Functional Network Analysis

This repository contains the source code for an R package that generates weighted network models describing "functional profiles" found in a feature network. For gene networks the included Gephi plugin can be used to explore the models and related Gene Ontology annotations.

1. Build workflow (skip this if you don't want to rebuild from source)
2. Installation
3. Dependency notes
4. Example usage

Build workflow

Open up one R session and one terminal session.

In R,

install.packages(c("devtools", "roxygen2", "testthat")) # If necessary
library(devtools)
library(roxygen2)
document('fna/')

You might already have all the dependencies installed, but if not go to Dependency notes first. In the terminal session:

R CMD INSTALL fna
R CMD build fna
R CMD check fna_1.0.0.tar.gz

or just

./build.sh

Gephi plugin (optional). Clone and set up the plugin-development repository from https://github.com/gephi/gephi-plugins by following directions for 'create a plugin'. Use the following answers to the 13 questions that appear after running mvn org.gephi:gephi-maven-plugin:generate:

org.msk
fna
0.9.0
FunctionalNetworkAnalysis
Functional Network Analysis
Filter
Jimmy Mathews
<enter>
<enter>
Apache 2.0
Viewer for results of FNA analysis.
Displays based on topology and edge weights coming from FNA. May also show annotations.
no

Note that on some systems, due to some sort of bug with their maven plugin, the prompt questions indicated in the Gephi plugin development documentation may not appear. In this case you need to enter the answers 'blind', as shown above.

Copy the directory

functional-network-analysis/gephi_plugin_code/fna

(containing source files) from this repository into the new folder created after configuration,

gephi-plugins/modules/FunctionalNetworkAnalysis/src/main/java

Then copy

functional-network-analysis/gephi_plugin_code/pom.xml

into

gephi-plugins/modules/FunctionalNetworkAnalysis/

Finally, from the top level (gephi-plugins), run:

mvn clean package

For the above you need a Maven installation. Maven works on Windows and Unix/MacOS.

The .nbm file created can be loaded directly into Gephi as a new plugin. It is located at:

gephi-plugins/modules/FunctionalNetworkAnalysis/target/fna-0.9.0.nbm

To use it, you can select 'Feature Network Reduction' from the 'Topology' category of filters when you are viewing a graphml file which is output of the FNA pipeline. You may find that the performance is improved by increasing the memory available to Gephi.

Installation

Assuming you built/checked the R package, do the following:

install.packages("fna_1.0.0.tar.gz", repos=NULL, type="source")
library(fna)

help(generate_reduction) # To see usage of the main function, for example

Dependency notes

install.packages(c("igraph","emdist","mclust","pbmcapply"))

If you want support for the GCT file format for gene expression data, use:

install.packages('BiocManager')
BiocManager::install("CePa")

For the GO annotation functionality you will need

BiocManager::install("rols")

and you will also need the goa_human.gaf annotation file available from the EBI. goa_human.gaf is a somewhat large file (75mb), which is why it is not included here in this repository. For the most updated information, the script uses web API calls for the term definitions rather than a term definition file. However a partial cache file system is used to speed up repeated lookups.

Example Usage

After installation with install.packages("fna_1.0.0.tar.gz", repos=NULL, type="source"), open up an R session or RStudio and then run:

source('lung_gtex_run.R')

This example uses lung tissue RNA expression data from GTEx. The file example_data/lung_tissue_expression_gtex_abridged.csv is abridged to the 1000 genes with the most variance in the original dataset available from the GTEx portal, for illustration purposes. You should see output like the following:

Calculating weighted network reduction based on

node_data_file:             example_data/lung_tissue_expression_gtex_abridged.csv
topology_file:              NA
correlation_transformation: none
normalization:              none
method:                     gmt

[1/6] Loading data and loading/building network
      Inferring feature network from data using Pearson correlation. (No topology_file supplied).
      Data set 427x1000 all numeric.
      Calculated correlations (1000x1000)
      Cutoff value for correlation: 0.65
      Inferred 6963 edges, out of possible 499500, with connectivity 0.0139399399399399
      Support of inferred graph contains 734 nodes.
      Average degree 18.9727520435967.
      Diameter 12.

[2/6] Integrating sample set and network data
      Number of nodes in network not in the data set: 0
      Number of nodes in the data set not in the network: 266
      RSL24D1 NXN CASP7 NDUFS4 LILRB5 ...
      Number of nodes in common: 734
      6963 edges after integrating dataset and feature network.

[3/6] Fitting Gaussian mixture models
      Number of cores according to parallel::detectCores(): 8
      Trying 7.
  |=========================================================================| 100%, Elapsed 01:12
      6963 2D models with 3 populations.

[4/6] Comparing all adjacent models
  |=========================================================================| 100%, Elapsed 01:22
      305546 adjacent edge pairs considered.

[5/6] Averaging over intermediating triangles to get virtual edges with weights
      305546 of 305546 edge pairs collated.
      40198 total virtual edges.

DONE
4.624307 mins

479120 lines to scan.
20608 relevant lines found in goa_human.gaf out of 479122 total lines read. (Elapsed 2.92 mins of expected total 2.92 mins)

You can open up the output file example_data/lung_gtex_hierarchy_1000_annotated.graphml in Gephi.

Gephi hints:

1. Install the plugin NBM file into Gephi with Tools > Plugins > Downloaded > Add Plugin. Then find it under Filters > Topology > FNA Plot.
2. Use the Force Directed Layout 2.
3. Zoom out with the scroll wheel.
4. Set the node size to be a function of the absorption_time attribute.
5. Turn on labels coming from the name attribute.

Annotation. The Gene Ontology annotations are assessed for statistical significance as follows. The gene/term pairs listed in goa_human.gaf are filtered for the genes in the final hierarchy graph (approximately 750 genes in the example). Each term (annotation) that appears is thereby associated with a certain gene subset, certain leaf nodes in our weighted hierarchy tree (the edges of the hierarchy tree are weighted by a scaled level). For each such subset, the mean of the node-to-node pairwise weighted graph distances is calculated and regarded as a dispersal or coordination statistic. The statistical significance is measured with 10000-trial bootstrapping by random permutation of the gene set, with a p-value recording the fraction of the trials in which the statistic was lower than the observed value.