ddepn-package: Dynamic Deterministic Effects Propagation Networks: Infer...

In ddepn: Dynamic Deterministic Effects Propagation Networks: Infer signalling networks for timecourse RPPA data.

Description

Uses high throughput array data (e.g. from Reverse Phase Protein arrays or mRNA microarrays) to infer regulatory or signalling relationships among the measured proteins after different stimuli/inhibitions. Each stimulus/inhibition is included as a separate node and edges to the measured proteins are inferred. The signal flow through the network is modelled in a boolean fashion and a set of reachable system states is created. Afterwards, the optimal path through the possible system states is searched for using an HMM. Estimation of Gaussian distributions for active and passive states for each node depending on the activity state of the protein is done and a likelihood based score is calculated for a network. Network structure search is performed in either a genetic algorithm that optimises a population of candidate networks or via a markov chain monte carlo sampling that summarises the distribution over the network structures via sampling.

Details

Package:	ddepn
Type:	Package
License:	GPL
LazyLoad:	yes

Author(s)

Christian Bender

Maintainer: Christian Bender <[e-mail]>

References

Bender et. al. 2010: Dynamic deterministic effects propagation networks: learning signalling pathways from longitudinal protein array data; Bioinformatics, Vol. 26(18), pp. i596-i602

See Also

ddepn, netga
Visit also the project page on R-forge: http://ddepn.r-forge.r-project.org/.

Examples

## Not run: 
## load package
library(ddepn)

## sample a network
n <- 6
signet <- signalnetwork(n=n, nstim=2, cstim=0, prop.inh=0.2)
phit <- signet$phi
stimuli <- signet$stimuli

## sample data
dataset <- makedata(phit, stimuli, mu.bg=1200, sd.bg=400, mu.signal.a=2000, sd.signal.a=1000)

## use original network as prior matrix
## reset all entries for inhibiting edges 
## to -1
B <- phit
B[B==2] <- -1

## Genetic algorithm, using BIC score as optimisation criterion
ret1 <- ddepn(dataset$datx, phiorig=phit,inference="netga",
              maxiterations=30, p=15, q=0.3, m=0.8,
              usebics=TRUE)	
x11()
plotdetailed(ret1$phi,stimuli=stimuli,fontsize=20)

## Genetic algorithm, using a uniform prior
ret2 <- ddepn(dataset$datx, phiorig=phit, inference="netga",
    maxiterations=15, p=15, q=0.3, m=0.8,
    usebics=FALSE, priortype="uniform")
x11()
plotdetailed(ret2$phi,stimuli=ret2$stimuli,fontsize=20)

## Genetic algorithm, using a laplaceinhib prior and the posterior
## probabilities as optimisation criterion
ret2 <- ddepn(dataset$datx, phiorig=phit, inference="netga",
              maxiterations=30, p=15, q=0.3, m=0.8,
              usebics=FALSE, lambda=0.01, B=B, priortype="laplaceinhib") 
x11()
plotdetailed(ret2$phi,stimuli=ret2$stimuli,fontsize=20)

## MCMC sampling using a uniform prior
ret3 <- ddepn(dataset$datx,phiorig=phit, inference="mcmc",
    maxiterations=300, burnin=100,
    usebics=FALSE, priortype="uniform")

## MCMC sampling using a laplaceinhib prior
ret3 <- ddepn(dataset$datx,phiorig=phit, inference="mcmc",
              maxiterations=300, burnin=100,
              usebics=FALSE, lambda=0.01, B=B, priortype="laplaceinhib")
              
x11()
plotdetailed(ret3$samplings[[1]]$phi,stimuli=ret3$samplings[[1]]$stimuli)

## MCMC sampling using a scale free prior
ret4 <- ddepn(dataset$datx,phiorig=phit, inference="mcmc",
              maxiterations=300, burnin=100,
              usebics=FALSE, gam = 2.2, it = 500, K = 0.8, priortype="scalefree")
x11()
plotdetailed(ret4$samplings[[1]]$phi,stimuli=ret4$samplings[[1]]$stimuli)



## End(Not run)

ddepn documentation built on May 2, 2019, 4:42 p.m.