Bayesian Networks in the Absence of Temporal Information - Bayesian Networks in R: With Applications in Systems Biology - page 48

The relevant arguments are idisc and ibreaks , which control how the data are
initially discretized, and breaks , which specifies the number of levels of each dis-
cretized variable. Choosing good values for these arguments is a trade-off between
quality and speed; high values of idisc preserve the characteristics of the origi-
nal data to a greater extent, whereas smaller values result in much smaller memory
usage and shorter running times.
Each variable in the dsachs data frame is a factor with three levels, correspond-
ing roughly to low, average, and high expression. Now that the data are ready for
the analysis, we can apply bootstrap resampling to dsachs to learn a set of 500
network structures to use for model averaging.

> boot = boot.strength(data = dsachs, R = 500,
+ algorithm = "hc",
+ algorithm.args = list(score = "bde",
+ iss = 10))
> boot[(boot$strength > 0.85) &
+ (boot$direction >= 0.5), ]

from to strength direction

1 praf pmek 1.000 0.573

23 plcg PIP2 1.000 0.624

24 plcg PIP3 1.000 0.994

34 PIP2 PIP3 1.000 0.997

56 p44.42 pakts473 1.000 0.616

57 p44.42 PKA 0.998 1.000

67 pakts473 PKA 1.000 0.995

89 PKC P38 1.000 0.530

90 PKC pjnk 1.000 0.982

100 P38 pjnk 0.962 1.000

Bootstrapped Network

Sachs et al. Network

PKC PKC

P38 P38

plcg plcg

PIP2 PIP2

pjnk pjnk

PIP3 PIP 3

p44.42 p44.42

pakts473 pakts473

PKA praf PKA praf

pmek pmek

Fig. 2.8 Averaged network learned from the observational data studied in Sachs et al. ( 2005 )(on
the left ) and the network learned with bootstrap resampling from the same data (on the right )