InferNetworks-methods: Network construction using Hierarchical Dirichlet Process

Description

This methods applies a Hierarchical Dirichlet Process (HDP) algorithm on the collection of proteins networks to infer the set of chromatin loop-maintainer proteins. HDP are non-parametric Bayesian models widely used in document classification as it enables us to model datasets with a mixtures of classes. In our case, we suppose that different kinds of networks are involved in maintaining the different loops. Thus, to make an analogy with topic modeling, we each DNA-interaction maintaining protein network as a document and each edge in this network as word. Thus, the task is to say which word (edge) belongs to which topic (chromatin-maintainer family). The method implementation is based on the C++ code of Chong Wang and David Blei with adaptation to Rcpp and removal of the dependency on the Gnu Scientific Library.

Usage

"InferNetworks"(object,thr =0.5,max_iter = 500L, max_time = 3600L, ...)

Arguments

object

a NetworkCollection object in which the list of protein interactions associated with each DNA interaction is already populated.

thr

Used to select the top protein interaction in each inferred chromatin-maintainer family. In HDP each topic (Chromatin-maintainer family) is considered as a distribution over words (edges), thus, for each topic we consider the words that capture threshold percent of the topic to be the top words. For example, in topic1, we first rank edges by partnership probability to topic1 in a decreasing order, and we take the top edges that capture 50% of the partnership.

max_iter

maximum number of iterations (befault 500).

max_time

maximum runing time (3600 sec).

...

not used for the moment, left for additional parameters.

Value

Returns a ChromMaintainers object that contains the list of inferred networks and the probability of each edge in each network.

References

https://www.cs.princeton.edu/~blei/topicmodeling.html (C. Wang's hdp code)

Chong Wang, John Paisley and David M. Blei, Online variational inference for the hierarchical Dirichlet process .In AISTATS 2011

Mohamed Nadhir D, Yang C et al 3CPET: Finding Co-factor Complexes in Chia-PET experiment using a Hierarchical Dirichlet Process, ....

Examples

Run this code

    ## get the different datasets path
    petFile <- file.path(system.file("example",package="R3CPET"),"HepG2_interactions.txt")  
    tfbsFile <- file.path(system.file("example",package="R3CPET"),"HepG2_TF.txt.gz")  
## Not run: 
#     x <- ChiapetExperimentData(pet = petFile, tfbs=  tfbsFile, IsBed = FALSE, ppiType="HPRD", filter= TRUE) 
#     ## build the different indexes
#     x <- createIndexes(x)
#     ## build networks connecting each interacting regions
#     nets<- buildNetworks(x)
# 
#     ## infer the networks
#     hlda<- InferNetworks(nets)
#     hlda
# ## End(Not run)