Biomedical Engineering Reference
In-Depth Information
residues have been catalogued,
99
while characteristic shift changes for cysteine
residues report on the formation of disulphide bonds
100-102
or ligation to zinc
atoms.
103
The PREDITOR program uses probabilistic hypersurfaces to predict
rotameric states (g
+
, g
-
or trans) for the side-chain dihedral angle x1, along with
a confidence score.
91
More recently, the availability of stereospecific methyl
group assignments has led to simple recipes for defining rotamer populations in
the branched side-chains of valine (x1), leucine (x2) and isoleucine (x2)
residues.
104-108
Extensions of this approach to threonine, tryptophan and
methionine side-chains are under investigation.
104,109,110
3.7 Predicting Isotropic Chemical Shifts from Atomic
Coordinates
Alongside the effort that has gone into deriving structural restraints from
chemical shift measurements, several groups have studied the inverse problem
of how to use protein structure coordinates to predict observed chemical shifts.
Successful shift-prediction routines can be used to verify chemical shift
referencing, to validate resonance assignment sets or experimental structures
from NMR or X-ray crystallography studies, to refine structural models, to
restrain MD calculations, or even to determine protein structures solely from
chemical shift data by scoring candidate conformations.
5
The various suggested prediction methods can be divided into sequence-
based and structure-based approaches.
111
Sequence-based strategies find
matches for the primary sequence and structure of the query protein with
entries in a chemical shift database, and use these to generate shift predictions.
The most successful examples of this approach are SPARTA
112
and
SPARTA+,
113
both of which make shift predictions for backbone (
1
H
a
,
1
H
N
,
13
C9,
13
C
a
,
15
N) and
13
C
b
nuclei. SPARTA uses the sequence and w, y,
and x1 dihedral angles of triplet fragments from the query protein to harvest
shift predictions from a database, which are then corrected for neighbouring
residue effects and adjusted using phenomenological terms to account for
hydrogen bonding and ring-current shift effects.
112
SPARTA+ has additional
modifications, making use of a larger chemical shift/structure database and a
more sophisticated neural network analysis of backbone and side-chain
conformations, hydrogen bonding, electric field and ring-current effects in the
query protein.
113
By contrast, structure-based methods calculate chemical
shifts for the query protein directly from its atomic coordinates using a wide
range of approaches, including: empirically derived chemical shift hypersur-
faces (SHIFTCALC
30
and SHIFTX
114
); neural network analysis
(PROSHIFT
115
); look-up tables obtained from density functional theory
calculations on model peptides (CheShift
116
); and parameterised atom-pair
distance equations (CamShift
117
and BioShift
118
).
Overall, sequence-based methods are more accurate if the query protein
has a close homologue (.40% sequence identity) with known chemical
shifts, but structure-based approaches are superior for less familiar motifs.
111
