1. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Conformational sampling in protein-ligand complex environment Zsolt Zsoldos SimBioSys Inc., © 2004 Contents: ● Introduction: conformation sampling requirements of scoring ● Statistics on bound ligand conformations ● Search space for exhaustive flexible docking, de novo design ● Comparative evaluation of conformation sampling algorithms http://www.simbiosys.ca/

2. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Acknowledgement ● A. Peter Johnson ● Val Gillet ● ICAMS, University of Leeds

3. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Contents: Introduction ● Introduction ● Pose sampling requirements for interactions ● Orientation and dihedral precision needed ● Statistics on bound ligand conformations ● Search space size for flexible ligand docking ● Comparative evaluation of conformation sampling algorithms for use in flexible docking, de novo design

4. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Pose sampling requirements of good interaction geometry for scoring ❑ H-bond geometry H-acceptor distance range 1.6Å to 2.2Å, i.e. 1.9ű0.3Å ❑ hydrophobic contact carbon-carbon distance range 3.2Å to 4.2Å, i.e. 3.7ű0.5Å ❑ discretization must be fine enough to differentiate any atom movement >=0.5Å D H A LpLp d HA CC vdw d CC

5. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Orientation and dihedral angle precision required ❑ Goal: differentiate any atom movement >=0.5Å ❑ Drug-like molecules can have heavy atoms at 7Å distance from a rotation axis (see figure) ❑ Simple trigonometric calculation: tangential movement of 0.5Å is caused by rotation of about 5° at a rotation radius of 7Å ❑ Consequence: orientation and dihedral sampling must be finer than 5° ~7Å PDB code: 1CX2

6. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Contents: Statistics ● Introduction ● Statistics on bound ligand conformations ● Dihedral angle distribution statistics ● Frequency of low energy conformations ● Statistics and examples of high energy dihedrals ● Search space for exhaustive flexible docking ● Comparative evaluation of conformation sampling algorithms for use in flexible docking, de novo design

7. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Receptor-ligand complexes from RCSB PDB High resolution: less than 2.5Å 5000 entries Ligand dihedrals of rotatable bonds considered only Dihedral angle distribution statistics on bound ligand conformations

8. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Frequency of low energy conformations in 5000 PDB ligands ❑ All dihedrals of the ligand at energy minimum: ❑ 86 (1.7%) with ±5° error ❑ 131 (2.6%) with ±10° error ❑ 186 (3.7%) with ±15° error ❑ All low energy dihedrals: ❑ 108 (2.2%) with ±5° error ❑ 211 (4.2%) with ±10° error ❑ 315 (6.3%) with ±15° error PDB code: 1hak

9. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Statistics on high energy dihedrals in 5000 bound ligand conformations ❑ High energy, strained torsion occurance frequency: ❑ 1836 (36.9%) with ±5° ❑ 2721 (54.7%) with ±10° ❑ 3192 (64.2%) with ±15° ❑ Worst possible dihedral: ❑ 240 (4.8%) with ±5° ❑ 288 (5.8%) with ±10° ❑ 339 (6.8%) with ±15° ❑ 486 (9.8%) with ±30° PDB code: 1cle

10. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Scatter plot of consecutive dihedrals in bound ligand conformations Dihedral sequences of three consecutive rotatable bonds e.g. in -CH 2 -CH 2 -CH 2 -CH 2 - The angles are fully scattered and fill the whole range

11. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Examples of high energy dihedrals... Sampling at every 60° for each rotatable bond: Miss 97% of X-ray ligand conformations by more than 5° error PDB code: 1c2t

12. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Examples of high energy dihedrals... PDB code: 1uvs

13. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Examples of high energy dihedrals... PDB code: 1yds

14. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Examples of high energy dihedrals... PDB code: 1kmv

15. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Search space size for exhaustive flexible ligand docking ❑ Number of poses to examine with sampling defined: Translations(0.5Å) * Rotations(5°) * Dihedrals(5°) = 20 3 * 72 3 * 72 n ~ n=6 rot.bonds => 2*10 20 poses per ligand ❑ Brute force evaluation 2000/s => 3 billion years ❑ Stochastic methods explore a tiny fraction of the space with no guarantee to find the best or even any good-enough solutions

16. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Contents: Algorithm comparison ● Introduction ● Statistics on bound ligand conformations ● Search space for exhaustive flexible docking ● Comparative evaluation of conformation sampling algorithms for docking, de novo design ● Conformational sampling test ● Accuracy comparison (RMSD from target) ● Speed comparison

17. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Conformational sampling test ● The test problem: ● A target hydrocarbon chain conformation is generated randomly – number of carbons is given ● The algorithms have to find the target conformation based on scoring function value – which is simply the RMSD ● Algorithms tested: ● Taboo search ● Genetic Algorithm ● Incremental construction with continuous tweak

18. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Accuracy Comparison of the algorithms: best RMSD found Number of atoms in chain range: 5-20 Number of iterations: 10,000 Population, taboo list size: 100 10 test runs each

19. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Speed comparison of the algorithms Hardware: 2.6GHz Pentium 4 laptop Runtime displayed: CPU seconds

20. SimBioSys Inc.© 2004http://www.simbiosys.ca/ Summary ❑ Receptor bound conformations of flexible ligands deviate widely from low energy conformations ❑ Strained dihedrals occur with high frequency (37%) ❑ Interaction scoring sets low error limit (5°) ❑ Sampling of few hundred low energy conformers miss 97% of ligand conformations in X-ray data ❑ Sufficient sampling implies huge search space ❑ Stochastic methods are slow and inaccurate