2. Docking Challenge
• Identification of the ligand’s correct
binding geometry in the binding site
(Binding Mode)
• Observation:
– Similar ligands can bind at quite
different orientations in the active
site.
3. Two main tasks of Docking
Tools
• Sampling of conformational (Ligand)
space
• Scoring protein-ligand complexes
4. Rigid-body docking algorithms
• Historically the first approaches.
• Protein and ligand fixed.
• Search for the relative orientation
of the two molecules with lowest
energy.
• FLOG (Flexible Ligands Oriented on
Grid): each ligand represented by up
to 25 low energy conformations.
5. Introducing flexibility:
Whole molecule docking
• Monte Carlo methods (MC)
• Molecular Dynamics (MD)
• Simulated Annealing (SA)
• Genetic Algorithms (GA)
Available in packages:
AutoDock (MC,GA,SA)
GOLD (GA)
Sybyl (MD)
6. Monte Carlo
• Start with configuration A (energy E A)
• Make random move to configuration B
(energy EB)
• Accept move when:
EB < EA or if
EB > EA except with probability P:
P = exp( − [ E A − E B ] kT )
7. Molecular Dynamics
• force-field is used to calculate forces on
each atom of the simulated system
• following Newton mechanics, calculate
accelerations, velocities and new
coordinates from the forces.
(Force = mass times acceleration)
• The atoms are moved slightly with respect
to a given time step
8. Simulated Annealing
Finding a global minimium
by lowering the temperature
during the Monte Carlo/MD simulation
9. Genetic Algorithms
• Ligand translation, rotation and
configuration variables constitute the
genes
• Crossovers mixes ligand variables from
parent configurations
• Mutations randomly change variables
• Natural selection of current generation
based on fitness
• Energy scoring function determines fitness
10. Introducing flexibility:
Fragment Based Methods
• build small molecules inside defined
binding sites while maximizing
favorable contacts.
• De Novo methods construct new
molecules in the site.
• division into two major groups:
– Incremental construction (FlexX, Dock)
– Place & join.
11. Placing Fragments and Rigid
Molecules
• All rigid-body docking methods have in
common that superposition of point sets is
a fundamental sub-problem that has to be
solved efficiently:
– Geometric hashing
– Pose clustering
– Clique detection
12. Geometric hashing
• originates from computer vision
• Given a picture of a scene and a set
of objects within the picture, both
represented by points in 2d space,
the goal is to recognize some of the
models in the scene
13.
14. Pose-Clustering
• For each triangle of receptor compute
the transformation to each ligand
matching triangle.
• Cluster transformations.
• Score the results.
15. Clique-Detection
•
•Nodes comprise of matches between protein and ligand
•Edges connect distance compatible pairs of nodes
•In a clique all pair of nodes are connected
16. Scoring Functions
• Shape & Chemical Complementary
Scores
• Empirical Scoring
• Force Field Scoring
• Knowledge-based Scoring
• Consensus Scoring
17. Shape & Chemical
Complementary Scores
• Divide accessible protein surface into
zones:
– Hydrophobic
– Hydrogen-bond donating
– Hydrogen-bond accepting
• Do the same for the ligand surface
• Find ligand orientation with best
complementarity score
19. Empirical scoring
∆G = ∆G0 + ∆Grot × N rot Loss of entropy during binding
+ ∆Ghb ∑ f (∆R, ∆α ) Hydrogen-bonding
neutral. H −bonds
+ ∆Gio ∑ f (∆R, ∆α ) Ionic interactions
ionic −int .
+ ∆Garom ∑ f (∆R, ∆α ) Aromatic interactions
arom.int
+ ∆Glipo ∑ f (∆R, ∆α ) Hydrophobic interactions
lipo.cont.
20. Force Field Scoring (Dock)
Aij Bij
lig prot
qi q j
Enonbond = ∑∑ 12 − 6 +c
i j ij
r rij r ij
Nonbonding interactions (ligand-protein):
-van der Waals
-electrostatics
Amber force field
21. Knowledge-based Scoring
Function
Free energies of molecular interactions
derived from structural information on
Protein-ligand complexes contained in PDB
Boltzmann-Like Statistics of Interatomic
Contacts.
[
P ( p , σ l )= Pref exp − βF ( p , σ l )
σ σ ]
22. Distribution of interatomic distances is converted
into energy functions by inverting Boltzmann’s law.
F P(N,O)
23. Potential of Mean Force (PMF)
i σ seg (r )
ij
Fij (r ) = −kB T ln fVol _corr (r ) ij
σ bulk
σ ij
seg
(r ) Number density of atom pairs of type ij
at atom pair distance r
σ ij
bulk
Number density of atom pairs of type ij
in reference sphere with radius R
25. Virtual screening by Docking
• Find weak binders in pool of non-
binders
• Many false positives (96-100%)
• Consensus Scoring reduces rate of
false positives
26. Concluding remarks
Scoring functions are the Achilles’ heel
of docking programs.
False positives rates can be reduced using several
scoring functions in a consensus-scoring strategy
Although the reliability of docking methods is
not so high, they can provide new suggestions for
protein-ligand interactions that otherwise
may be overlooked
Notes de l'éditeur
Explain docking is fitting ligand into the receptor, steric and electrostatic match
GA available in GOLD (Genetic Optimisation for Ligand Docking) MD a force field is used to calculate the forces an each atom. Following Newton Mechanics, velocities and accelerations are calculated Atoms are moved with respect to a time step MC local moves of atoms are performed randomly SA optimisation technique: 1 starting form conformation A with energy/score Ea 2 calculate random local move to configuration B with Eb 3 Accept on the Basis of Metropolis criterion: a) if Eb is lower than Ea b) with probability P=exp(-[Eb-Ea]/kT)
The first group places a single fragment, or seed, in a binding cavity, and in a stepwise manner, other groups are attached to the growing structure. The second group of methods places key functional groups into a binding site and then attempts to connected these together into a single structure. GROW, LUDI, SPROUT
first applied to molecular docking program sby Fischer, Norel Nussinov Wolfson (1993) CPM93, 20-34 and Fischer, Lin, Wolfson and Nussinov (1995) J. Mol. Biology 248, 459-477.
The distribution of interatomic distances is converted into energy functions by inverting Boltzmann’s law. It is not Boltzmann’s law that determines the distribution observed in the PDB in the first place. “ An ensemble of structural parameters obtained from chemically different compounds in different crystal structures does not even remotely resemble a closed system at thermal equilibrium” Assumption: pair interaction are independent.
Incremental construction 1 adding next fragment in all possible conformations to all placements befor 2 search for new protein-ligand interactions 3 optimising ligand position to improve interaction and reduce strain 4 select a subset of placement with high score 5 clustering of these placements Bohm = Empirical scoring; fit coefficients of physical contributions (LUDI, FlexX) Knowledge-Based Scoring = statistical preferences can be derived between protein And ligand that are similar to potentials of mean field Force Field or Energy scoring= speaks for itself Dock score best for apolar, FlexX best for polar
Creates a negative image of binding surface, matching of distances between receptor negative image and ligand positive image
Some of the failures represent protein-ligand interactions are not expressed in algorithm, e.g., those between electron rich and electron dense groups Also, has no entropic element so couldn’t predict binding energies