Transcript Conformational Searching

Chemical
Computing
Group Inc.
Docking and
Multi Fragment Search
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
1
Outline
Part A: Docking
Docking is based on a more complete and quantitative description and
analysis of ligand-protein interactions compared to Ph4 queries.
• Examines the ensemble of all actual and potential interactions
between ligands and their binding sites while optimizing the
geometry of the complex.
• Typically involves forcefield minimization and eventually further
terms to predict binding poses.
Part B: Multi-Fragment Search
• Attempts to predict poses of smaller functional groups rather than
whole molecules.
• The accuracy of those predictions increases while the search space
is less constrained to a given set of molecules.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
2
Docking
Objective
MOE-Dock identifies favorable poses of flexible ligands in rigid
binding sites of macromolecules, typically proteins. MOE-Dock consists
of a “toolbox” which offers different routines for conformational
sampling, placement and scoring. This enables the user to optimize
his workflow for a given target.
Note: The MOE-Dock algorithm
was completely rewritten in
MOE 2005.06.
Receptor
3D ligand
Annotation
Torsion Rules
Placement
3D conformations
Scoring
PH4 Filter
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
3
Placement Methods
• The Alpha PMI method generates poses by
aligning ligand conformations' principal moments of
inertia to a randomly chosen subset of alpha sphere
dummies in the receptor site. This method is
preferred for tight pockets.
• The Alpha Triangle placement method (default)
derives poses by random superposition of ligand
atom triplets and alpha sphere dummies in the
receptor site.
• The Triangle Matcher method generates poses by
aligning ligand triplets of atoms on triplets of alpha
spheres in a more systematic way than the Alpha
Triangle method. (most complete screen of poses)
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
4
Scoring Methods
• The Affinity dG scoring function estimates
the enthalpic contribution to the free energy
of binding using a linear function of
hydrophobic, ionic, hydrogen bond and
metal ligation terms.
• The DephtHB scoring function is a linear
combination of two terms:*)
- the burying of the ligand and
- the hydrogen bond effects.
• The London scoring function (default) estimates the free energy of
ligand binding from a given pose. The functional form is a sum of
energy term (ligand flexibility, H-bonds, desolvation) and geometric
imperfections.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
5
Affinity dG Scoring Function
ΔG is the sum of terms (fit from ~100 pKi complexes)
G  CH
f
H
hbonds : i  j
(rij )  CM
f
M
metal lig: i  j
(rij )  CI
q q
ionic: i  j
i
j
f I (rij )  CB
f
B
contacts: i  j
• H-bonds are between donor
and acceptor heavy atoms
• Metal ligations are between
transition metal and –O, –N and –S
• Ionic contacts are between
functional groups (not just ions)
• Contacts are between heavy atoms of receptor and ligand
O(-1)
O(-1/2)
C(+1)
O(-1)
O(-1)
O(-1)
(+2)S
(+1)P
N(+1)
O(-1/2)
(rij )
O(-1)
O(-1)
N(+1/2)
N(+1/2)
N(+1/2)
Na+1
N(+1)
Zn+2
N(+1/2)
N(+1/2)
1,2
1
Functions f decrease with distance
• fI and fB functions have
7.5 Å cutoff
• Close contacts not penalized
0,8
0,6
0,4
0,2
0
1
1,25 1,5 1,75
2
2,25 2,5 2,75
fH
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
3
fM
3,25 3,5 3,75
fI
4
4,25 4,5 4,75
5
fB
6
http://webpages.ull.es/users/bioquibi/temascompletos/InteraccionesNC/i
nicial/radio_de_van_der_waals.htm
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
7
London dG Scoring Function
• Sum of terms intended to estimate ΔG of binding
G  c  E flex 
c
hbonds
HB
f HB 
c
mlig
M
 D
fM 
i
atoms i
- c is average entropy loss/gain due to rotational/translational motion
- Eflex is entropy loss due to conformational flexibility (ligand topology only)
• H-bond fHB measures geometric imperfections
- cHB is H-bond maximum energy
- Distance, in-plane and out-of-plane angles into account
- Functional forms and distributions taken from MOE contact statistics
• Metal ligation fM measures geometric imperfections
- cM is metal ligation maximum energy
- Landis’s VALBOND sd3 hybridization potential for angles
• ΔDi estimates desolvation energy of each atom i
- HB and M terms recover desolvation energy for hydrophilic atoms
- Desolvation energy is the main component of the model
- No ionic effects are part of the model (difficult to incorporate consistently)
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
8
London dG Desolvation Model
A
A
+
B
ΔDi
i
B
i
solvent
For atom i with radius Ri estimate desolvation energy by integrating London
dispersion forces over solvent space


6
6
Di  ci R   | u | du   | u | du
uB
uAB

3
i
Implementation:
• Radii are OPLS-AA van der Waals radii plus 0.5 Å
• Coefficients ci are assigned for each of ~12 atom types (e.g., Csp3/Csp2)
• Integration approximated with GB/VI type integrals over solutes A and B
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
9
Molecular Docking Updates
Placement
• AlphaTriangle and TriangleMatcher
placement methods updated for speed
Scoring
• New London dG scoring function
for binding affinity estimation
• New DepthHB scoring function
for ranking poses (no affinity estimate)
Ligand strain energy no longer used in pose selection
• Strain energy rarely helps and often hurts pose selection
Structural binding affinity database
• A new database of 500+ complexes and experimental pKi data
($MOE/sample/mol/complex.mdb)
• Ligands verified for correctness, receptors verified for close contacts
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
10
Exercise: MOE-Dock – Binding Site Preparation
Perform a docking study with a kinase receptor and a ligand using alpha sphere
dummy atoms defining the binding site.
1. Open 1ke6_rec.moe
(MOE | File | Open)
2. Add hydrogens before docking
(MOE | Edit | Hydrogens | Add Hydrogens)
and calculate the partial charges
(MOE | Compute | Partial Charges)
with Amber99 forcefield*).
In addition, MOE-Dock requires definition of a
docking site in the receptor. This can be done in various ways:
• Select the ligand (it will be ignored during docking),
• Select the residues of the pocket,
• Select dummy atoms generated by MOE’s Site Finder
3. Open the Alpha Site Finder (MOE | Compute | Site Finder) to isolate the
active site. Identify and select the relevant site and create Dummies.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
11
MOE-Dock Panel
MOE-Dock can be launched with (MOE | Compute | Simulations | Dock).
Choose between various options for receptor, binding site and ligand
atoms definitions
Specifies which
atoms in MOE are
used as receptor.
Verification of the
selected atoms.
Specifies which
atoms in MOE are
used as docking
site.
Ph4 query can be
loaded.
Specifies which
atoms in MOE (or
mdb-file) are used
as ligand(s).
Specifies the
placement and
scoring method
Visualize backbone,
binding site residues
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
12
Exercise: MOE-Dock I
4. Launch MOE-Dock (MOE | Compute | Simulations | Dock).
5. Specify an output database file name (e.g. dock_tm_lon.mdb).
6. Ensure that Receptor is set to "Receptor
Atoms".
7. Ensure that Site is set to “Dummy Atoms".
8. Use the Render button to isolate the
docking site.
9. Switch Ligand to “MDB File” and select
the conformation database of the ligand
(conf_out_1ke6.mdb); disable
Conformational Search.
10. Select Placement and Scoring Methods:
Triangle Matcher and London dG
11. Click “OK”.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
13
Exercise: MOE-Dock II
12. Examine the results in the output database (dock_tm_lon.mdb) :
The poses of each ligand (mseq) are already
ranked according to the “best” score, i.e.
lowest S value at the top.
Scoring function
[kcal/mol]
Compare the position of the original,
co-crystallized ligand with the docked
conformations of the database.
13. Load the original ligand (copy the
structure from the crystal field in
1ke6_crystal.mdb into the MOE
window), select it, and color the
ligand orange to better distinguish the
crystal ligand from the docking poses.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
Distance based
scoring function
14
Exercise: MOE-Dock III
14. Use the Browser in the Database Viewer (DBV | File | Browser) to step
through the output database, one by one.
15. Select the first pose in the MOE window,
and render it by (MOE | Render | Stick).1)
16. Evaluate the position of the best scored
pose (lowest S) and the pose with the
lowest RMSD2) with the co-crystallized
ligand (orange).
Best scored pose:
S = -10.26 kcal/mol
RMSD = 1.6 Å
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
Lowest RMSD (rank 88):
S = -8.1 kcal/mol
RMSD = 0.69 Å
15
Validation of the Docking Scores I
Placement methodologies in MOE may be compared by calculating the
RMSDs of the docked poses and the X-ray structure of the ligand.
Ideally, the correlation plot of the ASE score vs. the RMSD should show
a clear trend from lower left to upper right.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
3 Docking Simulations with the Alpha Triangle Placement
Methodology
-2
-3
-4
ASE score
The figure shows three
independent docking runs with
the Alpha PMI placement
methodology. Smallest RMSDs
(most similar to the crystal pose)
correlate with the highest
scores.
As search space is not covered
systematically results may differ
slightly from one simulation to
another.
-5
-6
-7
-8
-9
0
1
2
3
4
5
6
7
8
9
10
RMSD [A]
Red diamonds: 1st run; green dots: 2nd run;
blue squares: 3rd run
16
Validation of the Docking Scores II
Successful scoring functions will rank lowest RMSD placements at the
top with about 80 % accuracy. The plot compares lowest E versus lowest
RMSD placements for three runs using different placement strategies. In
this case lowest RMSD solutions are ranked quite low (11-30). Since the
implementation of docking routines in MOE is in the form of a toolbox
derive an optimal combination of tools for each project.
Docking Methods
Energy [kcal/mol]
-5
Alpha Triangle
-4
(1)
Alpha PMI
(11)
(1)
Triangle Matcher
Triangle Matcher
-3
(25)
(30)
-2
Alpha
Triangle
-1
1.0
1.5
sorted by lowest E
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
(1)
Alpha PMI
2.0
2.5
3.0
RMSD [A]
sorted by lowset RMSD
3.5
4.0
4.5
(Rank indicated in brackets)
17
Validation of the Docking Scores III
Alpha PMI
10
8
RMSD
To compare the performance of different
placement strategies 5 different CDK2 ligands
are cross-docked into the 5 different pockets to
give an idea about the variability of results
obtained with different methods. This also
shows that some pockets tend to perform better
than others thus also validate the pockets
before starting a large scale screening.
1ke6
1ke7
6
1ke8
4
1oiy
1oit
2
0
1ke6 pocket 1ke7 pocket 1ke8 pocket 1oiy pocket
1oit pocket
protein pockets
As a rule of thumb alpha PMI tends to provide
best results for small pockets.
Alpha Triangle
Triangle Matcher
8
6
1ke6
5
1ke7
4
1ke8
3
1oiy
2
1oit
1
0
1ke6 pocket 1ke7 pocket 1ke8 pocket
protein pockets
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
1oiy pocket
1oit pocket
RMSD
RMSD
7
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
1ke6
1ke7
1ke8
1oiy
1oit
1ke6 pocket 1ke7 pocket 1ke8 pocket 1oiy pocket
1oit pocket
protein pockets
18
Multi-Fragment Search I
Objective
Multi-fragment search (MFS) attempts to determine preferred positions
and interactions for certain functional groups in a given receptor.
Docking small fragments of limited flexibility can be achieved at higher
accuracies than docking entire molecules. Results may be used as
input to a “de novo” design procedure or to derive a pharmacophore.
Miranker, A., Karplus, M. Functionality Maps of Binding Sites:
A Multiple Copy Simultaneous Search Method. Proteins:
Structure, Function, and Genetics, 1991, 11, 29-34
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
19
Multi-Fragment Search II
Methodology
The active site of a receptor is randomly populated with many copies of
a fragment and then energy minimized (fragments do not interact).
• Removes duplicates and
calculates interaction energies
with or without solvation effects
• Fragment output is sent to a
molecular database for analysis
• All or part of a receptor can be
held fixed during minimization
Input system
Save System
mfss_orgsys.moe
Place Fragments
Save Fragments
mfss_orgcop.moe
Fix Atoms
Minimize Classes
convergence
Minimize Receptor
Delete Duplicates
Save System
Save Fragments
Save Receptor
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
mfss_mincop.moe
mfss_output.mdb
mfss_minrec.moe
20
Exercise: Multi-Fragment Search I
First, prepare the receptor for Multi-Fragment Search:
1. Open the receptor file 1ke6_rec.pdb
(MOE | File | Open)
2. Add hydrogens
(MOE | Edit | Hydrogens | Add Hydrogens)
and calculate the partial charges
(MOE | Compute | Partial Charges)
with the Amber99 forcefield. After calculation
use the Potential Setup panel (MOE | Window | Potential Setup) to choose
an appropriate forcefield for receptor-ligand interactions (e.g. MMFF94x).*)
3. Select the atoms of the active site to guide placement of fragments. Be careful
to pick only the solvent exposed atoms of the active site (to improve the speed
of fragment placement). Use Site Finder dummies in absence of any other
ligand information.
Select the ligand atoms/ Site Finder dummies and extend the selection to 4.5Å
(MOE | Selection | Extend | Nearby (4.5A))
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
21
Exercise: Multi-Fragment Search II
4. Delete the chain of the ligand/alpha spheres in the SE (without deselecting the
pocket*).
5. Choose (Compute | Simulations | MultiFragment Search) in the MOE
window to open the MultiFragment Search panel.
6. Select the fragments that should be included:
E.g. acetate ion and phenol.
7. Click “Next”.
Fragment list contained in
the fragment database.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
22
Exercise: Multi-Fragment Search III
8. The next window contains parameters
for minimization (stay with defaults).
Click “Next”.
9. Before starting the MF-search choose
several output file names (stay with
defaults).
10. Click “Start”.
The selected fragments will be flooded
into the pocket and minimized. This
may take some time.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
23
Exercise: Multi-Fragment Search IV
11. Examine the results:
Open the *_orgcop.moe into the MOE window.
This is the molecular system after initial
placement of the fragments together with the
receptor but prior to energy minimization (the
original copies).
Open the *_mincop.moe into the MOE window.
These are all the resulting unique fragments and
the receptor after energy minimization (the
minimized copies).
The same picture will result if all entries from the
*_output.mdb are copied into the MOE
window.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
24
Exercise: Multi-Fragment Search V
12. Open the *_output.mdb
This fragment database contains 58 entries of
the acetate and 74 entries of the phenol
fragment after minimization including
interaction energy information.
Sort the entries according to different criteria:
best binding, lowest potential, best binding in
fragment class, or lowest potential in fragment
class.*) (All values are in kcal/mol.)
13. Open the receptor again (please use the
prepared pocket.moe file for examination;
otherwise the surface etc. has the be
generated).
Copy all entries of one fragment into the
receptor pocket to get a first impression if there
are some preferred regions (clusters) of the
fragment. In this phenol example there are two
main clusters – one inside and one more
exterior to the pocket.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
25
Exercise: Multi-Fragment Search VI
14. Examine each placement individually using
the Browser in the Database Viewer (BDV |
File | Browser) and step through the output
fragments. Each fragment will be placed in
the receptor context. (delete the fragments
of point 13 before.)
15. If a placement may be a suitable starting
point for “de novo” design, keep the entry in
the MOE window by clicking Keep in the
Browser panel.
16. Compare also the positions of the
fragments with the original ligand position.
With knowledge of preferred fragment
positions it may now be possible to join the
fragments to new ligands, build
combinatorial libraries, or derive
pharmacophores.
Copyright © 2006 Chemical Computing Group Inc.
All Rights Reserved.
26