Bayesian Networks in the Absence of Temporal Information - Bayesian Networks in R: With Applications in Systems Biology - page 55

Whitelisted Interventions Learned Interventions

INT

INT

p44.42

pmek

pakts473

praf

PKA

PKA

p44.42

PKC

PKC

pmek

P38

pakts473 P38 plcg

praf plcg pjnk

pjnk

PIP3

PIP3

PIP2

PIP2

bn.wh
Modified BDe

bn.tiers
Sachs et al. Validated Network

plcg

PKA

PKC

plcg

PKC

PIP3 PKA

praf

PIP2

PIP2

P38 praf pjnk

pmek

PIP3

pmek

p44.42 pjnk

p44.42

pakts473

P3 8

pakts473

bn.mbde

Fig. 2.10 Bayesian networks learned from isachs .Thefirsttwonetworks( bn.wh on the top
left , bn.tiers on the top right ) have been learned by including INT and adding arcs to model
stimulatory cues and inhibitory interventions. The third network ( bn.mbde ,onthe bottom left )
has been learned with model averaging and the mbde score; arcs highlighted with a thicker line
width make up the validated Bayesian network ( bottom right ) from Sachs et al. ( 2005 )

The approach used in Sachs et al. ( 2005 ) yields much better results. Instead of
including the interventions in the network as an additional node, Sachs et al. ( 2005 )
used a modified BDe score (labeled “ mbde ”in bnlearn ) incorporating the effects
of the interventions into the score components associated with each node ( Cooper
and Yoo , 1995 ).
Since the value of INT identifies which node is subject to either a stimulatory cue
or an inhibitory intervention for each observation, we can easily construct a named
list of which observations are manipulated for each node.