HYDEing the False Positives

Transcription

1 HYDEing the False Positives Matthias Rarey Center for Hamburg Gudrun Lange

2 HYDE = HYdrogen bonding and DEsolvation Center for 2 HYDE Thus, describes a primary consistently question in hydrogen molecular bonds, design the should hydrophobic be which donors effect and desolvation acceptors need to be satisfied and not how more hydrogen bonds can be formed. (Martin Stahl, A medicinal chemist s guide to molecular interactions, JMedChem, 2010) PDB: 1GKC

3 HYDE = HYdrogen bonding and DEsolvation Center for 3 HYDE Thus, describes a primary consistently question in hydrogen molecular bonds, design the should hydrophobic be which donors effect and desolvation acceptors need to be satisfied and not how more hydrogen bonds can be formed. (Martin Stahl, A medicinal chemist s guide to molecular interactions, JMedChem, 2010) PDB: 1GKC

4 HYDE Scoring Function * Concept Center for *Reulecke et al., ChemMedChem, G interaction ΔG ΔG ΔG binding atom i i desolvation G desolvation i interaction

5 HYDE Atom based Desolvation 5 ΔG 2.3RT acc acc plogp i i i desolvation free bound i plogp i Atom solvation parameter derived from octanol/water partition coefficients acc i Solvent accessibility of atom i with respect to the chemical model i i

6 plogp i Atom based Hydrophobicity * Hansch C. et al., American Chemical Society, Calibration dataset: 458 small, simple molecules taken from the Starlist * 21 plogp atom types used plogp(xygen sp2) = 1.42 G desolvation = 8.09 kj/mol plogp(aromatic Carbon) = 0.45 G desolvation = 1.37 kj/mol

7 HYDE Atom based Interaction 7 2.3RT ΔG sat sat plogp i i i interaction bound free f sat i i sat i Number of intermolecular and intramolecular interactions (degree of saturation) i f sat Saturation factor describing the incomplete saturation of the hydrogen bond network in bulk water 310K f sat = K f sat = 1

8 HYDE verview 8 Experimental logps of small molecules plogp increments f sat HYDE Correct prediction of: hydrophobic effect (~ 110J/A 2 ) H bonds in vacuum (~ 16 kj/mol) H bonds in water ( 2 to 6 kj/mol) affinity loss due to unsatisfied H bond function (~ 6 kj/mol) Score reflects affinity of inhibitor to protein G = RTlnK

9 Binding Mode Analysis with HYDE Center for * Foloppe N. et al., JMedChem HYDE color code: + G contribution G contribution no G contribution receptor amide oxygen ligand aromatic oxygen desolvation cost 8.2 kj/mol 2.4 kj/mol kj/mol receptor aromatic carbons 1.7 kj/mol N C H N7 8 C6 ligand aromatic carbon hydrophobic effect receptor amide nitrogen 1.9 kj/mol 3.6 kj/mol 6.3 kj/mol 7.4 kj/mol G experimental 28 kj/mol G HYDE 22 kj/mol ligand aromatic nitrogen 6.4 kj/mol 7.5 kj/mol 2BRB * hydrogen bond energy 2.2 kj/mol

10 Ideal Structures vs Real Life Center for *Lippert et al., J.Cheminformatics, Avoid clashes and optimize the hydrogen bond network Stage 2: 1: Crystal structure hydrogen avoid clashes, bond optimize network hydrogen bond optimization geometries * 1X8X G opt = LJpot inter + LJpot intraligand + Torsion Ligand + 2 G HYDEinteract + G HYDEdesolv G HYDE 7 29 kj/mol

11 Ideal Structures vs Real Life Center for *Lippert et al., J.Cheminformatics, Avoid clashes and optimize the hydrogen bond network Stage 2: 1: Crystal structure hydrogen avoid clashes, bond optimize network hydrogen bond optimization geometries * 1X8X G opt = LJpot inter + LJpot intraligand + Torsion Ligand + 2 G HYDEinteract + G HYDEdesolv G HYDE kj/mol

12 Redocking Workflow 12 Astex diverse set Docking 400 FlexX poses generated

13 Redocking Results verview 13 supplied PDB 41 % 75 % 87 % 92 % Hydrogen bond network optimized PDB 41 % 79 % 88 % 92 % RMSD [Å] RMSD [Å] Min RMSD 0.18 Max RMSD 5.76 Mean 0.79 STD 0.71 Median 0.56

14 Astex Diverse Set Redocking 14 RMSD best scoring pose of all pockets 76 % of the best scoring below 2 ÅRMSD

15 Astex Diverse Set Redocking 15 RMSD best scoring pose of all best pockets % of the best scoring below 2 ÅRMSD

16 Astex Diverse Set Redocking 16 RMSD best scoring pose of all best pockets 1meh 1n2v 1oq5 1sqn 1tt1 1u4d 86 % of the best scoring below 2 ÅRMSD No good pose generated Problematic complex geometries (soaked structures)

17 Unexpected Inhibition 17 1oq5 G experimental 49 kj/mol crystal ligand (grey) and poses (hyde coloring) Best RMSD Best scoring RMSD 0.9 Å G HYDE 20 kj/mol RMSD 5.1 Å G HYDE 41 kj/mol

18 Soaked Structure 19 1meh Best scoring Best RMSD crystal structure G HYDE 15 kj/mol RMSD 5.3 Å G HYDE 36 kj/mol RMSD 1.1 Å G HYDE 33 kj/mol

19 Poor Density 1u4d 21 RMSD 0.4 Å G HYDE 18 kj/mol High temperature factor Temperature factor High Best RMSD RMSD 5.8 Å G HYDE 26 kj/mol Low Best scoring

20 Small Hydrophilic Binder 22 1tt1 G experimental 43kJ/mol 8 perfect h bonds Chain A Best scoring RMSD 0.7 Å G HYDE 46 kj/mol Kainate Chain B Best scoring RMSD 0.4 Å G HYDE 44 kj/mol

21 Hydrophobic Binder 23 1sqn G experimental 54kJ/mol Chain A Best scoring RMSD 0.6 Å G HYDE 49 kj/mol Chain B Best scoring RMSD 0.6 Å G HYDE 54 kj/mol

22 Two Possible Binding Modes 24 1n2v G experimental 23kJ/mol Best scoring RMDS 0.4 Å G HYDE 22kJ/mol RMSD 2.2 Å G HYDE 6kJ/mol

23 Astex Diverse Set Redocking 25 RMSD best of 32 of all pockets % found below 2 Årmsdconsideringthe32 best scoring

24 Astex Diverse Set Redocking 26 RMSD best of 32 of all best pockets % found below 2 Årmsdconsideringthe32 best scoring 85 % found below 1 Å rmsd considering the 32 best scoring No good/better pose generated

25 Screening Workflow 27 Dataset of useful Decoys (DUD) Docking 40 FlexX poses generated

26 Virtual Screening Performance 28 Mean AUC 0.72 STD 0.16 Median AUC 0.70 Min AUC 0.25 Max AUC 0.94 Structure quality: Q 1 (no structural def.) Q 2 (high Temp. factor) Q 3 (crystal packing) Q 4 (Model, poor density) Mean AUC 0.78 STD 0.15 Median AUC 0.75 Min AUC 0.53 Max AUC 0.94

27 True Positive Rates 29 p38 ER agonist HMGA NA TK CMT

28 NULL Hypothesis 30 Mean AUC 0,14 targetauc non targetauc STD 0,18 Median AUC 0,13 Min AUC 0,19 Max AUC 0,57

29 P38 MAP Kinase Different binding sites 31 Different binding modes of the ligands: Conformational change in 1kv2 (DUD target) 1kv2 1a9u Met109A R N H R Val38A Leu74A Phe169A Ala51A Leu167A NH + Glu71A Leu108A N N HN N H Asp168A R H N Asp168A R 1a9u 1kv2 Leu75A Leu104A Ile84A Lys53A Thr106A Glu71A Clash

30 P38 MAP Kinase Different binding sites 32 Different binding modes of the ligands: Conformational change in 1kv2 (DUD target) 1kv2 1a9u 1a9u Met109A R N H Leu167A Clash R Val38A Leu74A Phe169A 1a9u Ala51A NH + Glu71A Leu108A N N HN N H Met109A R N H Leu75A Leu104A Ile84A Lys53A Thr106A Leu167A Asp168A 1kv2 R Val38A Phe169A Ala51A Leu74A NH + Glu71A Leu108A H N Glu71A N N 1kv2 HN R Asp168A Ile84A Lys53A N H Thr106A R Leu75A Leu104A Asp168A R H N Asp168A Glu71A R

31 Three Structural Classes in p38 Actives Pyridinyl Imidazole (PI) 190 Triarylpyrrol (TA) 44 Diaryl urea derivatives (DU) only DU included as actives nly DU bind to the p38 MAP Kinase conformation in the DUD

32 Thymidine Kinase (TK) 34 1kim Thymidine G experimental 30 kj/mol G HYDE 31 kj/mol

33 TK Quiz Show: Find the Active! 35...decoy molecules that are physically similar but topologically distinct... Huang et al., JMedChem, 2006 Figure 1. Nucleosidesand nucleosideanaloguesthataresubstratesof TK2. Perez Perez et al. MedResearchRev 2008

34 Catechol Methyltransferase (CMT) 37 Small, shallow binding site, MG in octahedral coordination 1h1d Binding mode caused by crystal packing G experimental 47kJ/mol G HYDE 56kJ/mol G HYDE 42 kj/mol

35 Catechol Methyltransferase (CMT) 38 Same binding mode of the 2 molecules, chelating the MG ion G HYDE 46kJ/mol active Some of the decoys: G HYDE 54 kj/mol decoy

36 NULL Hypothesis: Factor Xa and Thrombin Center for * Wiley, JMedChem, FXA and Thrombin Similar structure Similar inhibitors 1ba8 1f0r Thrombin with FXA inhibitors Thrombin was used as surrogate protein for the design of FXA inhibitors *

37 Estrogen Receptor Agonist (ER) 40 3ert Best scoring G HYDE 57 kj/mol crystal ligand (grey) and best active

38 HMG CoA Reductase 41 1hw8 crystal ligand (grey) and best active Best scoring G HYDE 69 kj/mol

39 Neuraminidase 42 1a4g Best scoring G HYDE 51 kj/mol

40 Conclusions 43 HYDE... What did we learn: is an intuitive and unsupervised scoring function improves redocking results significant avoids false positives due to desolvation penalties facilitates the analysis of lead structures we need accurate data to get reliably good results with our tools having correct datasets is more important than having large datasets to understand the successes and failures of a method

41 Acknowledgement 44 Matthias Rarey and the whole AMD group Sally Hindle Holger Claußen Christian Lemmen Gudrun Lange Robert Klein Jürgen Albrecht Hans Briem Kristin Engels Thomas Zuhl Frank Sonnenburg Martina Brümmer Marcus Gastreich Carsten Detering Markus Lilienthal... the whole team

42 HYDE Analysis Tool: LeadIT GUI 45 ther interesting dates: HYDE: Scoring for lead optimization Date: Tuesday, March 29 th, 2011 Time: 3pm 3:30pm Location: Anaheim Convention Center, Room 212 B LeadIT Workshop: Date: Tuesday, March 29 th, 2011 Time: 4pm 6pm Location: Anaheim Convention Center, Room 211 Fragment based drug discovery in teams of medicinal and computational chemists Date: Thurday, March 31 th, 2011 Time: 12:10am 12:30am Location: Anaheim Convention Center, Ballroom B