PASS (putative active sites with spheres)

1
PASS (putative active sites with spheres)

• Fast prediction and visualization of protein
  binding pockets with PASS
• G. Patrick Brady Jr. Pieter F.W.Stouten
• Journal of Computer-Aided Molecular Design, 14
  383-401, 2000 KLUWER/ESCOM
• Presented by Mareike-Danica Moritz

2
Overview

• Overview over PASS
• Introduction
• Methods
• Tests
• Discussion
• Conclusion

3
Overview over PASS

• Characterization of buried volume in proteins
  binding sites (BS)
• Based upon size, shape burial extent of these
  volumes
• Tested by predicting known BS of proteins in PDB
• ? PASS as an front-end to fast docking
• Analysis of a moderate-size protein in under 20s
• ? suitable for molecular modeling, protein
  database analysis, aggressive virtual screening
  efforts
• Output standard PDB files, suitable for any
  modeling package script files

4
Introduction

• 3D geometry of ligands dictated by size shape
  of protein cavities (hand in a glove)
• ? sterical fit as minimal requirement for drug
  activity
• Surface representation infer buried vol. from
  often occuled void space ? methods for direct
  display important
• increasingly tempting interfacing molecular
  docking virtual screening tools with functional
  genetics ? threading / homology modeling
• PASS virtual screening visualization aid for
  manual molecular modeling

5
Methods Probe sphere (PS) Calculation (Overview)

• Cavities in protein structures are filled with
  set of spheres
• Filling layers of spheres, 3-point-Conolly like
  sphere geometry
• 1st layer calculated with protein as substrate
• Additional layers accreted onto previously found
  PS
• Only PS with low solvent exposure are retained
• When accretion layer produces no new buried PS
  the calculation finishes

6
Methods calculation of PS

• Reading PDB coordinates of target protein
• Assigning elemental atom radii
• Compute 1st layer
• - looping over all unique triplets of protein
  atoms
• - if close enough together calculation of the
  two locations at
• which a PS may lie tangential to all 3 protein
  atoms
• Probe must survive 3 filters
• - 1. no overlap with any atoms of the accretion
  substrate
• - 2. not clash with any protein atoms
• - 3. buried within the protein

7
Methods calculation of PS

• Weeding of PS no probe centres closer together
  than R(weed) 1Å ! ? no clumping of additional
  layers
• - Iterative accretion onto existing layers
• - retention of the set of all PS from previous
  PS layers
• ? new accretion substrate
• - min centre-to-centre distance (defined) from
  each protein
• atom
• No newly-found PS survive the filters ? accretion
  phase stops

8
Methods calculation of PS

• Results of the calculations
• 1. Cavities filled with a set of fairly evenly
  spaced PS, all
• buried, no sterical clashes with the protein
• 2. Probes lying along protein surface are packed
  in ideal steric
• contact with 3 protein atoms

9
Methods calculation of PS
10
Methods Active site point (ASP) determination

• Final set of PS ? identification of small number
  of ASPs
• ASPs
• - represent potential binding sites for ligands
• - selection regions that contain many spheres
  with high burial
• counts (BC) ? central probes
• Assignment of probe weights (PW) for each probe
• Final ASPs
• - cycling through the probes in descending order
  of PW
• - only those with PW PW(min) are kept
• - separated by min distance P(ASP)8Å
• - rank-ordering to PW values of ASPs
• ?PASS predicting binding sites (BS)

11
PASS output

• PDB file with the final set of PS
• PDB files of ASPs
• PDB file for each ligand that optionally was read
  in
• Additional
• Visualization scripts for several popular
  molecular modeling packages (Insight III, Rasmol)
• Scripts simplify visualization by automatically
  loading, rendering, coloring the protein, PS,
  ASPs, ligand(s)
• Detailed runtime information
• Coordinates of bound ligands
• For each ligand Computation of distances from
  each ASP to nearest ligand atom and ligand center
  of mass
• Command-line options

12
Results (Test Complex)

• Test 30 protein-ligand complexes drawn from PDB
• Chosen by diversity, resolution, inclusion in
  previous theoretical studies, existence of
  corresponding apo-protein X-ray structures in PDB
• Hydrogen-free PASS parameters
• Binding site hit computed by coordinates of the
  known ligand(s)

13
Results (Test Complex)

• PASS successfully identifies the locations
  of known BS in complexed protein structures
• Correct location of pockets in all but 3 of the
  32 trials
• 11 times found multiple BS hits for a given
  ligand
• In 19 (of 32) trials the top-ranking ASP
  represents a BS hit
• In 26 trials one of the top 3 ASPs is a hit
• ?PASS identification of protein cavity in
  seconds
• ?Correlation between ASP rank (i.e.PW) and
  volume of the corresponding group of PS

14
Results (Test apo-proteins)

• More realistic 20 apo-protein structures from
  PDB (corresponding to table 3)
• Determination of known BS (binding site)
  positions
• - superposing native complexed structures
• - calculation proximity of ASPs to known ligand
  
• ? computation BS hits
• Protein deformation by ligand
• - calculation of rmsd between superposed
  structures
• - use only residues lying in the BS

15
Results (Test apo-proteins)

• ?PASS is able to predict BS locations when only
  the apo X-ray structure is known!
• Similarity of observed hit rates ? presence of a
  ligand in the structural data not necessary for
  al successful cavity detection algorithm?
• More mode enhanced set of probes ASPs

16
Results Experimental Test (Mattos Ringe)

• Comparison with experimental data organic
  solvents in elastase crystals
• PASS was run over elastase
• Graphically superimposed resulting ASPs with
  Ringe et al.s organic probes and bound ligands
• Results
• - several clusters of organic probes
• - i.a. large grouping in the active site (S1
  pocket)
• - only one organic probe within 8Å of top or 2nd
  ranked ASP
• - PASS places an ASP near 4 of the 5 largest
  clusters of
• probes

17
Results Organic probes more predictive than ASPs??

• Contrast no ASP near S4 pocket but cluster of
  organic probe!
• Ringe small, often polar molecules on surface ?
  PASS ASPs (large, apolar only cavity of
  critical size can sustain an ASP)
• ?PASS and Ringe et al. appear comparably
  predictive of known BS in elastase

18
Discussion PASS in a virtual screening environment

• Hit rates (table 4) ? PASS as front-end to
  virtual screening
• screening tool fast enough ? docking against
  multiple sites ? run separate screening
  calculations ? in most cases identification of
  optimal binding mode ( 71 hit rate of top 2
  ASPs)
• Selection of the true BS more detailed
  Docking routine
• ? for different ASP regions possibility of
  comparison of the affinity of a ligand

19
Results PASS in a virtual screening environment

• Screening small set of probe molecules against
  all ASPs, comparison of top binders
• ? identification of the stickiest region of
  the protein
• ? screening large database against this region
• ? construct combinatorial libraries
• ? characters to select site with the highest
  probability to
• have affinity for a given class of
  compounds
• Tempting aspects of PASS speed
• - fast analysis of entire structural databases
  (PDB, Corporate)
• - suitable bridge between 3D structural modeling
  ligand
• docking

20
Results PASS as an interactive visualization tool

• PASS less than 20s for calculation of a
  moderate-sized protein ? interactive usage in a
  molecular modeling environment utility as
  visualization tool for drug design
• PASS PDB-files can be loaded and rendered
  separately
• Or PASS visualization ? full display of PASS
  output
• ASP coloring PW denoted ? PS colored by
• - burial count (BC)
• - group identity -group
• - layer of accretion in which each was identified

21
Results PASS as an interactive
visualization tool
22
Discussion PASS as an interactive visualization
tool

• Advantage
• ASPs displayed relative to the protein, centrally
  located in cavities ? quick identification of
  residues in modulating binding
• PASS visualization scripts automatically define
  6, 8, 10Å residue-based subsets around each ASP ?
  facilitate coloring and specific display of these
  regions

23
Discussion PASS as an interactive visualization
tool

• PASS identifies multiple ASPs
• Computation of difference map (bound ? unbound
  forms) ? identification visualization of
  packing contacts in protein crystals or
  multimeric forms ? PASS can identify interfacial
  pockets
• PS representing buried volume can be viewed
  manipulated as a solid object ? visualization of
  shape complementary of a protein surface
• Ligands only bind to regions possessing enough
  buried volume to accommodate them ? buried volume
  as quantity of central importance in drug design

24
Discussion PASS as interactive visualization tool

• Property-based coloring of PS ? information
  equivalent to what is color-coded onto protein
  surface display ? color according to
  electrostatic potential
• No automated interaction-based coloring in PASS

25
Discussion Comparative study regular grid ? PASS

• Grid
• - coordinates of points lying in cavities ?
  identify boundaries
• - disadvantage consumes memory unnecessarily
  uncertainties
• - advantage purely algorithmic
• PASS
• - uses only 3-point geometry to obtain points
  lying in cavity regions
• ( sterical optimal packing)
• - might expect spotty distribution of probes
  poorly shaped buried
• volume
• - practical experience produces smooth
  well-shaped buried volume
• - 3-point geometry helps minimize number of
  points required to fill
• protein cavities

26
Discussion Comparative study sea-level

• Some operate by filling fully enclosed volumes ?
  artificial definition of sea-level
• Others definition as byproduct of the algorithm
  itself
• Kuntz et al. - Connolly surface ? sphere growth
• - angular condition for spheres (concave, flat,
  convex)
• - radial constraint for spheres (5Å)
• PASS size solvent accessibility as quantified
  by burial counts ? sea-level
• PASS rates favorably with all published methods
  in Speed ease of use (less familization
  training)
• Fastest CPU times belong to LIGSITE, but time
  increases strongly as grid spacing is reduced.
  PASS nearly the same speed (20s)

27
Discussion

• Disadvantages
• 3-point geometry ? no identification of cavities
  in flat (i.e. aromatic) proteins
• Prescribed distance of 8Å ? no identification of
  cavities in very small proteins
• Output input as PDB files ? no exact allegation
  of charge of ligand/protein

28
Conclusion

• PASS
• Simple cavity detection tool
• Utility in virtual screening interactive
  molecular modeling environments
• Reliably predicts the location of known BS ?
  utility as front-end to fast docking virtual
  screening
• Visualize buried volume in a protein (price 30s)
• Suggests alternate BS
• Simplifies detailed visualization of potential
  binding hot spots
• http//www. rcsb.org/pdb/software-list.html
• http//www.ccl.net/chemistry

29
References

• G. Patrick Brady Jr. F.W. Stouten, Fast
  prediction and visualization of protein binding
  pockets with PASS, Journal of Computer-Aided
  Molecular Design,14 383-401, 2000
• Ringe, D. Curr.Opin.Struct.Biol., 5 (1995) 825
• Ruppert, J., Welch, W. and Jain, A.N. Protein
  Sci., 6 (1997) 524
• Connolly, M.L., J.Appl. Crystallogr., 16 (1983)
  548