Structural and Functional Discrimination of Membrane Proteins - Ion Channels and Their Inhibitors - page 7

nonmembrane region, inside TMH-cap, membrane helix, outside TMH-cap, and
outside nonmembrane region.
Rost et al. [ 27 ] developed a method based on neural networks for identifying the
location and topology of transmembrane

-helices. Further, the method has been
refined by postprocessing the neural network output through a dynamic program-
ming-like algorithm, similar to the one introduced by Jones et al. [ 28 ]. Persson and
Argos [ 29 ] proposed a method based on multiple sequence alignment for predicting
transmembrane helical segments and their topologies. They have used two sets of
propensity values: one for the middle, hydrophobic portion and the other for the
terminal regions of the transmembrane sequence spans. Average propensity values
were calculated for each position along the alignment, with the contribution from
each sequence weighted according to its dissimilarity relative to the other aligned
sequences.
Although multiple sequence alignments improve prediction accuracy, there are
no homologues in current databases for 20-30% of all proteins [ 30 ]. To overcome
this situation, the so-called dense alignment surface (DAS) method was developed
[ 31 ]. DAS is based on the scoring matrix originally introduced to improve
alignments for G-protein-coupled receptors. It compares low-stringency dot-plots
of the query protein against the background representing the universe of nonho-
mologous membrane proteins using the scoring matrix.
Nugent and Jones [ 32 ] developed a method based on support vector machines for
predicting the topology of membrane proteins along with signal peptides and
reentrant helices. Osmanbeyoglu et al. [ 33 ] utilized an active learning approach
for predicting transmembrane helices in membrane proteins. Ahmed et al. [ 34 ]
presented a new transmembrane helix topology prediction method that combines
support vector machines, HMMs, and a widely used rule-based scheme.
Further, methods have been developed for predicting helix-helix interactions
and modeling the three-dimensional structures of membrane proteins. Park and
Helms [ 35 ] proposed a method to predict the assembly of transmembrane helices in
polytopic membrane proteins using sequence conservation patterns. Fuchs et al.
[ 36 ] developed a method for predicting the interacting helices and helix-helix
contacts in polytopic membrane proteins using neural networks. Michino et al.
[ 37 ] developed a protocol, FoldGPCR for modeling the transmembrane domains of
G-protein-coupled receptors. Recently, de Brevern [ 38 ] reviewed the developments
on the 3D modeling of membrane proteins. The general methodology composed of
the following: (1) use of secondary structure prediction to complete the compara-
tive modeling process, (2) perform refinement and assessment steps to a novel
comparative modeling process, and (3) consider the helix-helix and helix-lipid
interactions, and even build quaternary structures. However, the most important
factor when proceeding to correct structural models is taking the experimental data
into account.
The servers for identifying TMH proteins and predicting the membrane spanning
helical segments are listed in Table 2 .

a