Shape and ROCS

1
Shape and ROCS

• Bob Tolbert
• OpenEye Scientific Software, Inc.
• ACS - NYC
• September 10th, 2003

2
Agenda

• Shape Theory
• Basic ROCS usage
• Using VIDA to visualize results
• Hands-on examples
• Using the Color Force Field
• Hands-on examples with color

3
Gaussian Description of Shape

• Atoms are represented as Gaussians instead of
  hard spheres
• Easily integrable, analytic derivatives
• Product of 2 gaussians is another gaussian
• Easily calculate overlap between two atoms
• Easily calculate overlap between two collections
  of gaussians
• Shape Tanimoto and Tversky as measure of
  similarity

4
What is shape tanimoto?

• shape specific version of tanimoto that uses 3D
  overlap instead of bits for whats in common.
• Formula
• Sensitive to large size differences between two
  structures

5
What is shape Tversky?

• Tversky is an alternate similarity measure.
• Introduced for 2D similarity by John Bradshaw at
  Daylight
• Has weighting factor to deal with size
  differences.
• Useful for small structure vs large structure
  similarity

6
Tversky

• Tversky is assymmetric with ? 0.95 and ? 0.05
• ROCS reports 2 values Tversky-d and Tversky-q
  with the 0.95 weighting for query molecule and
  database molecule, respectively

7
What is ROCS

• Rapid Overlay of Chemical Structures
• Small molecule vs. small molecule
• Uses
• Scaffold jumping
• HTS Rescue
• Patent busting

8
ROCS Status

• Version 2.0 announced this week
• Completely rebuilt on top of OEChem
• Platforms include Linux, Linux-IA64, HPUX,
  MacOSX, AIX, Tru64(Alpha), Solaris, Win32 and
  Irix.
• PVM support for all platforms except Win32

9
ROCS in actual use

• How many 'hits' has ROCS helped to identify?
• 15-20, 4 classes, 15-20, 1, 1, 1, 50-75
• How many Targets attempted?
• 5, 4, 3, 1, numerous
• What are typical hit rates?
• 1-2, 6-15, 5-10, 2-5
• How many hits have made it into chemistry
• 0, 1-2, 3, 1, 1, 1
• How many HTS programs have been cancelled?
• gt1?

10
Minimum example

• ROCS requires only 2 inputs, a file containing
  shape query molecules and a file containing the
  database of interesting structures, i.e.
  corporate database, ACS-Screen, vendor databases
  etc.
• rocs -dbase vendor.oeb.gz -query 6cox.mol2

11
The query file

• Can be one or more structures, each with one or
  more conformers. By default, each
  conformer/structure is treated as a separate
  query
• File format can be any of SDF, MOL2, PDB, XYZ,
  MMOD, OEB
• Molecule(s) must already be 3D.

12
The database file

• Normally a pre-generated multi-conformer file.
• Can be one of several formats SDF, MOL2, PDB,
  XYZ, MMOD, OEB (OEBinary v2) or BIN (OEBinary v1)
• Contiguous conformers in file like SDF or MOL2
  will be combined into a single, multi-conformer
  molecule (by default)
• OpenEyes tool of choice is OMEGA for generating
  dbase files.

13
Other default settings

• -besthits 500
• number of results to keep in hitlist
• -cutoff 0.0
• minimum score to even consider
• -rankby tanimoto
• alternates tverskyd, tverskyq, scaledcolor,
  combo
• -prefix rocs
• text prefix for output file names
• -oformat sdf
• format of output structure file

14
Initial orientations for optimization

• Default - Inertial Frame alignment
• Overlay COM of query and target
• Large MOI aligned, then second largest. Including
  2-fold degeneracy of each yields 4 starting
  points.
• Extra 2 or 4 axes for top symmetry.
• Random
• -randomstarts N

15
A couple of simple examples

• rocs -dbase spam.bin -query acetsali.sdf -prefix
  ACET -cutoff 0.5 -besthits 100 -outputquery
• rocs -dbase spam.bin -query aminopy.sdf -rankby
  tverskyd -cutoff 0.4 -maxhits 10 -outputquery

16
Using VIDA to visualize results

• ROCS outputs (by default) two main files, a
  structure file and a report file.
• Structure file is by default SDF, scores are
  stored in SD tags and automatically loaded into
  VIDAs spreadsheet
• Report file is tab-delimited text, can be loaded
  into any spreadsheet for analysis or parsed by
  splitting each line on tabs.

17
Report File Format

• Tab-delimited
• Fields
• Name
• ShapeQuery
• Rank
• ShapeTanimoto
• Scaled Color
• ComboScore
• ColorScore
• SubTan

18
Hands-on examples

• Example molecules
• Several example molecules in data directory
• spam.bin is example multi-conformer database
• Documentation is available in HTML and PDF format.

19
Color Force Field

• Use SMARTS to describe color atoms or groups of
  atoms
• Post-shape scoring
• Color gradients can be used as part of
  optimization process

20
Color Force-field definition

• Define some atom types

TYPE donor TYPE acceptor TYPE cation TYPE
anion TYPE rings
21
Color Force-field

• Define patterns (SMARTS) that match the types

These definitions of donor and acceptor are the
general definitions of Mills Dean, JCAMD
10607-622, 1996. Donor an electronegative
atom with a proton (no S or C, see
above). PATTERN donor
7,8h,H Acceptor a lone pair on an
electronegative atom (O or N S was removed,
see reference). Note, N in an amide or in an
alkyl-aniline system is too conjugated to
accept, however, analinic NH2 is a potential
acceptor. PATTERN acceptor
OD1(O-,,6,15,16)!(1,2,3) PATTERN
acceptor nH0,N,8!(nX3)((-,
,ee)-,,ee)! (NC)!(ND2,D3-a)!(
1,2,3) PATTERN acceptor
NX3((-,,ee)(-,,ee)-,,ee)!(
1,2,3)
22
Color Force-field

• Define interactions between types
• Weight is strength of interaction, relative to
  shape gradients.
• Radius affects range of interaction

INTERACTION donor donor attractive gaussian
weight1.0 radius1.0 INTERACTION acceptor
acceptor attractive gaussian weight1.0
radius1.0 INTERACTION rings rings attractive
gaussian weight1.0 radius1.0
23
Example Force-field

• ROCS includes a very simple color force field
  simple.cff
• Also included is a more complete force field that
  include Mills-Dean definitions of donors and
  acceptors as well as types for rings, anions and
  cations. (MillsDean.cff)

24
Using Color

• To just score shape hits
• rocs -dbase spam.bin -query aminopy.sdf -chemff
  MillsDean.cff
• To use color gradients
• rocs -dbase spam.bin -query aminopy.sdf -chemff
  MillsDean.cff -optchem
• To also rank by color
• rocs -dbase spam.bin -query aminopy.sdf -chemff
  MillsDean.cff -optchem -rankby scaledcolor

25
Two extra scores with Color

• Scaledcolor
• Actual color score is sum of each best color
  interaction. Scaled color divides actual score
  into self-color of query giving score between 0
  and 1.0
• Comboscore
• To use shape and color together for ranking,
  comboscore is the sum of shape tanimoto and
  scaled color giving a score between 0.0 and 2.0

26
Advanced Features

• -maxconfs
• retain more than one db molecule conformer
• -scdbase
• treat dbase as single conformer, separate
  molecules
• -report each, one, none
• -stats hits, best, all
• -nostructs
• dont write structure file at all
• -pvmconf, -pvmpass

27
Roadmap I

• MOCS
• Multiple Overlay of Chemical Structures
• result is grid for query
• General grid query
• Pharmacophore pre-screening
• User-directed starting positions

28
Roadmap II

• ElectroROCS
• Electrostatic tanimoto
• Electrostatic gradients
• Shape fingerprint pre-screening
• Torsion tweak with MMFF
• Query Optimization
• Database Optimization

29
Acknowledgements

• Anthony Nicholls - OpenEye El Presidente
• author of Shape toolkit
• The OEChem hive-mind
• Geoff Skillman, Roger Sayle, and Matt Stahl
• Hewlett Packard
• computer loans for booth and seminar