simultaneously Criteria
logic, geometric and structural
Logic three conformations of the same residue
Geometric clash between residues
21
Receptor flexibility United protein description
Recombines elements from different ensemble
structures (conformations or point
mutations) Uses a graph search to find the
optimal combination of ensemble structures for
each ligand United protein description
Superposition of receptor structures. Merge
similar parts and treat dissimilar parts as
separate alternatives (instances) Incompatibilit
y graph Two instances are incompatible if they
can not be realized simultaneously Criteria
logic, geometric and structural Nodes
instances Edges incompatible instances Valid
structure unconnected set of instances
22
Receptor flexibility Computational approaches
United protein description Superposition of
receptor structures. Merge similar parts and
treat dissimilar parts as separate alternatives
(instances) Incompatibility graph Two instances
are incompatible if they can not be realized
simultaneously Criteria logic, geometric and
structural Nodes instances Edges
incompatible instances Valid structure
unconnected set of instances Docking The
incremental construction algorithm is used to
place the ligand fragment by fragment into the
active site Evaluation of interactions between
partially placed ligand with all instances of the
united Scoring of each solution Protein
structure which suites best the ligand fragment

Claussen H, Buning C,
Rarey M, Lengauer, T., J. Mol Biol
2001308377395.
B
S
B
S
S
S
S
S
23
Receptor flexibility Computational approaches
FlexE Evaluation 10 protein ensembles and 105
crystal structures 1 homology model For
comparison all ligands were docked separately
into each structure of the ensemble Aldose
reductase Superposition of 4 structures, 3 with
ligand and 1 apoenzyme Two different
conformations for tolrestat and
zopolrestat Sorbinil does not contact
the flexible region RMSDs FlexE vs Cross
docking Sorbinil can be docked into all
receptors (see above) Using rigid docking the
other ligands can be docked to their receptor
only Claussen H, Buning C, Rarey M, Lengauer,
T., J. Mol Biol 2001308377395 Urzhumtzev A. et
al., Structure 1997, 5, 601-612
Phe122
flexible region
Sorbinil
24
Examples
Examples of bioactive molecules which were
identified, designed or optimized with docking
methods Structure Target Method Aldose
reductase inhibitors 3D database
searching Diabetes HIV protease (AIDS) 3D
database searching Structure based
design Carboanhydrase (Glaucoma) Structure
based design Dorzolamide Thrombine Combin
atorial docking, de novo design 2) Van
Drie JH, Lajiness MS, Drug Discovery Today 1998,
3, 274-283 3) Drugs, New, Perspect. 1995, 8,
237. 4) Bohm HJ et al, J. Comput.-Aided Mol. Des.
1999, 13, 51-56