The Long and Rocky Road from a PDB File to a Protein Ligand Docking Score. Protein Structures: The Starting Point for New Drugs 2

Transcription

1 The Long and Rocky Road from a PDB File to a Protein Ligand Docking Score Protein Structures: The Starting Point for New Drugs 2 Matthias Rarey Stefan Bietz Nadine Schneider Sascha Urbaczek University of Hamburg Bundesstraße Hamburg, Germany info@zbh.uni hamburg.de Pteridine Reductase PDB: 2x9g Protein Structures: The Starting Point for New Drugs 3 Detailed Structural Analysis of PDBbind 2007 by Consulting Electron Density Maps complexes with structural deficiencies 1HK4 1GNI Missing electron density Alternative conformations High temperature factors 1AJP 1PXO Contact with symmetry related subunits 1

2 Protein Structures: The Starting Point for New Drugs 5 Protein Structures: The Starting Point for New Drugs 6 Does it Matter? Contribution of a Single Hydrogen Bond to Binding Affinity 7 Example: Thrombin/Trypsin inhibitor complexes Abb.: G. Klebe, Wirkstoffdesign Spektrum Akad. Verlag, 2009 Exchange NH CH 2 : slight loss in IC 50 2BRB G= 28 kj/mol G= 34 kj/mol Exchange NH O: loss of two hydrogen bonds upon dehydration no formation of a new hydrogen bond IC 50 drops by about three orders of magnitude * Foloppe N. et al., JMedChem

3 Urbaczek et al, JCIM (2013), 53, Urbaczek et al, JCIM (2013), 53, PHASE 1: Perception of Chemical Structure Valence States 9 10 Valence state: Chemically valid combination of bond orders and formal charge for a particular element. Aim: Find a valid valence state combination (VSC) Urbaczek et al, JCIM (2013), 53, Urbaczek et al, JCIM (2013), 53, Valence States From Valence States to Valence Bond Structures Atoms get sets of all valence states compatible to the detected bond structure Each valence state is scored by the geometry of surrounding covalent bonds Valence states get probabilities converted to scores Iterative refinement by removing valence states incompatibleto those of neighboring atoms Partitioning of the molecules into zones of covalently bound atoms with yet undefined valence state Branch&Bound procedure enumerating and selecting highest scoring valence states Consider aromaticity, favorize certain functional group topologies in case of nearly equal scores 3

4 Urbaczek et al, JCIM (2013), 53, Results: Quality of Chemical Structure Perception PHASE 2: Prediction of Protonation PDB files with hand curated structures, 563 ligands in total NAOMI vs. Ligand Expo: 92% identical structures differences found: (NAOMI structure validated by literature): Prediction of Hydrogen Coordinates Prediction of Hydrogen Coordinates Existing hydrogen placing tools: WHAT IF 1 Protonate3D (MOE) 2 HINT Comp. Titration 3 Protoss 4 YASARA 5 1 R.W.W. Hooft et al. Proteins, 26(4): , P. Labute et al. Proteins, 75(1):187, A.S. Bayden et al. Journal of Computer-Aided Molecular Design, 23(9): , T. Lippert et al. Journal of Cheminformatics, 1(1):13, E. Krieger et al. Computational Drug Discovery and Design, 819: ,

5 Error Case Examples 17 Frequency of Tautomerism 18 Ligand Expo 6 : 17,563 small molecules 7 According to the ProToss protonation model 81 % exhibit alternative tautomers or / and protonation states 17 % only contain commonly used groups (primary amines, carboxylats, imidazoles) portion portion of of molecules [%] [%] ensemble size k ensemble size 3k3i 1jj9 1y6t Only 19 % do not show variability 1802 structurally different variable mode regions (VMRs) three or more aromatic rings 21% aliphatic groups 19% classical VMRs 0.1% other rings 3% Repulsive donor interactions YASARA Hydrogen-metal clashes Protonate3D Unsaturated donors / acceptors Protoss v , isolated two aromatic rings aromatic rings 19% 38% 6 Z. Feng et al., 20(13):2153-5, Downloaded from ( 01/03/2014 ) different VMR type portions Challenges to an Advancement of ProToss 19 Protoss Workflow 20 Find the tautomer / protonation state that leads to the best hydrogen bonding network in the protein-ligand complex Keep the runtime low variable groups hydrogen-bonding network optimal network 5

6 Urbaczek et al, J Chem Inf Model (2014), 45 (3): Generation of Tautomers and Protonation States Generation of Tautomers and Protonation States Step 1: Conjugated zone identification Pemetrexed (PTR1 inhibitor) Urbaczek et al, J Chem Inf Model (2014), 45 (3): Urbaczek et al, J Chem Inf Model (2014), 45 (3): Generation of Tautomers and Protonation States Generation of Tautomers and Protonation States Step 2: Atom state selection Step 3: Enumeration of tautomers and protonation states tautomer generation 6

7 Urbaczek et al, J Chem Inf Model (2014), 45 (3): Generation of Tautomers and Protonation States ProToss Workflow Extended Step 4: State scoring Step 5: Inclusion into ProToss optimization procedure Results: Comparison on the sc PDB Undesirable Donor Contacts in the Protein Ligand Interface Dataset: sc-pdb protein-ligand complexes (37 with errors excluded) Druggable binding-sites Mainly drug-like ligands less contacts 8 Paul et al., Proteins: Structure Function and, 54: , ftp://cheminfo.u-strasbg.fr/databases/scpdb/ (01/07/2013) 7

8 Undesirable Donor Contacts in the Protein Ligand Interface Undesirable Donor Metal Contacts in the Protein Ligand Interface less contacts less contacts Undesirable Donor Metal Contacts in the Protein Ligand Interface Undesirable Contacts in the Protein Ligand Interface less contacts 8

9 Astex Diverse Set 9 : Protonation Comparison Astex Diverse Set: Protonation Comparison Hartshorn et al., J.Med.Chem, 50(4): , 2007 Astex Diverse Set: Protonation Comparison Astex Diverse Set: Protonation Comparison

10 Astex Diverse Set: Protonation Comparison Astex Diverse Set: Protonation Comparison Å 1.34 Å Astex Diverse Set: Protonation Comparison ProToss Runtime Evaluation Whole complex optimization on the Astex Diverse Set (including all ligands, co-factors and water molecules) Runtimes include File IO Complex preprocessing Network optimization Intel Core i CPU ( 3.4 GHz ) and 8 GB memory Mean: 2.47 s Median 0.93 s 10

11 PHASE 3: Scoring Protein Ligand Complexes 41 Schneider et al, J. Comput. Aided Mol. Des., 27(1), (2013) HYDE Towards a Consistent Description of Hydrogen Bonding, Dehydration and the Hydrophobic Effect 42 G HYDE atoms i i G dehydration i G H-bond i G dehydration 2.3RT plogp i acc i i G H-bond 2.3RT plogp i sat i F sat unfavorable energy favorable energy 1GKC Schneider et al, J.Comput. Aided Mol. Des : HYDE Benchmarks Estimating the Cost of a Single Hydrogen Bond PDBbind 2007 coreset [1] Binding affinity prediction Astex diverse set [2] Redocking experiments Dataset of Useful Decoy [3] Virtual screening study [1] Wang R. et al, JMedChem, 2004, 2005 [2] Hartshorn M. J. et al, JMedChem, 2007 [3] Huang N. et al, JMedChem, 2006 [4] Schneider N. et al, JCAMD, sahh 46 % rxr 77 pr % ppar Mean 88 % AUC % parp STD 0.17 p38 na mr hsp90 Median AUC 0.78 hmga hivrt 75 % hivpr Min 67 % AUC 0.50 gr gpb fxa er_antagonist Max AUC 0.95 er_agonist cox1 46 % ada pde5 alr2 ache ace trypsin 16 % src pnp gart 0.5 Å dhfr cox2 1 Å 1.5 Å 2 Å cdk2 ampc RMSD vegfr2 thrombin pdgfrb inha fgfr1 egfr comt Rank <= 32 Rank <= 20 Rank <= 5 Rank <= 4 Rank <= 3 Rank <= 2 Rank 1 2BRB G HYDE = 32 kj/mol G= 28 kj/mol G= 34 kj/mol G HYDE = 22 kj/mol * Foloppe N. et al., JMedChem

12 A Few Things to Remember Contributions / Acknowledgements PDB files are models consult electron density. Perception of chemical structures from PDB is difficult cross check with literature. Assigning protonation and tautomeric states is not simpler cross check theresultsofautomatic assignment procedures Tools like ProToss help in large scale applications in which manual curation is not possible. Stefan Bietz ProToss development Sascha Urbaczek Tautomer generation Tobias Lippert First version of ProToss Nadine Schneider Hyde development Benjamin Schulz ProToss development, testing and validation Holger Claussen Testing, testing Christian Lemmen AMD Group Software Availability 47 More on ProToss ProToss in action: More on tools and servers from ZBH: