Dashboard for the Inference and Analysis of Networks from Expression data

DIANE is a shiny application for the analysis of high throughput gene expression data (RNA-Seq). Its function is to extract important regulatory pathways involved in the response to environmental changes, or any perturbation inducing genomic modifications.

We designed this tool to process, explore, and perform advanced statistical analysis on multifactorial expression data using state of the art methods. It includes :

Raw count data pre-processing and sample-wise normalization
Customizable differential expression analysis
Gene ontology enrichment analysis for model organisms
Expression based clustering in the framework of Poisson Mixture Models, and characterization of those clusters with generalized linear models and GO enrichment analysis
Machine learning based Gene regulatory network inference

All of the features in DIANE are accessible via a single page shiny application that can be locally launched, or accessed online at https://diane.ipsim.inrae.fr{.uri}.

The steps should be performed in the order of the different tabs. For instance, before running clustering or network inference, differential expression should be performed first. The figure above summarizes DIANE’s main workflow.

For more advanced users, all server-side functions in DIANE are exported so they can be called from R scripts.

Fore more information, please find full documentation and examples in the github page https://oceanecsn.github.io/DIANE.

Once the application is launched, if the resolution poorly fits your screen, you can adjust it with the keyboard shortcuts ctrl + or ctrl - (use cmd on Mac).

Please report any bug or suggestion via github or at oceane.cassan@cnrs.fr.

To cite DIANE in publications use:

Cassan, O., Lèbre, S. & Martin, A. Inferring and analyzing gene regulatory networks from multi-factorial expression data: a complete and interactive suite. BMC Genomics 22, 387 (2021). https://doi.org/10.1186/s12864-021-07659-2

A BibTeX entry for LaTeX users is

@Article{cassan2021Inferring,
    title = {Inferring and analyzing gene regulatory networks from multi-factorial expression data: a complete and interactive suite},
    author = {Océane Cassan and Sophie Lèbre and Antoine Martin},
    journal = {BMC Genomics},
    year = {2021},
    volume = {22},
    number = {387},
    url = {https://bmcgenomics.biomedcentral.com/articles/10.1186/s12864-021-07659-2}}

Use DIANE locally

To use DIANE locally, download and install DIANE in your R console as follows (you need the remotes package installed) :

remotes::install_github("OceaneCsn/DIANE")

DIANE relies on R 4.0.0, available for all OS at https://cloud.r-project.org/.

You can then launch the application :

library(DIANE)
DIANE::run_app()

License

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

Deploy DIANE on your server

We provide a solution based on Docker and Shiny server to deploy DIANE on any linux server, just as it is at https://diane.bpmp.inrae.fr. To do so, see the following command line instructions.

Get DIANE source code via Git :

git clone https://github.com/OceaneCsn/DIANE.git

Install Docker engine, as described in the Docker docs.

Go to DIANE’s folder.

First, you can change the default settings for the dockerized shiny-server by editing the file shiny-customized.config (like changing the port, the user to run with, and more)

Now let’s build the image, that we’ll name diane, from the Dockerfile (superuser rights required).

cd DIANE
docker build -t diane .

This might take a while. You can check that the container image was built with docker images. Then, you can start the container diane, by setting appropriately the following options in the above command:

/path/to/app/on/host/ is the path to DIANE on the host, that is to say the location where you cloned it, containing the app.R file. /path/to/logs/on/host/ is the folder you want to store your app logs.

-p 8086:8086 is the port to use, change the first 8086 to use another one on the host.

--user shiny allows to run as non root, with the shiny user that must have been created before, and granted rights to the folder /path/to/app/on/host/logs and /path/to/logs/on/host/.

-d --rm are options for the detached mode.

Plus, in the following example, a session of DIANE will be allowed to use 16 CPU cores :

docker run -d --cpus 16 --user shiny --rm -p 8086:8086 -v /path/to/app/on/host/:/srv/shiny-server/ -v /path/to/logs/on/host/:/var/log/shiny-server/ diane

You can check that the container is running with docker ps.

Authors : Océane Cassan, Antoine Martin, Sophie Lèbre

Dev : Océane Cassan, PhD Student at IPSIM (Institute for Plant Sciences in Montpellier) research unit, SUPAGRO Montpellier, with contributions from Alexandre Soriano.

Upload expression data and experimental design

Expression file upload

Separator :

Comma Semicolon Tab

Choose CSV/TXT expression file

Expected expression file

You should feed DIANE the raw counts that were obtained after the bioinformatic pipeline of mapping and quantification of your reads. It gives, for each gene, the transcript aboundance found in each of your experimental condition and repliactes. The needed matrix-like file shoud contain a column named “Gene”, containing all your gene IDs. The other columns should be your sample names, noted as follow : conditionName_replicate (see example below). Expression values should be zeros or positive integers.

In order for your input to be compatible with the proposed organisms, it should contain gene IDs as follow :
- For Arabidopsis thaliana : TAIR IDs (ex: AT1G01020, or AT1G01020.1)
- For Human : ensembl IDs (ex: ENSG00000005513)
Here is the head of our companion expression file, from 24 Arabidopsis thaliana’s samples (8 conditions with triplicates) :
```
Gene,C_1,C_2,C_3,S_1,S_2,S_3,M_1,M_2,M_3,H_1,H_2,H_3,SM_1,SM_2,SM_3,SH_1,SH_2,SH_3,MH_1,MH_2,MH_3,SMH_1,SMH_2,SMH_3
AT1G01010.1,127.0,67.9,65.5,94.0,88.1,95.9,65.1,100.3,126.8,95.4,135.0,117.2,96.7,104.4,98.1,94.7,96.1,101.3,82.8,107.4,97.1,100.1,96.7,121.8
AT1G01020.1,207.9,220.8,186.8,192.5,225.1,197.8,234.2,196.9,179.4,312.9,366.0,318.0,169.0,179.6,186.5,340.8,352.6,345.0,331.2,315.8,327.5,267.7,313.7,319.3
AT1G01030.1,32.7,34.4,55.8,33.6,15.9,31.3,21.4,29.8,33.5,47.7,39.0,51.1,25.2,31.2,34.4,61.0,65.5,61.2,45.0,36.2,55.9,46.0,42.7,56.8
```

Browse...

Separator :

Tab

Choose CSV/TXT gene information file (optional)

Gene information file requirements

This file gives additional information about the genes in your analysis.

It must contain a column named “Gene”. This column should contain the genes IDs present in the row names of the expression file, or the locus corresponding to your splice variants. Genes can be duplicated, in that case the final annotation fields will be a concatenation of the common annotations.

The annotation fields must be either “label”, “description”, or both. The column “label” is typically used for common gene names. “description” is usually a brief description of the known roles/pathways of the gene.

The summary tables containing genes will contain those additional columns.

Here is an example of a tab separated file that would work :
```
label   description Gene
HTT2	target of trans acting-siR480/255 protein	AT5G18040
HRS1	histidyl-tRNA synthetase	AT3G46100
HIPP20	Heavy metal transport/detoxification superfamily protein	AT1G71050
```

Browse...

Preview of the expression matrix

This might help you visualize the general aspect of the data and different sequencing depths of your conditions.

Design and gene information files

Separator :

Comma Semicolon Tab

Expected design file

You can also feed DIANE the experimental design corresponding to your data.

Indeed, DIANE can anlysis any combinatorial design and give valuable insights about the different factors that are studied.

To do so, you can provide, for each condition name, the level of the factors corresponding to that condition in your study.

Let’s take our companion data as an example. Here, the experiment we chose for our demo included 3 factors, heat stress, mannitol stress, as well as salinity stress.

You should first identify which level of each factor can be considered as the control level, and which is a perturbation. In our demo, the control condition is referred to as C. Then, the perturbation are specified by their corresponding letter, from simple to triple stress combination. For example, SH corresponds to salt and heat stresses in the control level of mannitol. As a consequence, its levels would be 1,0,1.

The design file is thus a matrix with condition names as rows, and factor names as columns. It MUST contain a column named “Condition”, as you can see in our example below :
```
Condition,Salt_stress,Mannitol_stress,Heat_stress
C,0,0,0
H,0,0,1
S,1,0,0
M,0,1,0
SM,1,1,0
SH,1,0,1
MH,0,1,1
SMH,1,1,1
```
Many RNA-Seq anlysis only study one factor, which would not be a problem to use DIANE. If you have two conditions named control and trt, the design would contain only one factor column:
```
Condition,treatment
contr,0
trt,1
```

Choose CSV/TXT design file (optional)

Browse...

Describe the levels of each factors for your conditions

Data filtering and normalisation

Settings

Normalization

Low counts filtering

Minimal gene count sum accross conditions :

Samples distributions
Summary
Data exploration

Explore normalized gene expression

PCA
Visualize gene expression levels
Explore normalized data

Principal Component Analysis

Performing PCA on normalized RNA-Seq counts can be really informative about the conditions that impact gene expression the most. During PCA, new variables are computed, as linear combinations of your initial variables (e.g. experimental conditions). Those new variables, also called principal components, are designed to carry the maximum of the data variability.

We can plot the correlations of the initial variables to the new variables, the principal components, to see which ones contribute the most to the overall expression changes.

Each principal component explains a certain amount of the total variability, and those relative percentages are shown in what is called the screeplot.

Genes and conditions choice

‌

Expression based clustering

Settings

Mixture models clustering

Min number of clusters :

Max number of clusters :

Clustering quality descriptors

The coseq package tests a range of different clusters in order to give the best fit to the data.

For each number of cluster, the ICL (Integrated Completed Likelihood) is computed. It combines two elements :

The global likelihood of the clustering. It quantifies how accurate the clustering seems, regarding the posterior probability of each element to belong the its predicted cluster. It can be computed using the Poisson probability densities resulting from the proposed clustering, for all the genes.
The number of clusters. As the likelihood tends to grow monotonously with the number of clusters, resulting in a very big number of groups, that would not be very informative for the user. Thus, the ICL penalizes the clustering quality criteria with the number of clusters.

This is why the maximal value of ICL can be interpreted as an approximation of the ideal number of clusters.

The last line informs you of the number of clusters chosen to maximise the ICL, the clustering quality criteria.

Analyse the genes of a specific cluster

Expression profiles

GO Background

The absolute value of a coefficient gives information about the intensity of its effect on gene expression. The highest coefficient(s) thus are the one(s) driving the profiles in a specific cluster. The genes in this cluster are potentially involved in the response to that factor. The sign of a coefficient gives information about the way it impacts expression. If it is positive it increases the expression when the facot is in its perturbation level. If negative, it decreases it.

Network inference

Inference Settings

GENIE3 Gene regulatory network inference

Or choose your own :

Expected regulators list

You can use this file upload if there is no list provided for your organism, or you would like to provide yout own list.

Those regulators should be in the same format as the gene identifiers in the rownames of your expression file. In addition, your expression file should contain at least a portion of those regulators.

Please use a file containing one regultaors per line.

Here is the head of a file that would be accepted :
```
AT1G04880
AT1G20910
AT1G55650
AT1G76110
AT1G76510
AT1G76510
AT2G17410
```

Upload custom CSV/TXT regulators list

Browse...

Recommended : group regulators correlated over ( %)

Number of trees for GENIE3 Random Forests :

Thresholding settings

Without thresholding, we would obtain a fully connected weighted graph from GENIE3, with far too many links to be interpretable. In order build a meaningfull network, this weighted adjacency matrix betwen regulators and targets has to be sparsified, and we have to determine the N higher regulatory weights that we consider significant.

As biological networks are known for their pronounced sparsity, considering all possible regulator-target pairs would be of very little interest. Studies found that the typical values of density (ratio between the number of edges, and the number of edges of the complete graph) in state of the art biological networks lie approximately between 0.1 and 0.001. We therefore propose to create a first graph, topologically consistent with biological network standards, which will be further refined by statistical testing. During the testing phase, the null ditributions of regulatory weights are estimated by the permutations, thus providing an empricial p-value for the observed importance.

Network's conectivity density

Network analysis

Click on any node (gene) to get its description and network information

The super nodes (dark green squares) of the network are detailed here. Each edge represents a correlation above the specified threshold between the regulators. The community detection in this graph was used to group highly correlated variables before network inference with GENIE3.

Topolocical clusters correspond to the network structural communitities.

The expression profiles of genes within a community can be positively correlated, but also, unlike with expression based clustering, negatively correlated.

GO Background

Legal mentions

For DIANE’s online version <https://diane.bpmp.inrae.fr>{.uri}

Website publication

Webmaster and content publisher : Océane CASSAN

Website hosted at : Campus INRAE / SupAgro – IBIP – UMR BPMP

Bât. 7 – 2 place Pierre Viala

34060 Montpellier cedex 2

Affiliations

DIANE was developped with the following academic organisations support:

INRAE

Établissement public à caractère scientifique et technologique

147, rue de l’université

75338 Paris cedex

07 01 42 75 90 00

CNRS

3, rue Michel-Ange

75794 Paris cedex 16 - France

+33 1 44 96 40 00

BPMP and Agro Institute

Campus INRAE / SupAgro – IBIP – UMR BPMP

Bât. 7 – 2 place Pierre Viala

34060 Montpellier cedex 2

University of Montpellier

163 rue Auguste Broussonnet

34090 Montpellier

Data privacy and cookies

All data submitted to DIANE for analysis will remain confidential as it is not stored on the server side.

DIANE doesn’t collect any private information on users. It only makes use of technical cookies as well as anonymous usage data.

License

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

Version changes in DIANE

1.0.1 : DIANE is released

Version at the moment of the official BMC publication (May 26, 2021).

1.0.2 : Minor bug fixes and improvements.

Improvements :

Citation was added
Gene IDs can now have blank space round them in the exploration of specific genes profiles
MDS was removed
PCA can now handle expression data with 4 or less experimental conditions
This document was added to the interface in a new tab

Bug fixes :

Custom annotation/gene information file now allows duplicated genes
Custom annotation/gene information now can only have columns named label and/or description
DEA csv download does not require annotations any more

1.0.3 : Minor bug fixes and improvements.

Fixed error in DAINE’s programming interface (wrong value of the defaut “conditions”” argument in normalize function) 22/11/2021
Used namesapce operator in data import module to speed up loading

1.0.4 : Minor changes, mainly aesthetic, with no impact one DIANE’s main functions and backend. Changes are from merging pull request #36 from Alexandre-So.

Some code based on bootstrap grid system has been rewrite. This avoid the grey (anthracite) background to show up at the bottom of some pages, and improve shinydashboardPlus::box alignment between different dashboard tabs. This also increases a little bit the width of most of the shinydashboardPlus::box.
Numbers in the DEA table show less digits when displayed in DIANE (up to 10-3). The downloaded table is not affected.
When a row is clicked in the DEA table, display a popup with RNAseq count associated with the related gene. (Very cool feature, worth the highlight)
The little loading square position is now based on user screen resolution. The position can still be improved. Used CSS code is in app_ui.R.
The little help buttons in the import dataset page have been moved to new locations. This can be done for other tabs.
Add “y axis” scrollbar to some tables which were exceeded screen boundaries in low resolutions.
Update contact email to new cnrs one.

1.0.5 : Minor changes and update caused by breaking changes in dependency

rfPerumte was updated and code had to be change to comply with new function names
The vignette was improved (mention of seed, and rendering adjustements)

1.0.6 Fixes

Changed url to diane.ipsim.inrae.fr added license to welcome pages
Fixed cutom gene info file upload in the case of splicing aware gene IDs in expression file
Fixed not displaying go results for custom org in network communities
In draw expression levels, genes filtered because of low expression can still be viewed

January 8, 2024 - Fix

Fix a bug with normalization : Normalization was using total count instead of the normalization factors computed by the chosen method.

Dashboard for the Inference and Analysis of Networks from Expression data

Use DIANE locally

License

Deploy DIANE on your server

Upload expression data and experimental design

Expression file upload

Expected expression file

Gene information file requirements

Seed setting for reproducible analyses

Preview of the expression matrix

Design and gene information files

Expected design file

Data filtering and normalisation

Settings

Normalization

Normalisation methods

Low counts filtering

Removing low count genes

Minimal gene count sum accross conditions :

Explore normalized gene expression

Principal Component Analysis

Genes and conditions choice

‌

Differential expression analysis

Settings

Estimation of disperion

Differential expression methods

Conditions to compare for differential analysis :

Once more than one differential expression analysis were performed, you can visualise and compare the different genes lists in a Venn diagram.

Expression based clustering

Settings

Mixture models clustering

Mixture models for clustering

Clustering quality descriptors

The last line informs you of the number of clusters chosen to maximise the ICL, the clustering quality criteria.

Analyse the genes of a specific cluster

Expression profiles

Generalized Poisson regression on a cluster of genes

Network inference

Inference Settings

GENIE3 Gene regulatory network inference

GENIE3 network inference

Or choose your own :

Expected regulators list

Variable grouping in network inference

Thresholding settings

Regulatory links thresholding

Network analysis

Click on any node (gene) to get its description and network information

The super nodes (dark green squares) of the network are detailed here. Each edge represents a correlation above the specified threshold between the regulators. The community detection in this graph was used to group highly correlated variables before network inference with GENIE3.

Topolocical clusters correspond to the network structural communitities.

The expression profiles of genes within a community can be positively correlated, but also, unlike with expression based clustering, negatively correlated.

Legal mentions

Website publication

Affiliations

Data privacy and cookies

License

Version changes in DIANE

1.0.1 : DIANE is released

1.0.2 : Minor bug fixes and improvements.

1.0.3 : Minor bug fixes and improvements.

1.0.4 : Minor changes, mainly aesthetic, with no impact one DIANE’s main functions and backend. Changes are from merging pull request #36 from Alexandre-So.

1.0.5 : Minor changes and update caused by breaking changes in dependency

1.0.6 Fixes

January 8, 2024 - Fix