|
|
|
eHiTS: A new approach to scoring
receptor-ligand interactions
Zsolt Zsoldos1, A. Peter Johnson2,
Aniko Simon1, Irina Szabo1, Zsolt Szabo1
and David Fung1
(1) SimBioSys Inc., 135 Queen's
Plate Drive, Unit 355, Toronto, ON, M9W 6V1 Canada
(2) Institute for Computer Applications
in Molecular Sciences (ICAMS), Department of Chemistry, University
of Leeds, Leeds, UK LS2 9JT
Abstract:
The eHiTS® software of SimBioSys employs a very
rapid, exhaustive graph matching technique on chemical feature mapped polyhedra
graph representation of ligands and receptor cavities. This method typically
generates millions of flexible ligand poses for a given receptor structure
in a matter of seconds. The evaluation of such a vast set of poses demands
a quick scoring function with reliable filtering abilities.
After evaluating a wide range of published scoring functions, it was concluded
that none of them was sufficiently fast and accurate to distinguish the crystal
structure binding mode from other generated poses. Then a generalized function
was investigated. This included all the possible components, even some novel
penalty terms for unfavorable interactions. The large number of terms (18+)
were combined with adjustable coefficients. Furthermore, most of the terms
contained internal parameters (e.g. distance cut-offs or ideal contact distance
values) that may also be adjusted. Optimization studies have been performed
on the coefficients and the parameters of the terms with the goal to recognize
the crystal structure pose, i.e. the RMSD of the best scoring pose from the
X-ray structure was minimized on a set of 390 PDB complexes. Given recent
experiences of other workers in this area, it is perhaps not surprising that
even this method did not yield a generally usable scoring function. In most
of the cases eHiTS generated hundreds of poses with high RMSD and better
score than the X-ray structure. This suggested, that some significant components
are still missing from the scoring function terms.
A new approach was taken, based purely on local surface contact evaluation.
Properties of surface points were assigned with fine granularity, e.g. surface
point properties of a polar atom in an aromatic ring would be very different
along the edge of the ring from the faces of the ring. Hydrogen bonding is
also expressed as localized and concentrated surface property along the specific
proton or electron pair donation directions. Receptor surface points were
also assigned pocket-depth information to express differences in dielectric
constants on solvated surface points and deeply embedded cavity points.
This radically new scoring approach has yielded much better (it would be
good to provide some quantification of this) recognition of the X-ray structure
than any of the traditional atom based scoring functions.
|
Full Presentation
|
|
|
