Toward Fully Flexible Docking

1
Toward Fully Flexible Docking

• Bashir S. Sadjad
• bssadjad_at_uwaterloo.ca
• School of Computer Science, University of
  Waterloo, Canada
• Simulated Biomolecular Systems, Toronto, Canada

2
Outline

• How does Aspirin work?!
• Structure-based rational drug design
• What is Docking?
• Current approaches and software packages
• Why is protein flexibility important?
• One step forward in flexible docking ...

3
What Is Aspirin?

• Acetosal
• Route of administration Oral
• Structure
• NOTE Images without reference are taken from
  public domain (mostly Wikipedia)

4
How does Aspirin Work?

• A painkiller, also used against fever
• Reduce production of Prostaglandin, Thromboxane
• Prostaglandin
• Prostaglandin binds to some trans-membrane
  proteins of spinal neuron cells causing pain.

5
How does Aspirin Work? (cont'd)

• Cyclooxygenase is inhibited (breaking the
  pathway).
• Thromboxane pathway
• Many of drugs interfere in a protein function.
• Aspirin effect is irreversible, we like
  reversible!

6
Enzymes (Review)

• Enzymes are great catalyzers (may speed up a
  reaction by 5 to 17 order of magnitude).
• An explanation of how they work

7
Enzyme Inhibition

• Change of shape or chemical properties of active
  site on an enzyme.
• Example blocking active site in HIV protease by
  ritonavir.

8
Outline

• How does Aspirin work?!
• Structure-based rational drug design
• What is Docking?
• Current approaches and software packages
• Why is protein flexibility important?
• One step forward in flexible docking ...

9
How Was Aspirin Discovered?

• Roots can be traced back to 5th century BC, in
  Hippocrates notes regarding a bitter powder
  extracted from willow bark easing pain.
• Similar references found in other ancient notes.
• Willow produces salicin
• which acts similarly to
• aspirin in the body.
• Main discovery was done in 19th century.

10
Rational Drug Design

• The discovery was by trial and error anyway!
• How a similar drug may be discovered in 21st
  century?
• Identify related pathways
• Select target proteins
• Identify active sites and allosteric sites
• Try to inhibit target proteins
• Inhibition may happen by ligand binding

11
Rational Drug Design (Example)

• Zanamivir (a ligand) used for treatment of
  Influenza virus.
• Inhibits Neuraminidase, an enzyme on the surface
  of Influenza virus.
• One of the first rationally designed drugs, by
  Biota (1989).
• Marketed by GSK (1999).

12
Outline

• How does Aspirin work?!
• Structure-based rational drug design
• What is Docking?
• Current approaches and software packages
• Why is protein flexibility important?
• One step forward in flexible docking ...

13
High Throughput Screening (HTS)

• Once the target protein is purified a library of
  ligands might be tested against it.
• The size of a practical library is in 105 106
  range.
• Ligands active against the target protein can be
  selected by automated mechanisms.
• This requires significant resources including
  expensive labs.

14
Virtual HTS

• How if we can predict the HTS result by a
  computer program?
• This is Virtual High Throughput Screening.
• One way to do this is by simulating the binding
  process using properties of involved molecules.
• Free Energy of Binding determines the affinity.

15
Free Energy

• Each conformation and binding mode has a specific
  free energy

16
Docking

• Determine the best binding mode
• An approximation of free energy is used (scoring
  function).
• The search engine finds the minimum of the
  scoring function.
• Carbonic anhydrase and a bound ligand

17
Outline

• How does Aspirin work?!
• Structure-based rational drug design
• What is Docking?
• Current approaches and software packages
• Why is protein flexibility important?
• One step forward in flexible docking ...

18
Classification Criteria

• Type of scoring function
• Force-field based
• Empirical
• Statistical
• Search method
• Systematic
• Stochastic
• Degree of ligand-protein flexibility

19
Scoring

• Approximating reality a bad approximation won't
  work even with the best search method.
• A general form for an empirical scoring function

20
Search

• Stochastic and heuristic
• MCDOCK (simulated annealing)
• AutoDock2 (simulated annealing)
• GOLD (genetic algorithm)
• Directed
• FlexX (incremental construction)
• eHiTS (exhaustive search)
• Combined
• Glide (systematic pose gen. stochastic
  optimization)

21
eHiTS

• eHiTS approach 4
• Ligand fragmentation
• Fragment rigid-dock
• Fragment matching
• Ligand reconstruction
• Local optimization

22
Example (Rigid Docking)

• Each fragment is docked in cavity.
• For sufficient accuracy a fine sampling should be
  done.

23
Example (Matching)

• All fragments are scored.
• A diverse set of matching poses with high scores
  are selected.
• A full ligand pose is generated from each
  matching set.

24
Outline

• How does Aspirin work?!
• Structure-based rational drug design
• What is Docking?
• Current approaches and software packages
• Why is protein flexibility important?
• One step forward in flexible docking ...

25
Protein Structure and Ligand Binding

• Protein structure may significantly be changed by
  ligand binding.
• Calmodulin (a calcium-binding protein)
• Movie http//molmovdb.org/cgi-bin/morph.cgi?ID78
  252-5656

26
The Allosteric Effect

• Binding of ligands may regulate the protein
  function.
• Example binding of oxygen and carbon-dioxide to
  hemoglobin

27
Binding Site Flexibility

• Ligand binding changes the binding site of the
  protein. This is called induced fit.
• In many of protein-ligand complexes in PDB, the
  cavity surrounds the ligand with a small open
  part.
• Rigid treating of binding site (as done by most
  docking programs), makes binding energy
  prediction difficult.

28
Example (Ligand Binding)

• Conformational change at the binding site of
  Renin.

29
Example (Cavity Closure)

• L-Arabinose-binding protein complexed with
  L-Arabinose. (PDB 1ABE)

30
A Note on Structure-Function Assumption

• Amino Acid Structure Function assumption.
• Consider a highly hydrophilic protein sequence,
  is it folded in water? Does it have any
  functions?
• Indeed it is not in a single folded state but it
  can be functional! There are functional
  intrinsically unstructured proteins.
• They may fulfil different tasks and have
  different fold for each task.

31
Example (Unstructured)

• The pKID domain of CREB protein, complexed with
  KIX domain of CREB-binding protein.

32
Example (Structural Change)

• The TAZ1 domain of CREB-binding protein complexed
  with two different domains.

33
Outline

• How does Aspirin work?!
• Structure-based rational drug design
• What is Docking?
• Current approaches and software packages
• Why is protein flexibility important?
• One step forward in flexible docking ...

34
Truly Flexible Docking

• A truly flexible docking application is in fact a
  folding program!
• eHiTS is an ab-initio method folding complexity
• Different types of protein mobility
• Movement of large domains
• Multiple conformations observed in a few residues

35
Movement of Domains

• Patterns of domain movement
• Ribose-binding protein movie (2DRI, 1URP)
• http//molmovdb.org/cgi-bin/morph.cgi?ID645772-17
  065

36
Conformations of a Few Residues

• Acetylcholinesterase (PDB 2ACE, 1EVE, 1VOT, 1ACL)

37
Truly Flexible Docking

• A truly flexible docking application is in fact a
  folding program!
• eHiTS is an ab-initio method folding complexity
• Different types of protein mobility
• Movement of large domains
• Multiple conformations observed in a few residues
  (to be addressed in first step)

38
Binding Site Side-Chains

• Modeling side-chain flexibility of binding site
  residues in eHiTS.
• First the candidate chains should be selected.
• Solvent exposed?
• More statistics

39
Binding Site Side-Chains (cont'd)

• Same technique of fragmentation can be applied to
  side-chains.
• Rigid docking and pose matching with the backbone
  constraints.

40
The Problem Size (Difficulty)

• Run statistics for a set of 20 PDB codes (all
  numbers are averages)
• rigid fragments 3.05
• poses tried in RD 60 million
• poses accepted in RD 493,354
• Best Match RMSD 0.60 A
• Best Match Found 1.01A
• (NOTE Finding the best match is NP-hard for a
  general scoring function.)

41
Pose Match Example

• Closest match for an HIV protease inhibitor
  (1AAQ)

42
Training

• eHiTS uses a statistical scoring function.
• Training is done by known structures.
• Pose Match specific training is done by linear
  programming modeling and using CLP package.
• For receptor flexibility modeling we can either
• Generate receptor decoys
• Use PDB complexes with same receptor

43
Goals and Previous Works

• Induced fit modeling in Glide 5
• Docking into rigid receptor using softened
  scoring func.
• Receptor active site sampling
• Complex optimization (minor backbone flexibility)
• Differences with our approach
• Simultaneous handling of ligand/receptor
  flexibility
• Same scoring function (no softened version)
• The set of cross-docking data can be used for
  training and benchmarking.

44
Selected References

1. S. J. Teague, Implications of Protein Flexibility
   for Drug Discovery, Nature Reviews (Drug
   Discovery), vol. 2, pp. 527-541, 2003.
2. D. B. Kitchen, H. Decornez, J. R. Furr, J.
   Bajorath, Docking and Scoring in Virtual
   Screening for Drug Discovery Methods and
   Applications, Nature Reviews (Drug Discovery),
   vol. 3, pp. 935-949, 2004.
3. H. J. Dyson, P. E. Wright, Intrinsically
   Unstructured Proteins and Their Function, Nature
   Reviews (Molecular Cell Biology), vol. 6, pp.
   197-208, 2005.
4. Z. Zsoldos, D. Reid, A. Simon, B. S. Sadjad, A.
   P. Johnson, eHiTS A New Fast, Exhaustive
   Flexible Ligand Docking System, J. of Mol.
   Graphics and Modeling, (to appear available
   online).
5. W. Sherman, T. Day, M. P. Jacobson, R. A.
   Friesner, R. Farid, Novel Procedure for Modeling
   Ligand/Receptor Induced Fit Effects, J. Med.
   Chem., vol. 49, pp. 534-553, 2006.

45
Questions?