[Chimera-users] Intersurf, protein interface display

[Thomas Goddard]

Hi Conrad,

  I read the paper

    Intersurf: dynamic interface between proteins
    Nicolas Ray, Xavier Cavin, Jean-Claude Paul, Bernard Maigret
    Journal of Molecular Graphics and Modelling, 23 (205) 347-354

It addresses 3 problems: how to find a surface interface between 2 proteins,
ways to color such a surface, and how to flatten that surface into two
dimensions.

The surface calculation was a neat algorithm -- take the
Delauney tetrahedralization of all atom positions from both proteins,
then find tetrehedra with vertices from both proteins and cut those
tetrahedra in half.  The tetrahedra are cut in half with one or 2
triangles using vertices on tetrahedron edges that connect an atom
in one protein to an atom in the other protein.  The vertices are
placed midway between the atoms moved in by their vdw radii.
 
They proposed 2 ways of coloring the surface.  One colored vertices
by how far apart the atoms were in the two proteins that were used to
create that vertex.  Red to blue shows short to long distances.  The
other coloring shows types of possible interactions between the two
residues associated with the two atoms for the tetrahedron edge that
the surface vertex splits.  They gave different colors for predicted
hydrogen bonding, hydrophobic packing, pi ... X, pi ... pi, same charges,
opposite charges.  Not sure what the pi bonding notation means.

They also described how to flatten the 3-dimensional to a 2-dimensional
surface with one to one correspondence between vertices and edges.
Their method tried to come as close as possible to a conformal map.
Conformal means the angles of the triangles stay the same to the extent
possible.

These calculations took a few seconds on 2000-5000 atom proteins.

Here's my assessment.  I think the concept of showing a protein interaction
interface and coloring it to show interactions sounds useful for protein-
protein interaction people.  I think the flattening into 2 dimensions
is probably not useful.  Conformal maps greatly distort the area of
the surface if it has curvature (you've seen the maps where Greenland
is bigger than Africa, Americas, Asia and Europe combined).  So the flattened
surface is going to be hard to recognize as the interface, unless it
is already pretty flat.  If it is already pretty flat you might as
weill just look at the 3-d surface.

If we wanted to do something with interface surfaces I would suggest
talking to Tanya Korteme.  We already have all the infra-structure to
create the surfaces.  I would not use the Delauney surfaces, instead
just use MSMS surfaces on both proteins showing only the parts associated
with atoms in one protein that are within 5 angstroms of atoms in the
other protein.  The zone calculation code I wrote for multiscale could
do that fast.  We could offer a range of atom colorings that map onto
the surface (hydrogen bonds, distance to nearest atom in other protein,
charge complementarity, etc...).

I don't have time or enough interest to work on this project.  But
perhaps someone else will.  We have talked about collaborating with
Tanya Korteme in the past and this would be a good opportunity.

      Tom