Biomedical Engineering Reference
In-Depth Information
3.5.2 Machine-Learning Techniques
Several machine-learning techniques including HMMs, support vector machines
and neural networks have been widely used to predict the membrane spanning
b
-strands in TMBs. Jacoboni et al. [
71
] proposed a neural network method
(B2TMPRED) for predicting the membrane spanning
-strands with 12 proteins
and reported the residue accuracy of 69% using sequence information only. Further,
they have raised the accuracy up to 78% by using the evolutionary information as
input to the network. Bagos et al. [
53
] developed a method based on HMM (PRED-
TMBB) and reported the residue accuracy of 84.2% in a set of 14
b
b
-barrel
membrane proteins. Bigelow et al. [
60
] proposed a method based on profile-based
hidden Markov model (ProfTMB) for prediction. Natt et al. [
54
] used neural
networks and support vector machines (TMMPred) for predicting the membrane
spanning regions of TMBs. They obtained the accuracy of 70.4% with only
sequence information and improved the accuracy up to 80.5% using the multiple
sequence alignment obtained from PSI-BLAST. Gromiha et al. [
72
,
73
] developed
a neural network-based method (TMBETA-NET) and it could predict the mem-
brane spanning regions of 13 TMBs with the accuracy of 73% using only the
sequence information. In addition, our method would provide the probability of
each residue to be in the transmembrane segment. Bagos et al. [
74
] combined
different methods and proposed a consensus approach (ConBBPRED)
for
predicting membrane spanning
-strand segments.
Recently, we have developed a novel method for predicting the topology of
b
b
-barrel membrane proteins [
75
]. The procedure involves two steps (1) predicting
the number of
-strand segments in TMH proteins and (2) predicting the residues in
membrane spanning-strands. In the first step, we have developed a protocol to
predict the number of
b
-strands in TMBs, and the flowchart is shown in Fig.
3
.
The information about the total number of residues in a protein is sufficient to
identify the
b
-barrel membrane proteins with 22 and 8 strands. For example, the
proteins with more than 590 residues are predicted to have 22
b
b
-strands. The
number of
b
-strands is 8 if the number of residues in a protein is less than 200.
22 β strands
Length > 590
8 β strands
Length < 200
10, 12, 14, 16 or 18 β strands
Classification
Fig. 3 Flowchart for predicting the number of
b
-strands in TMBs. Length denotes the total
number of residues in a protein [
75
]
