Scoring Functions

1
Scoring Functions

• What works and what doesnt

Mark McGann OpenEye Scientific Software www.eyesop
en.com
2
What is a scoring function?

• Approximate binding energy
• Easy to evaluate (less than 1ms/pose)
• Rule based, not physics based

3
The Scoring Functions

• Gaussian Scoring Function1
• ChemGauss2
• PLP3
• Chemscore4
• Shape Surface2
• Electrostatic Surface2

• Mark R. McGann, Harold R. Almond, Anthony
  Nicholls, J. Andrew Grant, and Frank K. Brown,
  Gaussian Docking Functions'", Biopolymers, Vol.
  68, pp. 76--90, 2003.
• No publication at present.
• Gennady M. Verkivker, Djamal Bouzida, Daniel K.
  Gehlaar, Paul A. Rejto, Sandra Arthurs, Anthony
  B. Colson, Stephan T. Freer, Veda Larson, Brock
  A. Luty, Tami Marrone Peter W. Rose,
  Deciphering common failures in molecular docking
  of ligand-protein complexes, Journal of
  Computer-Aided Molecular Design, Vol. 14,
  pp.731-751, 2000.
• Matthew D. Eldridge, Christopher W. Murray,
  Timothy R. Auton, Gaia V. Paolini and Roger P.
  Mee, Empirical scoring functions I. The
  development of a fast empirical scoring function
  to estimate the binding affinity of ligands in
  receptor complexes, Journal of Computer-Aided
  Molecular Design, Vol. 11, pp. 425-445, 1997.

4
Test Case The PDB

• Pros
• Readily available
• Many structures, 24244 total
• Cons
• Not all structures are complexes
• Not all complexes are with high affinity binders
• PDB files not exactly high quality

5
PDB Complexes
5462 out of 24244 (23) of entries in the PDB
determined to be complexes

• Ligand candidate rules
• Not covalently bound to protein
• Between 10 and 50 heavy atoms
• Only H,C,N,O,F,P,S,Cl,Br or I atoms
• At least 1 Carbon atom
• Ligand Selection Rules (when there are 2 or more
  candidates)
• Favor ligands with closer to 30 heavy atoms
• Favor ligands that are more deeply buried
• Discard rules
• Abnormal valences on ligand (e.g., carbons with
  more than 4 bonds)
• Abnormal bond lengths on ligand (any bond greater
  than 2.75Å)
• Ligand protein heavy atom contacts less than 1.5Å
  (center-center)
• Ligands makes few close contacts with the protein
• Ligands with more than 12 rotatable bonds

6
Repeated Ligands
7
Testing Method

• Omega
• Conformer generation (Omega)
• Fred
• Pose ensemble generation
• Filter poses
• Score poses

8
Conformer Generation Omega

• Start with connection table (SMILES string)
• Generate initial structure using distance
  geometry
• Refine structure with MMFF
• Enumerate conformers using torsion and ring
  library

9
Conformer Generation Results (Best conformer
overlay with binding mode)
10
Pose Generation Ensemble Generation
Rotations
Translations
Systematic rotation and translation of conformers
within the active site 1 Angstrom step
size Conformers are rigid
11
Visualization of systematic search
Rotations
Translations
12
Negative Image Filter
13
Pose Generation Negative Image Filter
Negative image

Median Poses
Average Poses
50,000
170,000
14
Pose Ensemble Results(best pose after negative
image)
15
The Gaussian Scoring Function

• Shape based (no chemical awareness)
• Smooth pair-wise Vdw like potentials
• Potential exist between heavy atoms

16
Shape Surface

• Shape based surface matching (no chemical
  awareness)
• Smooth gaussian potentials
• Potentials exist between points on a molecular
  surface

17
PLP

• Chemically aware of
• Hydrogen bond acceptors
• Hydrogen bond donors
• Non-polar atoms
• Sulphurs
• Metals
• Interaction potentials
• Pair-wise linear potentials
• HB and metal interactions are partly angle based

18
Electrostatic Surface

• Shape based surface matching with electrostatic
  coloring
• Smooth gaussian potentials
• Favorable between positive-negative and neutral
  surface points
• Negative potential between point without
  favorable interactions

19
Chemical Gaussian Scoring Function

• Chemically aware of
• Hydrogen bond acceptors
• Lone-pair positions
• Hydrogen bond donors
• Shape
• Aromatic Rings
• Metals
• Interaction potentials
• Smooth gaussian potentials
• Shape interaction is the same as the gaussian
  scoring function

20
Chemscore

• Chemically aware of
• Hydrogen bond acceptors
• Hydrogen bond donors
• Lipophilic atoms
• Frozen rotatable bonds
• Metals
• Interaction potentials
• Most potentials are simple linear potentials
• 4 body hydrogen bonds potentials

21
Potential Outcomes of docking
Soft Docking Failure
Hard Docking Failure
Success!
Energy
Correct Structure
Configuration of ligand in active site
22
1 Rank Pose is RMSD lt 2.0
23
1 Rank Pose is RMSD lt 1.5
24
M.A.S.C.

• Pre-dock ligand into multiple reference active
  sites
• Calculate score average and standard deviation in
  the reference sites
• Correct score in actual docking runs with the
  following formula

• Reduces systematic biases in scoring functions
• Pre-dock calculations are expensive but only need
  to be done once for a given database
• Can be used with any scoring function

25
MASC vs. Non-MASC Enrichment
26
Acknowledgements

• Anthony Nicholls Zap Bind
• Roger Sayle OEChem
• Matt Stahl OEChem, MMFF, Omega
• Geoff Skillman OEChem, Omega
• Bob Tolbert OEChem, Zap Bind
• Stanislaw Wlodek MMFF