Description Usage Arguments Details Value Note Author(s) References See Also Examples
Provides three types of structure priors. laplaceinhib
and laplace
penalise the
difference between the actual network and a reference network. scalefree
penalises high
node degrees.
1 2 3 4 |
phi |
The candidate network. |
lambda |
Laplace prior hyperparameter describing the prior influence strength. |
B |
Laplace prior probability matrix. |
Z |
Laplace prior normalisation factor for the prior. (Not used at the moment.) |
gam |
Scalefree prior degree distribution coefficient: P(k) ~ k^gam or exponent for difference term in laplaceinhib prior. |
K |
Scale-free prior scaling factor/Strength |
it |
Scale-free prior number of iterations for prior sampling. |
priortype |
String. Either |
For the laplaceinhib and laplace prior types, the matrix B is of
central importance. The matrix has the same dimensions as the network to be inferred,
each entry corresponding to a confidence in the existence of the respective edge. This
confidence can be aquired by using external pathway sources, e.g. the KEGG database.
See the vignette for a description of how to get the prior matrices. No matter how
the confidence scores are obtained, there are two options, either use the laplaceinhib
prior type, in which knowledge about the type of the edges is present in the
external pathway source. Each confidence score for an edge, that is found as
inhibiting edge in the reference pathways, is multiplied by -1 to obtain a
negative value for the inhibiting edges. The entries of the prior matrix B
thus lie in the interval [-1;1], where -1 means strong confidence that
an edge is an inhibition, 1 means strong confidence that the edge is an activation
and 0 means that nothing is known about the presence or type of the edge.
If no information on the type of the edges is available in the external data source,
priortype="laplace" should be used, where the edge confidence ranges in the
interval [0;1], where 1 means strong confidence that the edge is present
and 0 means that nothing is known about the presence of the edge.
Argument gam
is used either as exponent in the scalefree prior, as it is described
in the reference, or in laplaceinhib
and laplace
as exponent in the following formula:
P(phi_{ij}|lambda,gam,B) = \frac{1}{2 \cdot lambda} exp(\frac{-|phi_{ij}-B_{ij}|^{gam}}{lambda})
It controls how strong the differences between an inferred edge and the probability
for seeing this edge in a reference set of networks are to be weighted. Defaults to 1,
if omitted. The prior curve rapidly decayes with increasing difference
|phi_{ij}-B_{ij}|, while for gam
larger than 1, the prior curve is
changed to an s-shaped curve with a plateau at the upper bound of
P(phi_{ij}|lambda,gam,B) and an exponential decay when a certain threshold of
|phi_{ij}-B_{ij}| is reached.
See also the references for a description of the priors.
The calcpr
function is a helper that calculates the laplace prior probability.
Returns a double for the prior probability of network structure phi
.
TODO
Christian Bender
Laplace prior
Froehlich et. al. 2007, Large scale statistical inference of signaling pathways from RNAi
and microarray data.
Scale free prior
Kamimura and Shimodaira, A Scale-free Prior over Graph Structures for Bayesian Inference of Gene Networks
1 2 3 | ##---- Should be DIRECTLY executable !! ----
##-- ==> Define data, use random,
##-- or do help(data=index) for the standard data sets.
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.