Computer Science meets Biology: Guiding “in vitro” experiments with “in silico” predictions.

1. Computer Science meets Biology:Guiding “in vitro” experiments with “in silico” predictions. Gidalevitz T, Biswas C, Ding H, Schneidman D, Wolfson HJ, Stevens F, Radford S, Argon Y.

2. Agenda • Introduction to the docking problem • The PatchDock algorithm • Biological problem • Real experimental results

3. T Complex Receptor Ligand What is Docking? • Given two molecules find their correct association: = +

4. Problem Importance • Computer aided drug design – a new drug should fit the active site of a specific receptor. • Understanding of biochemical pathways - many reactions in the cell occur through interactions between the molecules. • Despite the advances in the Structural Genomics initiative, there are no efficient techniques for crystallizing large complexes and finding their structure.

5. Docking Algorithm Bound Docking • In the bound docking we are given a complex of 2 molecules. • After artificial separation the goal is to reconstruct the native complex. • No conformational changes are involved. • Used as a first test of the validity of the algorithm.

6. Unbound Docking • In the unbound docking we are given 2 molecules in their native conformation. • The goal is to find the correct association. • Problems: conformational changes (side-chain and backbone movements), experimental errors in the structures. + = ?

7. Brute force enumeration of the transformation space: FFT – Katchalski-Katzir et al. (1992) (Walls & Sternberg, Vakser, Gabb et al., Camacho et al., Chen & Weng) Soft Docking – Jiang & Kim (1991), Palma et al., Randomized algorithms: GA, Monte-Carlo - Jones et al., Gardiner et al. Local shape featurematching: Dock- Kuntz (1982) ‘knobs’ and ‘holes’ – Connolly (1986) Geometric Hashing - Norel et al., Fischer et al. (1994) Flexible docking - Sandak et al. FlexX: hydrogen H-bonding – Rarey et al. Docking Algorithms

8. PatchDock … • is an efficient method for unbound docking of rigid molecules. • The molecular shape is used explicitly avoiding the exhaustive search of the 6D transformation space. • The algorithm focuses on local surface patches divided into three shape types: concave, convex and flat. • The geometric surface complementarity scoring is extremely fast and accurate. It employs advanced data structures for molecular representation: Distance Transform Grid and Multi-resolution Surface. http://bioinfo3d.cs.tau.ac.il/PatchDock Duhovny, D., Nussinov, N Wolfson, H.J. Lecture Notes in Computer Science 2452, pp. 185-200, Springer Verlag, 2002

9. PatchDock Method PDB files Surface Representation Patch Detection Matching Patches Scoring & Filtering Candidate complexes

10. Dense MS surface (Connolly) Sparse surface (Shuo Lin et al.) Surface Representation

11. Patch Detection • Shape representation by patches. PatchDock applies a segmentation algorithm to divide the surface into shape- based patches. • Connolly surface representation • Sparse surface [2]: local minima and maxima of Connolly surface. The surface topology graph is obtained by connecting neighboring points. • PatchDock focuses on sparse surface features, preserving the quality of shape representation. • The sparse features reduce the complexity of the matching step.

12. Receptor patches Ligand patches Matching Patches Matching 2 points and their associated normals is sufficient to compute transformation in 3Dspace. Transformation Base:1 critical point with its normal from one patch and 1 critical point with its normal from a neighbor patch. Base signature: distances and angles. Match every base from the receptor patches against all the bases from complementary ligand patches with similar signatures. Geometric Hashing of base signatures is used to speed up the search. dE, dG, α, β, ω

13. -1 0 +1 Penetrations Filtering Distance Transform Grid stores the distances from the surface of the molecule. The distance is negative inside the molecule and positive outside. Steric clashes are checked by accessing the receptor grid with ligand surface points.

14. Scoring The surface of the receptor is divided into five shells according to the distance function: S1-S5 The number of ligandsurface points in every shell is counted. The geometric score is a weighted sum of the number of ligand surface pointsinside every shell. Multi-resolution surface data structure was developed to speed up this stage.

15. Dataset and Results Protein-Proteincases from protein-protein docking benchmark [6]: Enzyme-inhibitor – 22 cases Antibody-antigen – 16 cases Protein-DNAdocking: 2 unbound-bound cases Protein-drugdocking: tens of bound cases (Estrogen receptor, HIV protease, COX) Performance: Several minutes for large protein molecules and seconds for small drug molecules on standard PC computer. Estrogen receptor Estradiol molecule from complex docking solution DNA endonuclease Estrogen receptor with estradiol (1A52). RMSD 0.9Å, rank 1, running time: 11 seconds docking solution Endonuclease I-PpoI (1EVX) with DNA (1A73). RMSD 0.87Å, rank 2

16. Results Enzyme-Inhibitor docking 1 Number of highly penetrating residues in unbound structures superimposed to complex

17. Results Antibody-Antigen docking 1 Number of highly penetrating residues in unbound structures superimposed to complex

18. The Real Challenge:Can we help biologists? + = ?

19. Identification of the N-terminal peptide binding site of GRP94 • GRP94 - Glucose regulated protein 94 • VSV8 peptide - derived from vesicular stomatitis virus Gidalevitz T, Biswas C, Ding H, Schneidman-Duhovny D, Wolfson HJ, Stevens F, Radford S, Argon Y. J Biol Chem. 2004

20. Biological motivation • The complex between the two molecules highly stimulates the response of the T-cells of the immune system. • The grp94 protein alone does not have this property. The activity that stimulates the immune response is due to the ability of grp94 to bind different peptides. • Characterization of peptide binding site is highly important.

21. GRP94 molecule • There was no structure of grp94 protein. Homology modeling was used to predict a structure using another protein with 52% identity. • Recently the structure of grp94 was published. The RMSD between the crystal structure and the model is 1.3A.

22. Docking • PatchDock was applied to dock the two molecules, without any binding site constraints. • Docking results were clustered in the two cavities:

23. GRP94 molecule • There is a binding site for inhibitors between the helices. • There is another cavity produced by beta sheet on the opposite side.

24. Experimental Verification Goals: • Try to eliminate one of the cavities. • Find the positions of the amino acids which are important for peptide binding.

25. Experimental Verification 1 • Experimental data shows that inhibitor and peptide can bind simultaneously. • Two residues in the inhibitor binding site were mutated. • The mutant did not bind inhibitor, however it could still bind peptide. • The binding sites of the inhibitor and peptide are distinct. • The abolition of the inhibitor does not affect peptide binding.

26. Experimental Verification 2 • The peptide binding was pH sensitive. Therefore involvement of His residue was suspected. • His125 was mutated to Asp and Tyr. The first mutated protein did not bind the peptide at all and the second had only partial activity. • Both mutants were soluble and could bind the inhibitor.

27. Computational Verification 2

28. Conclusions • Computational prediction can help in guiding “in vitro” experiments. • Further algorithmic improvements will yield in more reliable predictions.