GRAND: a database of gene regulatory network models across human conditions

Marouen Ben Guebila, Camila Miranda Lopes-Ramos, Deborah A. Weighill, Abhijeet Rajendra Sonawane, Rebekka Burkholz, Behrouz Shamsaei, John Platig, Kimberly Glass, Marieke L. Kuijjer, John QuackenbushNucleic Acids Res 2022 · DOI

TL;DR

This paper introduces GRAND, a comprehensive database of gene regulatory network (GRN) models that integrate diverse biological data. GRAND allows researchers to explore and compare GRNs across various human conditions, tissues, and experimental perturbations, facilitating the discovery of disease mechanisms and potential therapeutic targets.

What Problem Does This Paper Address?

Understanding how genes are regulated is fundamental to comprehending cellular identity, function, and response to disease or treatment. This regulation occurs through complex networks involving transcription factors (TFs) and microRNAs (miRNAs). While many methods exist to infer these gene regulatory networks (GRNs), a significant challenge has been the lack of a centralized, accessible repository for these computationally derived models. Existing GRN databases often have limitations in terms of scale, interactivity, or the ability to compare networks across different biological contexts. This makes it difficult for researchers to leverage these powerful models for broad discovery, to assess reproducibility, or to perform large-scale analyses. The paper addresses this by creating GRAND, a database designed to host and facilitate the analysis of thousands of genome-scale GRN models.

Methods

The GRAND database is built upon a suite of GRN inference tools collectively known as the "Network Zoo." Key methods include:

  • PANDA: A foundational method that infers GRNs by integrating TF binding motifs, protein-protein interaction data, and gene co-expression relationships.
  • PUMA: Extends PANDA to incorporate miRNAs as regulators.
  • LIONESS: A novel approach that infers sample-specific GRNs, allowing for the analysis of individual biological samples rather than just aggregate networks. This is crucial for capturing phenotypic variation.
  • OTTER: Uses a graph matching approach to build multi-omic networks.
  • DRAGON: Employs Gaussian Graphical Models to construct multi-omic networks.

The database construction involved collecting existing GRN models and generating new ones using these tools from various public datasets. The backend of the GRAND website is built using Django, with network visualization powered by Vis.js.

Datasets

GRAND integrates data from several major sources to build its GRN models:

  • GTEx: Used to construct TF and miRNA GRNs for 36 "normal" human tissues.
  • TCGA: Provided data for inferring GRNs for 28 cancer types.
  • Connectivity Map (CMAP): Contributed gene expression data from human cell lines treated with 2858 approved and investigational drugs, used to generate GRNs for small molecule screening assays.
  • Cancer Cell Line Encyclopedia (CCLE): Provided gene expression and miRNA data for over 1000 cell lines, used to build both aggregate and sample-specific GRNs.

The database itself hosts over 12,468 GRNs and 173,013 TF and gene targeting scores.

Key Insights and Ideas

  • The Power of Sample-Specific Networks: The development and inclusion of LIONESS highlight the critical need for sample-specific GRNs. Aggregate networks can mask important biological differences between individuals, and sample-specific models are essential for understanding disease heterogeneity and personalized medicine.
  • Bridging the Gap Between In Silico and In Vivo: By integrating GRNs with phenotypic information and drug perturbation data, GRAND aims to bridge the gap between computational predictions and experimental validation, facilitating drug repurposing and the identification of therapeutic strategies.
  • A Centralized Resource for Reproducibility and Discovery: The creation of GRAND addresses a major bottleneck in GRN research by providing a single, accessible platform for thousands of GRN models, promoting reproducibility and enabling new avenues of research that were previously hindered by data accessibility and scale.
  • Network Properties as Biomarkers and Therapeutic Predictors: The paper demonstrates how network properties, such as TF targeting scores, can be used to identify potential drug candidates that can reverse disease-associated regulatory changes.

Key Results

  • GRAND contains 12,468 genome-scale GRN models covering 36 human tissues, 28 cancers, and 1378 unperturbed cell lines.
  • It also includes 173,013 TF and gene targeting scores derived from networks of cell lines treated with 2858 small molecule compounds.
  • The database offers interactive visualization tools and analysis modules, including CLUEreg for drug repurposing and TF enrichment analysis.
  • An example analysis demonstrated how GRAND could identify potential therapeutic compounds (e.g., MK-5108) for colon cancer by comparing disease-specific GRNs to normal tissue GRNs and analyzing TF targeting patterns.

Limitations and Future Work

  • Scalability of Data Storage: While GRAND hosts a significant amount of data, the authors acknowledge that the sheer size of GRN data (over 6TB) required a tailored design approach, and future growth will necessitate continued infrastructure development.
  • Continued Expansion of Network Types and Contexts: The authors plan to expand GRAND with more gene regulatory network models from an increasing number of biological contexts, including additional cancer types and networks inferred from single-cell expression data.
  • Development of Additional Analytical Tools: Future work will focus on incorporating more analytical tools and features based on user feedback to enhance the utility of the database.
  • Reproducibility of Inference: While GRAND provides access to models, the paper emphasizes the importance of making the inference code available (which they do via GitHub) to ensure reproducibility.

Summary updated

BibTeX
@article{guebila2022grand,
  author = {Marouen Ben Guebila and Camila Miranda Lopes-Ramos and Deborah A. Weighill and Abhijeet Rajendra Sonawane and Rebekka Burkholz and Behrouz Shamsaei and John Platig and Kimberly Glass and Marieke L. Kuijjer and John Quackenbush},
  title = {{GRAND: a database of gene regulatory network models across human conditions}},
  journal = {Nucleic Acids Res.},
  year = {2022},
  doi = {10.1093/nar/gkab778},
}
Figure