Biology Reference
In-Depth Information
The most interesting characteristic of ARTIVA is that it considers
simultane-
ously
all possible combinations of CPs and all possible network topologies within
the different phases. Furthermore, ARTIVA allows network structures associated
with different nodes to change with time in different ways, and at the same time it
allows each variable to be simultaneously parent and target. This flexibility proves
problematic for time series with a low number of measurements, such as the ones
typically available in systems biology, leading to overfitting or inflated inference
uncertainty. This limitation can be overcome with the semiflexible model proposed
in
Dondelinger et al.
(
2013
), which is based on a piecewise homogeneous dynamic
Bayesian network regularized by gene-specific intersegment information sharing.
This approach is implemented in the
R
package EDISON (Dondelinger et al. 2012).
3.5 Dynamic Bayesian Network Learning with
R
In this section, we discuss popular
R
packages for investigating multivariate time
series including several diagnostic tests to assess the goodness of fit and justifying
the inherent assumptions in VAR models. Subsequently, we present examples using
R
packages for dynamic Bayesian network modeling. In this regard, we discuss
examples from time-homogeneous (Sects.
3.5.2
and
3.3.2
) as well as time-varying
multivariate time series (Sect.
3.5.4
).
3.5.1 Multivariate Time Series Analysis
Auseful
R
package for analyzing time seriesasaVARprocessesis
vars
. Consider
an example below using the data set
Canada
with 4 macroeconomic indicators
(
prod
, labor productivity;
e
, employment;
U
, unemployment rate; and
rw
, real
wages).
> library(vars)
> data(Canada)
A vector auto-regressive process VAR(
p
) can be fitted from these data with the
VAR
function.
> VAR(Canada, p = 2)
The object of class
varest
returned by
VAR
contains information on several as-
pects of the VAR process and its parameters. A summary of the results of
VAR
can
be obtained using the function
summary
as follows.
> summary(VAR(Canada, p = 2))
Function
VAR
has an argument called
type
which allows the use of different
types of deterministic regressors. It defaults to
const
, which adds the intercept
Search WWH ::
Custom Search