SimBioSys Inc.© 2003 eHiTS: Novel algorithm for fast, exhaustive flexible ligand docking and scoring Zsolt Zsoldos, Aniko Simon,

SimBioSys Inc.© 2003http://www.simbiosys.ca/

eHiTS: Novel algorithm for fast, exhaustive flexible ligand docking and scoring

Zsolt Zsoldos, Aniko Simon, Irina Szabo, Zsolt Szabo, David Fung and A. Peter Jonson

SimBioSys Inc., © 2003

Contents:

●Introduction: docking methods, exhaustive search definition●Method: new docking algorithm implemented in the eHiTS

software●Scoring function: surface map based, including weak

interactions●Validation results and summary

http://www.simbiosys.ca/


Contents: Introduction

• Introduction

• Categorisation of the docking algorithms• Definition of exhaustive docking search• The search space size

• New docking algorithm of the eHiTS software

• Scoring function

• Validation results


Yet Another Docking Algorithm, YADA, yada...

GOLDDOCK FlexX


Exhaustive Conformation and Pose Search in our definition

• Discretisation must be fine enough to differentiate any atom movement >=0.5Å

• Rotation and dihedral sampling must be finer than 5°

• 30° sampling can miss perfect hydrohob fit by 1.5Å clashes

PDB code: 1HRI

~7Å

PDB code: 1CX2


Search Space Size

• Number of poses to examine with sampling defined:• Translations(0.5Å) * Rotations(5°) * Dihedrals(5°) =

203 * 723 * 72n ~ n=6 rot.bonds => 2*1020 poses per ligand

• Brute force evaluation 2000/s => 3 billion years

• Stochastic methods explore a tiny fraction of the space with no guarantee to find the best or even any good-enough solutions


Contents: new docking algorithm

• Introduction

• New docking algorithm of the eHiTS software:

• Overview of the algorithm

• Rigid fragment docking

• Graph matching by clique detection

• Flexible chain fitting




eHiTS algorithm overview: Divide and Conquer

• Ligand is divided into • rigid fragments and

• connecting flexible chains

• All rigid fragments are docked independently

• Graph matching

• Flexible chain fitting

• Local energy minimisation

Patent pending algorithm:


Rigid fragment docking based on Chemical feature mapped polyhedra

• Polyhedron shrink-wrapped onto molecular surface (Connolly)

• Chemical feature flags on vertices

• Analogue cavity representation

• Rapid mapping of ligand and cavity polyhedra

D

D

D

DA

A


Graph matching by clique detection

• Each rigid fragment is now individually docked to several alternative sites

• Task is to find sets of poses such that:• fragments can be linked by flexible chains

• they do not bump into each other

• Graph nodes: ligand fragment poses

• Graph edges: pose pairs at distance compatible with chain length

• Cliques encode docking solutions


Flexible chain fitting

• Find chain conformation to link given pair of fragments:

• Double diamond latice to start from

• Tweak the chain to fit end points and avoid boundary (deterministic)


Contents: scoring function

• Introduction


• Scoring function:

• Surface map based ranking score• Chemical feature flags• Weak (secondary) interaction flags



Surface map based ranking score

• Chemical property flags are mapped onto 0.5Å resolution Connolly surface

• Receptor surface point weights determined by depth in pocket

• Flag compatibility matrix score for receptor-ligand contacts

• Exposed receptor points get penalty based on depth value


Feature flags indicate properties

• H-bond Donor: {D/d} • H-bond Acceptor: {A/a}• Hydrophobic flags: {H/h}• Aromatic Pi-stack:ring {R},

center {O}, edge {E}• Partial charges: {p/n}• Formal charges: {P/N}• Metal ions and ligands: {M/m}

HdE

HdE

nhR

nhR

HR

HRHO

DApnmE

DApnmE


Weak interaction flags

• Secondary (weak) H-bond donor flags are assigned to Carbon atoms in polar environment (e.g. in aromatic ring, next to Nitrogen)

• Heteroatoms in aromatic rings get secondary hydrophobic flags in direction of electrons

• Secondary flags contribute to score and avoid penalties

PDB code: 1AHA


Contents: validation results

• Introduction



• Validation results:

• Examples: 1NIS, 1TDB, 1SRJ, 1FKG, 3CPA, 4SLI

• Graph and table of RMSD results on validation set

• Speed performance in different modes

• Summary


Example: PDB code 1NIS

Stick & ball = X-ray crsytal structure ligand pose

Closest (0.6Å RMSD) Top rank (0.9Å RMSD)

Closest solution found by GOLD had 3.5Å RMSD


Example: PDB code 1TDB,that GOLD has failed to dock




Example: PDB code 1SRJ, where“FlexX failed due to structural errors”




Example: PDB code 1FKG, “too complex task for FlexX”




Example: PDB code 3CPA, where DOCK failed




Example: PDB code 4SLI, where DOCK failed




RMSD results on the validation set

0 1 2 3 4 50

10

20

30

40

50

60

70

80

90

100

Closest

Top-rank

Accuracy Top-rank Closest16% 53%60% 98%81% 100%

Average:

RMS<1ÅRMS<2ÅRMS<3Å

2.17Å 1.03Å

The Astex/CCDC validation set of GOLD was used:

105 PDB to tune parameters200 PDB complexes to test


eHiTS speed

Accurate Default Fast VHTS

Rigid frag. 513 87 3.8 0.3

Graph match 35 4 0.6 0.6

Optimize 124 35 0.7 0.7

Total 672 126 5.1 1.6

Seconds per ligand on single CPU Linux PC2.4GHz Pentium 4, 512MB RAM

VHTS mode employs an SQL database to store rigid fragment poses and re-use it for other ligands having the same fragment


Summary

• Advantages of the eHiTS software:

• Exhaustive search – don't miss a potential drug!

• High accuracy – the most important

• Deterministic – no need to wait for a lucky day

• Fully automated - no manual setup, seed selection

• Highly configurable scoring function

• Very fast - extremely High Througput Screening of large databases or virtual libraries

http://www.simbiosys.ca/


Acknowledgements

• Team: Irina Szabo, Zsolt Szabo, Aniko Simon, David Fung, A. Peter Johnson and Zsolt Zsoldos

• Funding support by Canadian government grants:• NRC - IRAP project #411461

• GeoDock SR & ED project (2001)

• NRC - IRAP project #468906

• EnerMin SR & ED project (2002)

Documents

SimBioSys Inc.© 2003 eHiTS: Novel algorithm for fast, exhaustive flexible ligand docking and scoring Zsolt Zsoldos, Aniko Simon,