Bayesian Networks in the Presence of Temporal Information - Bayesian Networks in R: With Applications in Systems Biology - page 80

G1 G

365

365

521 521

60 60

424 789 424 789

333 333

799 799

512 512

441 441

260 260

141

141

578 578

Fig. 3.7 Network structures learned by G1DBN in the first step ( G ( 1 ) ,onthe left ) and in the second
step ( G ,onthe right )

> edgesG = BuildEdges(score = step2,
+ threshold = 0.05, prec = 6)
The output of the second step is the matrix of the coefficients associated with the arcs
of the dynamic Bayesian network; the elements corresponding to arcs not present in
thegrapharesetto NA . The network structures learned in the first and second step
of G1DBN are shown in Fig. 3.7 . Essentially, the first step (left, Fig. 3.7 ) performs
dimension reduction, while the second step (right, Fig. 3.7 ) gives the set of arcs
defining the network.

3.5.4 Non-homogeneous Dynamic Bayesian Network
Learning: ARTIVA

The ARTIVA package provides several functions for structure learning, parameter
learning, and inference in order to facilitate the application of the ARTIVA approach
to dynamic Bayesian network learning and the interpretation of the its results.
An example of ARTIVA network learning is given below using a synthetic data
set called simulatedProfiles , which contains 55 genes and 30 time points
from the ARTIVA package.

> library(ARTIVA)
> data(simulatedProfiles)
Unlike the case of homogeneous dynamic Bayesian network, it is important to iden-
tify the target and parent genes prior to learning using simulatedProfiles .

> targets = c("1", "10", "20", "TF3", "45", "50")
> parents = c("TF1", "TF2", "TF3", "TF4", "TF5")