Biomedical Engineering Reference
In-Depth Information
0.08
10mins
30mins
60mins
120mins
0.07
0.06
0.05
0.04
0.03
0.02
0.01
10
721
744
849
Attractor Cycle Order Number
Fig. 2.3
Average predictor error difference on melanoma attractor data using MiniSat with random
variable selection modification
average difference in all the predictors' occurrence frequency in the complete All-
SAT result with the results obtained with shorter All-SAT runtimes (10, 30, 60, and
120 min). Figure
2.3
shows the average error difference of all predictors' frequency
for the four orderings, using MiniSat with the random variable selection modification
(100 % random variable frequency), while Fig.
2.2
shows the same results without
random variable selection (2 % random variable frequency). Across the four order-
ings analyzed, the average error difference of all predictors' occurrence frequency
(shown in Figs.
2.3
and
2.2
) is significantly lower using the random variable selection
modification than without. Furthermore, the average error difference decreases with
increasing runtime when using random variable selection. From this experiment, we
determine that 30 minutes with random variable selection was sufficient to achieve
an average of
5 % difference in the predictors' occurrence frequency compared to
the full All-SAT results.
The following presents our results after collection of All-SAT results from all valid
attractor cycle orderings. In Fig.
2.4
, we display a histogram of all logically valid
predictors and their frequency of occurrence, across all attractor orderings. In the
sequel, a predictor label of 2367 means that gene
g
2
is predicted by genes
g
3
,
g
6
, and
g
7
. From this chart, we can observe that certain predictors occur with significantly
higher frequency than others. For example with gene
g
1
, the predictor
≤
{
x
3
,
x
5
,
x
7
}
(
PIRIN
predicted by
RET1
,
HADHB
,
WNT5A
) occurs with much higher frequency
than all other predictors for gene
g
1
. This indicates that this predictor is most likely
to be present in the final predictor set. From this data, we propose three methods (A,
B, AB) for selecting the predictor set.
