Graphics Reference
In-Depth Information
Table
.
.
he prediction results of eSEM with PSO_Set
applied to the alpha data set with
(
T
s
, T
e
)=(
,
)
Training set
Score
Num. of
Test set
Type func
ċ
α
β
T H
C u t o ff
predicted
Accuracy
accuracy
gene
pairs
TD
E
Cost
.
.
.
.
%
%
%
TC
(Std=
.
) (Std=
.
)
(Std=
.
)
%
TD
E
Cost
.
.
.
.
%
%
%
TC
(Std=
.
)
(Std=
.
)
(Std=
.
)
%
control group, respectively. Log ratios of R to G were used to reconstruct the genetic
interactions.
Among the RT-PCR-confirmed gene pairs, two-thirds (randomly chosen) were
used as the training set to tune the parameters of eSEM. Leave-one-out cross-valid-
ation wasperformedinthetraining settogive theaveragedaccuracy,whichwasused
to check against the prediction accuracy in order to detect any overfitting problem.
Finally, the trained eSEM was applied to the test set (
pairs) to yield the prediction
accuracy. eSEM was first applied to one cell cycle of data, consisting of the fourth to
the twelth time points, which correspond to the first and second peaks of the two
sine waves for genes exhibiting sinusoidal patterns. he prediction results of eSEM
with thefirstsetofthe PSOcostfunction aresummarized inTable
.
.heprediction
accuracy is defined to be the ratio of the number of correctly predicted pairs to the
total number of gene pairs in the test set (
). Checked against the RT-PCR results,
thepredictionaccuracyofeSEMwiththefirstsetofthePSOcostfunctionisabout
%. Note that if we consider only the
TC interactions in the test set, prediction
accuracies range from
% to
%.
Next,theperformanceofeSEMwasevaluatedfortheentiretimecourse(
time
points) of the alpha data set. Table
.
summarizes the results of eSEM with the first
Table
.
.
he prediction results of eSEM with PSO_Set
applied to the alpha data set with
(
T
s
, T
e
)=(
,
)
Training set
Score
Num. of
Test set
Type func
ċ
α
β
T H
C u t o ff
predicted
Accuracy
accuracy
gene
pairs
TD
E
Cost
.
.
.
.
%
%
%
TC
(Std=
.
) (Std=
.
)
(Std=
.
)
%
TD
E
Cost
.
.
.
.
%
%
%
TC
(Std=
.
)
(Std=
.
)
(Std=
.
)
%
