Marc Delarue
Unite d'Immunologie Structurale, Institut Pasteur, 25 rue du Dr. Roux,
75015 Paris, France (delarue@pasteur.fr).
Patrice Koehl
UPR 9003 (Cancerogenese et Mutagenese Moleculaire et Structurale)
E.S.B.S. Blvd S. Brant, 67400 Illkirch-Graffenstaden, France
(koehl@sulawesi.u-strasg.fr)
We have sent essentially four pdb files:
- 1. ndpk.pdb, which represents the predicted coords of Nucleotide
Diphosphate Kinase, as modelled from an alignment with the Drosophila
melanogaster sequence and structure. The structure predicted is
essentially complete.
- 2. edn.pdb, which is the eosinophil derived neurotoxin modelled on
pancreatic bovine ribonuclease A. The main difficulty was the insertion
of 9 residues, and a deletion from 10 to 4 residues between two alpha
helices in the N-term part of the protein. We had to suppose the the
Nterm residue of the second helix and the C term one of the first helix
were not in a alpha helical conformation. Also, the last three C-term
residues have not been modelled.
- 3. hpr.pdb (histidine containing phospho carrier): hpr1.pdb is
modelled upon the S. faecalis structure and hpr2.pdb is modelled upon
the E.coli one. The first and last residues have not been modelled in
this last case.
Here follows a brief word about the methodology employed: The method
is entirely automatic and uses as an entry the alignment of two
sequences (the one to be modelled and the one of the known related
structure) plus the coords of the model. Side chains are placed
according to a recently published method based on a self-consistent
mean field method (Koehl & Delarue, J. Mol. Biol. 239:249-275). A
conformational matrix, whose M(i,j) elements represent the probability
of finding the ith rotamer at the jth position, is iteratively refined
using a simple mean field theory: each residue "feels" the average of
all possible rotamers (copies) of all its neighbours, weighted by their
respective probabilities; in this way, energies for each possible
rotamer are being evaluated for each position; these energies are then
transformed into probabilities using Boltzmann formula, and used to
update the conformational matrix before going on to the next cycle. The
procedure converges in a few cycles. The backbone is essentially fixed
and the energy used is only van der waals. This is just a preliminary
version of the program. More elaborated energies are being incorporated
right now. The program also automatically takes care of insertions and
deletions by generating multiple copies of possible backbones joining
the two ends of the loop, as found in the protein data base (in a
method very similar to the one of Jones and Thirup). In principle, the
method could use multiple copies of the entire backbone: it has been
shown recently to converge towards the right result in test cases
(Koehl and Delarue, Nature, Structural Biology, submitted). However,
for computer time limitations reasons, this has not been used here.
Asilomar Conference home page