Omega Redux

1
Omega Redux

• Matthew T. Stahl

2
Why Omega2?

• First accuracy, then speed
• Tight integration of new toolkits
• Search routines based on MMFF and derivatives
• Studies other than bioactive conformer generation
• Model builder and ring conformations

3
Why not Omicron?

• Stochastic procedure makes reproducible answers
  challenging
• Full structure MMFF optimization introduces
  asymmetry (omicron.sdf)
• Small fragments optimize to symmetric structures
• Ring conformations must be produced off-line for
  speed
• Why not generate all fragments off-line and
  assemble them to build models?

4
Omega2 Builder

• Fragment molecule by cleaving ?-bonds
• Exocyclic attachments
• Carbon-heteroatom bonds
• Preserve conjugation
• Generate and store fragment conformers using
  makefraglib (fraglib.sdf)
• Generate models by assembling on-disk fragments
• Enumerate ring conformers using templates
• Enumerate invertible nitrogens

5
Makefraglib Statistics

• 2.4M vendor compounds
• 129 Mb fragment library
• 81,392 fragments
• 40 CPUs 24 hrs

6
Omega2 Builder

• Fragment molecule by cleaving ?-bonds
• Exocyclic attachments
• Carbon-heteroatom bonds
• Preserve conjugation
• Generate and store fragment conformers
  (makefraglib)
• Generate models by assembling on-disk fragments
• Enumerate ring conformers using templates
• Enumerate invertible nitrogens

7
Generating Ring Conformers

• Exocyclic double bonds on aromatic systems (i.e.
  pyridone)
• Aliphatic nitrogen geometry highly dependent on
  attached aromatic
• Ring conformations may be dependent on exocyclic
  attachments (1qbt)
• Sampling, sampling, sampling

8
Invertible Nitrogens

• Geometry must be pyramidal
• No stereochemistry specification
• No more than 2 ring bonds
• Example 1i8z

9
User Defined Templates

• -fixfile option in Omega2
• Molecule fragment provided as file (.mol2, .sdf)
• Entire molecule is constructed around the used
  defined fragment
• Example 4phv

10
Reproducing Bioactive Conformations

• RMS distance is an imperfect measure
• Poor crystal structures muddy the waters
• Goal RMS is a moving target
• 4phv 1.3
• 1qbu 1.6
• Distance to reproduce binding pose lt1.5Å

11
Reproducing Bioactive Conformations

• All tests from connection table (SMILES)
• RMS distance used to measure goodness of fit
• 3 Data Sets
• 36 structures, Boström, J., Greenwood J.R., and
  Gottfries, J., J. Mol. Graph. Mod., 2003, 21,
  449-462
• 100 structures, Perola, E. and Charifson, P.S.,
  J. Med. Chem., 2004, 47, 2499-2510
• 263/300 structures http//www.ccdc.cam.ac.uk/produ
  cts/life_sciences/validate/astex/

12
Omega2 Builder vs X-ray
Gold2
13
Version Comparison
Boström
Perola
Gold2
14
Comparison of Maxtime
15
Force Field Selection

• mmff
• mmff_NoEstat
• mmff_Trunc
• mmff_Sheff
• mmff94s
• mmff94s_NoEstat
• mmff94s_Trunc
• mmff94s_Sheff

16
Force Field Comparison
Perola
17
Force Field Comparison
Perola
18
Force Field Comparison
Perola
19
Force Field Comparison
Perola
20
Effect of Optimization
Perola
1 hour cpu time
21
Effect of Optimization
Perola
40 min cpu time
22
Dont Change Versions Unless

• Fixfile active site conformer generation
• Require custom templates
• Solution phase conformer generation
• Working with very challenging structures
• You want to be on the bleeding edge

23
Conclusions

• Omega2 sucks less
• Improved ring handling
• Improved model builder
• More accurate than Omega 1.8
• Omega2 isnt just for bioactive conformer
  generation anymore
• Omega2 is a new starting point

24
Acknowledgements

• Anthony Nicholls
• Geoff Skillman
• Stan Wlodek