Information Technology Reference
In-Depth Information
X
20
i
¼
1
a
i
b
i
DotProductSco
ða; bÞ¼
ð
1
:
8
Þ
This can be interpreted as the probability of identical amino acids being produced
from distributions
independently. A variant of this scoring function is to
calculate dot product using log-odds values as follows.
α
and
β
X
20
DotOddSco
ða; bÞ¼
log
ð
a
i
=
p
i
Þ
log
ðb
i
=
p
i
Þ
ð
1
:
9
Þ
i
¼
1
Jensen-Shannon function
This scoring function was introduced by Yona and
Levitt [
56
], which measures similarity of two probability distributions using
information theory. The similarity measure is based only on the observed proba-
bility distributions, so it is independent of any evolutionary models. The similarity
score of two pro
le columns is de
ned as a combination of their statistical simi-
larity and the signi
cance of the statistical similarity. In particular, the scoring
function involves the calculation of a divergence score,
"
#
2
X
X
20
i
¼
1
a
i
log
20
i
¼
1
b
i
log
a
i
ða
i
þ b
i
Þ=
b
i
ða
i
þ b
i
Þ=
1
DivSco
a; ðÞ¼
þ
ð
1
:
10
Þ
2
2
and a signi
cance score.
"
#
þ
2
X
X
20
i
¼
1
a
i
log
20
i
¼
1
b
i
log
a
i
p
i
b
i
p
i
1
SigSco
a; ðÞ¼
ð
1
:
11
Þ
le scoring functions and their comparison, please refer to
[
58
]. Experimental results indicate that the log-odds-based scoring functions, such
as DotOddSco and Jensen-Shannon, more likely to perform better than many others
[
56
].
For more pro
le-pro
1.5 Contribution of This Topic
To signi
cantly advance remote homology detection and fold recognition, this topic
focuses on pro
le alignment, although the method presented in this topic can
be easily adapted for sequence-pro
le-pro
le alignment. In particular, this topic describes
a Markov Random Fields (MRFs) representation of sequence pro
le. That is, we
use MRF to model a multiple sequence alignment (MSA) of close sequence
homologs. Compared to Hidden Markov Model (HMM) that can only model
local-range residue correlation, MRFs can model long-range residue interactions
Search WWH ::
Custom Search