Next: Lower Bound Comparison
Up: Evaluation of the Technique
Previous: Predictor Evaluation
In order to evaluate chooser performance the following simulation
was performed: A network with 20 nodes, 24 edges and maximum
in-degree 4 was generated. The expression matrix E consisted of
the wild-type and all single perturbations. Next, 8 parsimonious
networks were inferred, all with 21 edges, which were consistent
with E. The chooser was used to select a double perturbation
which had maximal entropy score over the 8 networks, and the
process was repeated iteratively until only a single network was
inferred. The results are summarized in the figure
14.20. They show a pattern of jumps and decays in
the number of network solutions, correlated with an increase of
number of inferred edges.
Figure 14.20:
Summary of chooser evaluation: progress through experimental design
|
Peer Itsik
2001-03-04