ApplicationwithTBdataset_TianW.RmdnetZooR is an R package which consists of five main algorithms and is able to construct, analyse and plot gene regulatory networks.
PANDA(Passing Attributes between Networks for Data Assimilation) is a message-passing model to gene regulatory network reconstruction. It integrates multiple sources of biological data, including protein-protein interaction, gene expression, and transcription factor binding motifs information to reconstruct genome-wide, condition-specific regulatory networks.[Glass et al. 2013]
LIONESS(Linear Interpolation to Obtain Network Estimates for Single Samples) is a method to estimate sample-specific regulatory networks by applying linear interpolation to the predictions made by existing aggregate network inference approaches.[LIONESS arxiv paper]
CONDOR (COmplex Network Description Of Regulators) implements methods to cluster biapartite networks and estimatiing the contribution of each node to its community’s modularity.[Platig et al. 2016]
ALPACA(ALtered Partitions Across Community Architectures) is a method to compare two genome-scale networks derived from different phenotypic states to identify condition-specific modules. [Padi and Quackenbush 2018]
SAMBAR(Subtyping Agglomerated Mutations By Annotation Relations) is a method to identify subtypes based on somatic mutation data.[Kuijjer et al.].
Using this pacakage requires Python (3.X) and some Python libraries, R (>= 3.3.3), and stable Internet access.
Some plotting functions will require the Cytoscape installed.
How to install Python libraries depends varies from different platforms. More instructions could be find here.
The following Python libraries (or packages) are required by running PANDA and LIONESS algorithms:
The required Python packages are: pandas, numpy, networkx, matplotlib.pyplot.
This package could be downloaded via install_github() function from devtools package.
# install.packages("devtools")
library(devtools)
# install netZooR pkg with vignettes, otherwise remove the "build_vignettes = TRUE" argument.
devtools::install_github("netZoo/netZooR", build_vignettes = TRUE)Here is some PANDA-ready transcription factor binding motifs data by species.
This package includes a function source.PPI may source a Protein-Protein Interactions (PPI) througt STRING database given a list of proteins of interest. The STRINGdb is already loaded while loading netZooR.
# TF is a data frame with single column filled with TFs of Mycobacterium tuberculosis H37Rv.
# PPI <- source.PPI(TF, species.index=83332, score_threshold=0)Use search() to check all loaded package currently.
library(netZooR)
# search()Accessing the help pages for the usage of core functions.
?panda.py
?create.condor.object
?panda.to.condor.object
?lioness.py
?alpaca
?panda.to.alpaca
?sambarThis package will invoke the Python in R environment through reticulate package. Configure which version of Python to use if necessary, here in netZooR, Python 3.X is required. More details can be found here
#check your Python configuration and the specific version of Python in use currently
py_config()
# reset to Python 3.X if necessary, like below:
# use_python("/usr/local/bin/python3")Use example datasets within package to test this package. Refer to four input datasets files: one TB expression dataset control group , one TB expression dataset treated, one transcription factor binding motifs dataset, and one protein-protein interaction datasets in inst/extdat.
# retrieve the file path of these files
treated_expression_file_path <- system.file("extdata", "expr4.txt", package = "netZooR", mustWork = TRUE)
control_expression_file_path <- system.file("extdata", "expr10.txt", package = "netZooR", mustWork = TRUE)
motif_file_path <- system.file("extdata", "chip.txt", package = "netZooR", mustWork = TRUE)
ppi_file_path <- system.file("extdata", "ppi.txt", package = "netZooR", mustWork = TRUE)Assign the above file paths to flag e(refers to “expression dataset”), m(refers to “motif dataset”), and ppi(refers to “PPI” dataset), respectively. Then set option rm_missing to TRUE to run PANDA to generate an aggregate network without unmatched TF and genes.
Repeat with control group.
treated_all_panda_result <- panda.py(expr = treated_expression_file_path, motif = motif_file_path, ppi = ppi_file_path,mode_process="legacy", rm_missing = TRUE )
control_all_panda_result <- panda.py(expr = control_expression_file_path, motif = motif_file_path, ppi = ppi_file_path,mode_process="legacy", rm_missing = TRUE )Vector treated_all_panda_result and vector control_all_panda_result below are large lists with three elements: the entire PANDA network, indegree (“to” nodes) nodes and score, outdegree (“from” nodes) nodes and score. Use $panda,$indegree and $outdegree to access each list item resepctively.
Use $pandato access the entire PANDA network.
treated_net <- treated_all_panda_result$panda
control_net <- control_all_panda_result$pandaCytoscape is an interactivity network visualization tool highly recommanded to explore the PANDA network. Before using this function plot.panda.in.cytoscape, please install and launch Cytoscape (3.6.1 or greater) and keep it running whenever using.
How to run LIONESS is mostly idential with method how to run PANDA in this package, unless the return values of runLioness is a data frame where first two columns represent TFs (regulators) and Genes (targets) while the rest columns represent each sample. each cell filled with estimated score calculated by LIONESS.
# Run LIONESS algorithm for the first two samples
# removing start_sample and end_sample to generate whole LIONESS network with all samples.
control_lioness_result <- lioness.py(expr = control_expression_file_path, motif = motif_file_path, ppi = ppi_file_path, rm_missing = TRUE, start_sample=1, end_sample=2)PANDA network can simply be converted into condor.object by panda.to.condor.object(panda.net, threshold) Defaults option threshold is the average of [median weight of non-prior edges] and [median weight of prior edges], all weights mentioned previous are transformationed with formula w'=ln(e^w+1) before calculating the median and average. But all the edges selected will remain the orginal weights calculated by PANDA.
treated_condor_object <- panda.to.condor.object(treated_net, threshold = 0)The communities structure can be plotted by igraph.
treated_color_num <- max(treated_condor_object$red.memb$com)
treated_color <- viridis(treated_color_num, alpha = 1, begin = 0, end = 1, direction = 1, option = "D")
condor.plot.communities(treated_condor_object, color_list=treated_color, point.size=0.04, xlab="Target", ylab="Regulator")ALPACA community structure can also be generated from two PANDA network by panda.to.alpaca
alpaca<- panda.to.alpaca(treated_net, control_net, "./TB", verbose=TRUE)Browse with browseVignettes("netZooR")
If there is an error like Error in fetch(key) : lazy-load database.rdb' is corrupt when accessing the help pages of functions in this package after being loaded. It’s a limitation of base R and has not been solved yet. Restart R session and re-load this package will help.