Receptor-based Virtual Screening of Very Large Chemical Datasets Bohdan Waszkowycz, Tim Perkins, Carol Baxter, Jin Li and John Liebeschuetz Protherics Molecular Design Ltd Macclesfield, UK
Virtual Screening Computational compound selection for assay/synthesis Benefits focused subsets with enhanced hit rates analysis before assay established evaluate virtual combinatorial libraries before synthesis
Virtual Screening Methodologies 2-D similarity to known ligands low chance of novelty 3-D pharmacophore search imitates known binding mode limited steric constraints receptor-based docking 3-D structure allows open-ended query requires effective scoring procedures
PRO_LEADS Flexible-ligand docking software Empirical scoring function: ChemScore hydrogen bonding; lipophilic; flexibility calibrated against X-ray ligand protein complexes Direct estimate of free energy of binding Tabu searching algorithm
Docking Validation 86% correct over 70 protein ligand complexes Raloxifene in 1err r.m.s. error 0.77 Å docked solution green; X-ray orientation grey
Virtual Screening Experiment DockCrunch: dock 1 million ACD-SC compounds into estrogen receptor DockCrunch Objectives demonstrate technical practicability recover known active compounds identify novel ligands Collaboration with SGI
Estrogen Receptor Conformations Agonist (1ere) Antagonist (1err)
ACD-SC Preparation Strip counter-ions Add hydrogen atoms Check/fix valency problems Protonate as physiological ph Calculate 2-D properties Convert to 3-D with MSI Converter
2-D Property Selection Property Min Max Molecular weight 200 600 log P -8 7 Rotatable bonds 0 10 Chiral carbon atoms 0 3 Hydrogen-bond acceptors 0 8 Hydrogen-bond donors 0 8 Heavy atoms 11
DockCrunch Experiment 1.1 million compounds Spiked with known agonists/antagonists Dock against both receptor forms 64-processor Origin 2000 at SGI 30 s per ligand 6 days per million ligands
Results and Analysis: Aims Navigate and visualize 1.1M dockings Identify known ligands Select subset for bioassay objective selection criteria docked energy plus complementarity identify novel ligands
Docked-Energy Profiles Agonist Receptor Antagonist receptor 25 30 20 ACD-SC Agonists 25 ACD-SC Antagonists Relative Frequency (%) 15 10 Relative Frequency (%) 20 15 10 5 5 0-60 -50-40 -30-20 -10 Docked Energy (kj/mol) 0-60 -50-40 -30-20 -10 Docked Energy (kj/mol)
Antagonist Enrichment Rates Energy Cutoff % total n Rate none 100.0 20 1.0 35 kj/mol 22.0 20 4.5 40 kj/mol 6.7 20 15.0 45 kj/mol 1.2 19 78.0
Selectivity Docked Energy in Antagonist ER (kj/mol) ACD-SC Agonists Antagonists Docked Energy in Agonist ER (kj/mol)
Receptor-derived Descriptors for Post Processing Docked energy & component energies can over-predict random compounds Receptor-ligand complementarity quality of steric complementarity polar-lipophilic surface area mismatch 3-D similarity to known ligands in terms of ligand to receptor contacts
Complementarity Space Polar-lipophilic Clash Area (Ų) Low-energy ACD-SC Agonists Steric Penalty
PropertyViewer Interactive analysis Multipleproperty filters Coupled 3-D views Graphing
Agonist Selection Process Selection Criteria n None 1,152,379 Docked Energy 73,961 Energy components 12,265 Surface complementarity 2571 2-D properties (non-steroidal) 1520 3-D similarity 293
Sample Virtual Hits O O O N O H H H
Validated ER Hits from DockCrunch Approximately 100 hits were proposed for purchase. 37 virtual hits (non-steroidal) were actually purchased and assayed (Panlabs) 7 with Ki between 100 nm and 1 µm 12 with Ki between 10 and 100 nm 2 with Ki < 10 nm 10 drug-like, predicted poor binders were also purchased and assayed as negative control. No actives found at 10 µm.
Characteristics of ER Hits MACCS 2D similarity to the reference structures ER Hits Ki (nm) Oestradiol DiethylStil. Tamoxifen Raloxifen PMD4164 6.9 0.5 0.4 0.2 0.3 PMD4148 7.7 0.3 0.3 0.4 0.4 PMD4165 13 0.6 0.4 0.3 0.4 PMD4163 14 0.3 0.3 0.2 0.3 PMD4183 27 0.5 0.5 0.4 0.4 PMD4170 29 0.2 0.2 0.2 0.2 PMD4157 35 0.5 0.6 0.3 0.2 PMD4151 37 0.4 0.3 0.2 0.4 PMD4169 52 0.3 0.5 0.2 0.3 PMD4146 62 0.4 0.5 0.2 0.3 PMD4178 63 0.4 0.3 0.5 0.7 PMD4176 88 0.4 0.4 0.2 0.3
Summary of Other Virtual Screening Datasets of 10,000 screened against Thrombin and Xa. ~85% recovery of seeded known inhibitors, in top 1000 scoring compounds. 264 non-amidines selected for testing from Thrombin hit list, after post screen filtering. 8 hits with K i < 30 µm (3%). Several hits with good looking modes of binding and scope for rapid SAR investigation via combinatorial chemistry - PRO_SELECT.
Linux Implementation of PRO_LEADS Linux Based System open, stable widely supported and used 100 P3 750 MHz Processor Farm - Costeffective solution to large scale computing requirement Estimated time to run DockCrunch: 3 Days per receptor
Conclusions Docking very large databases both practicable and useful find known and novel hits Need good filtering protocols using energetic and structural criteria Fast and accurate receptor based virtual screening a reality http://www.protherics.com/crunch/ john.liebeschuetz@protherics.com
Acknowledgements Bohdan Waszkowycz Tim Perkins Carol Baxter Richard Sykes Chris Murray Jin Li SGI MSI