Genomic Region Set & (Ranked) Gene Set Enrichment Analysis & Visualization Snakemake Workflow for Human and Mouse Genomes.

Given human (hg19 or hg38) or mouse (mm9 or mm10) based genomic region sets (i.e., region sets) and/or (ranked) gene sets of interest and respective background region/gene sets, the enrichment within the configured databases is determined using LOLA, GREAT, GSEApy (over-representation analysis (ORA) & preranked GSEA), pycisTarget, RcisTarget and results saved as CSV files. Additionally, the most significant results are plotted for each region/gene set, database queried, and analysis performed. Finally, the results within the same "group" (e.g., stemming from the same analysis) are aggregated per database and analysis in summary CSV files and visualized using hierarchically clustered heatmaps and bubble plots. For collaboration, communication and documentation of results, methods and workflow information a detailed self-contained HTML report can be generated.

This workflow adheres to the module specifications of MR.PARETO, an effort to augment research by modularizing (biomedical) data science. For more details and modules check out the project's repository. Please consider starring and sharing modules that are interesting or useful to you, this helps others to find and benefit from the effort and me to prioritize my efforts!

If you use this workflow in a publication, please don't forget to give credits to the authors by citing it using this DOI 10.5281/zenodo.7810621.

Table of contents

• Authors
• Software
• Methods
• Features
• Usage
• Configuration
• Examples
• Links
• Resources
• Publications

Authors

• Stephan Reichl
• Daria Romanovskaia
• Christoph Bock

Software

This project wouldn't be possible without the following software and their dependencies:

Software	Reference (DOI)
Enrichr	https://doi.org/10.1002/cpz1.90
ggplot2	https://ggplot2.tidyverse.org/
GREAT	https://doi.org/10.1371/journal.pcbi.1010378
GSEA	https://doi.org/10.1073/pnas.0506580102
GSEApy	https://doi.org/10.1093/bioinformatics/btac757
LOLA	https://doi.org/10.1093/bioinformatics/btv612
pandas	https://doi.org/10.5281/zenodo.3509134
pheatmap	https://cran.r-project.org/package=pheatmap
pycisTarget	https://doi.org/10.1038/s41592-023-01938-4
RcisTarget	https://doi.org/10.1038/nmeth.4463
rGREAT	https://doi.org/10.1093/bioinformatics/btac745
Snakemake	https://doi.org/10.12688/f1000research.29032.2

Methods

This is a template for the Methods section of a scientific publication and is intended to serve as a starting point. Only retain paragraphs relevant to your analysis. References [ref] to the respective publications are curated in the software table above. Versions (ver) have to be read out from the respective conda environment specifications (workflow/envs/\*.yaml files) or post execution ({results_dir}/envs/enrichment_analysis/\*.yaml files). Parameters that have to be adapted depending on the data or workflow configurations are denoted in squared brackets e.g. [X].

The outlined analyses were performed using the programming languages R (ver) [ref] and Python (ver) [ref] unless stated otherwise. All approaches statistically correct their results using expressed/accessible background genomic region/gene sets from the respective analyses that yielded the query region/gene sets.

Genomic region set enrichment analyses

LOLA. Genomic region set enrichment analysis was performed using LOLA (ver) [ref], which uses Fisher’s exact test. The following databases were queried [lola_databases].

GREAT. Genomic region set enrichment analysis was performed using GREAT [ref] implemented with rGREAT (ver) [ref]. The following databases were queried [local_databases].

pycisTarget. Genomic region set TFBS (Transcription Factor Binding Site) motif enrichment analysis was performed using pycisTarget (ver) [ref]. The following databases were queried [pycisTarget_databases].

Furthermore, genomic regions (query- and background-sets) were mapped to genes using GREAT (without background) and then analyzed as gene sets as described below for a complementary and extended perspective.

Gene set enrichment analyses (GSEA)

Over-representation analysis (ORA). Gene set ORA was performed using Enrichr [ref], which uses Fisher’s exact test (i.e., hypergeometric test), implemented with GSEApy's (ver) [ref] function enrich. The following databases were queried [local_databases].

Preranked GSEA. Preranked GSEA was performed using GSEA [ref], implemented with GSEApy's (ver) [ref] function prerank. The following databases were queried [local_databases].

RcisTarget. Gene set TFBS (Transcription Factor Binding Site) motif enrichment analysis was performed using RcisTarget (ver) [ref]. The following databases were queried [RcisTarget_databases].

Aggregation The results of all queries belonging to the same analysis [group] were aggregated by method and database. Additionally, we filtered the results by retaining only the union of terms that were statistically significant (i.e. [adj_pvalue]<=[adjp_th]) in at least one query.

Visualization All analysis results were visualized in the same way.

For each query, method and database combination an enrichment dot plot was used to visualize the most important results. The top [top_n] terms were ranked (along the y-axis) by the mean rank of statistical significance ([p_value]), effect-size ([effect_size]), and overlap ([overlap]) with the goal to make the results more balanced and interpretable. The significance (adjusted p-value) is denoted by the dot color, effect-size by the x-axis position, and overlap by the dot size.

The aggregated results per analysis [group], method and database combination were visualized using hierarchically clustered heatmaps and bubble plots. The union of the top [top_terms_n] most significant terms per query were determined and their effect-size and significance were visualized as hierarchically clustered heatmaps, and statistical significance ([adj_pvalue] < [adjp_th]) was denoted by *. Furthermore, a hierarchically clustered bubble plot encoding both effect-size (color) and statistical significance (size) is provided, with statistical significance denoted by *. All summary visualizations’ values were capped by [adjp_cap]/[or_cap]/[nes_cap] to avoid shifts in the coloring scheme caused by outliers.

The analysis and visualizations described here were performed using a publicly available Snakemake (ver) [ref] workflow [10.5281/zenodo.7810621].

Features

The five tools LOLA, GREAT, pycisTarget, RcisTarget and GSEApy (over-representation analysis (ORA) & preranked GSEA) are used for various enrichment analyses. Databases to be queried can be configured (see ./config/config.yaml). All approaches statistically correct their results using the provided background region/gene sets.

• enrichment analysis methods
  • region set (\*.bed)
    • LOLA: Genomic Locus Overlap Enrichment Analysis is run locally using configured databases (lola_databases) taken from LOLA Region Databases or custom created using these instructions
    • GREAT using rGREAT: Genomic Regions Enrichment of Annotations Tool runs locally using configured databases (local_databases), additional resources are downloaded automatically during the analysis.
    • pycisTarget: Motif enrichment analysis in region sets to identify high confidence transcription factor (TF) cistromes is run locally using configured databases (pycistarget_parameters:databases) from the cisTarget resources.
  • gene set (\*.txt) over-representation analysis (ORA_GSEApy)
    • GSEApy enrich() function performs Fisher’s exact test (i.e., hypergeoemtric test) and is run locally using configured databases (local_databases).
    • RcisTarget: Motif enrichment analysis in gene sets to identify high confidence transcription factor (TF) cistromes is run locally using configured databases (Rcistarget_parameters:databases) from the cisTarget resources.
  • region-based gene set (\*.bed) over-representation analysis (ORA_GSEApy) & TFBS motif enrichment analysis (RcisTarget)
    • region-gene associations for each query and background region set are obtained using (r)GREAT, without accounting for background for improved performance and more genes. Correction for background is anyway included in the gene-based analyses downstream.
    • they are used for a complementary ORA using GSEApy and TFBS motif enrichment analysis using RcisTarget.
    • thereby an additional enrichment perspective for region sets can be gained through association to genes by querying the same and/or more databases, that are not supported/provided by region-based tools.
  • preranked gene set (\*.csv) enrichment analysis (preranked_GSEApy)
    • GSEApy prerank() function performs preranked GSEA and is run locally using configured databases (local_databases).
    • Note: only entries with the largest absolute score are kept and +/- infinity values are set to max/min, respectively.
• databases have to be provided by the user
  • databases (local_databases) for rGREAT and GSEApy
    • local GMT (\*.gmt) files e.g., from MSigDB or Enrichr
    • local (custom) JSON (\*.json) files e.g., { "MyDB_Term1": ["geneA","geneB","geneC"],"MyDB_Term2": ["geneX","geneY","geneZ"]}
    • always use gene symbols e.g., STAT1
    • the local databases are (converted,) copied and saved as GMT files in /resources.
  • LOLA databases for LOLA
    • downloaded from LOLA Region Databases
    • custom databases created using these [instructions]
    • pre-cached databases as .RData files are supported by simpleCache
  • cisTarget databases for pycisTarget and RcisTarget
    • downloaded from the cisTarget resources
    • custom databases using these instructions
• group aggregation of results per method and database
  • results of all queries belonging to the same group are aggregated per method (e.g., ORA_GSEApy) and database (e.g., GO_Biological_Process_2021) by concatenation and saved as a long-format table (CSV).
  • a filtered version taking the union of all statistically significant (i.e., adjusted p-value <{adjp_th}) terms per query is also saved as a long-format table (CSV).
• visualization
  • region/gene set specific enrichment dot plots are generated for each query, method and database combination
    • the top {top_n} terms are ranked (along the y-axis) by the mean rank of statistical significance ({p_value}), effect-size ({efect_size} e.g., log2(odds ratio) or normalized enrichemnt scores), and overlap ({overlap} e.g., coverage or support) with the goal to make the ranking more balanced and interpretable
    • significance (adjusted p-value) is presented by the dot color
    • effect-size is presented by the x-axis position
    • overlap is presented by the dot size
  • group summary/overview
    • the union of the top {top_terms_n} most significant terms per query, method, and database within a group is determined.
    • their effect-size (effect) and statistical significance (adjp) are visualized as hierarchically clustered heatmaps, with statistical significance denoted by \* (PDF).
    • a hierarchically clustered bubble plot encoding both effect-size (color) and significance (size) is provided, with statistical significance denoted by \* (PNG).
    • all summary visualizations are configured to cap the values ({adjp_cap}/{or_cap}/{nes_cap}) to avoid shifts in the coloring scheme caused by outliers.
• results ({result_path}/enrichment_analysis)
  • the result directory contains a folder for each region/gene set {query} and {group}
  • {query}/{method}/{database}/ containing:
    • result table (CSV): {query}\_{database}.csv
    • enrichment dot plot (PNG): {query}\_{database}.{png}
  • {group}/{method}/{database}/ containing
    • aggregated result table (CSV): {group}\_{database}\_all.csv
    • filtered aggregated result table (CSV): {group}\_{database}\_sig.csv
    • hierarchically clustered heatmaps visualizing statistical significance and effect-sizes of the top {top_terms_n} terms (PDF): {group}\_{database}\_{adjp|effect}\_heatmap.pdf
    • hierarchically clustered bubble plot visualizing statistical significance and effect-sizes simultaneously (PNG): {group}\_{database}\_summary.{png}

Note:

• Despite usage of the correct parameter, rGREAT was not using the provided cores during testing. Nevertheless, it is still provided as parameter.
• (r)GREAT performs two statistical test (binomial and hypergeometric), results of both are reported and should be considered. Which results are used for the visualization can be configured column_names:GREAT.
• pycisTarget for region set enrichment does not use the provided background regions. In case this is desired (e.g., conensus regions or TF ChIP-seq data) follow the instructions for custom cisTarget databases using your own regions and use them as database

Usage

Here are some tips for the usage of this workflow:

1. Download all relevant databases (see Resources).
2. Configure the analysis using the configuration YAML and an annotation file (see Configuration)
3. Run the analysis on every query gene/region set of interest (e.g., results of differential analyses) with the respective background genes/regions (e.g., all expressed genes or consensus regions).
4. generate the Snakemake Report
5. look through the overview plots of your dedicated groups and queried databases in the report
6. dig deeper by looking at the
   • aggregated result table underlying the summary/overview plot
   • enrichment plots for the individual query sets
7. investigate interesting hits further by looking into the individual query result tables.

Configuration

Detailed specifications can be found here ./config/README.md

Examples

We provide four example queries across all tools with four different databases:

• three are region sets from a LOLA Vignette. Download the example data by following the instructions below.
• one is a preranked gene-score set derived from the GDS289 fgsea R package example data (score=-log10(p-value) * sign(LFC)).
• the total runtime was ~45 minutes on an HPC with 1 core and 32GB RAM per job.
• note: we are using a hg38 database for pycistarget, because the respective hg19 database is not compatible with the current version.

Follow these steps to run the complete analysis:

1. Download all necessary data (query and resources)

   # change working directory to the cloned worklfow/module enrichment_analysis
   cd enrichment_analysis
   
   # download and extract the region set test data
   wget -c http://cloud.databio.org.s3.amazonaws.com/vignettes/lola_vignette_data_150505.tgz -O - | tar -xz -C test/data/
   
   # create and enter resources folder
   mkdir ../resources
   cd ../resources
   
   # download LOLACore databases and move to the correct location
   wget http://big.databio.org/regiondb/LOLACoreCaches_180412.tgz
   tar -xzvf LOLACoreCaches_180412.tgz
   mv nm/t1/resources/regions/LOLACore/ .
   rm -rf nm
   
   # download a local database
   wget https://data.broadinstitute.org/gsea-msigdb/msigdb/release/2023.2.Hs/c2.cgp.v2023.2.Hs.symbols.gmt
   
   # download cisTarget resources
   mkdir cistarget
   wget https://resources.aertslab.org/cistarget/databases/homo_sapiens/hg38/refseq_r80/mc_v10_clust/gene_based/hg38_500bp_up_100bp_down_full_tx_v10_clust.genes_vs_motifs.rankings.feather
   wget https://resources.aertslab.org/cistarget/databases/homo_sapiens/hg38/screen/mc_v10_clust/region_based/hg38_screen_v10_clust.regions_vs_motifs.rankings.feather
   wget https://resources.aertslab.org/cistarget/motif2tf/motifs-v10nr_clust-nr.hgnc-m0.001-o0.0.tbl
   
   # change your working directory back to the root of the module
   cd ../

2. activate your conda Snakemake environment, run a dry-run (-n flag), run the workflow and generate the report using the provided configuration

   conda activate snakemake
   # dry-run
   snakemake -p --use-conda --configfile test/config/example_enrichment_analysis_config.yaml -n
   # real run
   snakemake -p --use-conda --configfile test/config/example_enrichment_analysis_config.yaml
   # report
   snakemake --report test/report.html --configfile test/config/example_enrichment_analysis_config.yaml

Links

• GitHub Repository
• GitHub Page
• Zenodo Repository
• Snakemake Workflow Catalog Entry

Resources

• Recommended compatible MR.PARETO modules
  • for upstream processing and analyses:
    • ATAC-seq Processing to quantify chromatin accessibility.
    • scRNA-seq Data Processing & Visualization for processing (multimodal) single-cell trascnriptome data.
    • Split, Filter, Normalize and Integrate Sequencing Data to process sequencing data.
    • Differential Analysis with limma to identify and visualize statistically significant features between bulk sample groups.
    • Differential Analysis with Seurat to identify and visualize statistically significant features between single-cell groups.
• Package for simplifying enrichment results using the ComplexHeatmap package.
• Web versions of some of the used tools
  • LOLA
  • GREAT
  • Enrichr
• Databases & resources for region/gene sets
  • LOLA region databases
    • custom LOLA database instructions
  • cisTarget resources
    • custom cistarget database instructions
  • Enrichr gene set databases
  • The Molecular Signatures Database (MSigDB)

Publications

The following publications successfully used this module for their analyses.

• ...