Navigating gene set hierarchies with multi-resolution maps
🌐 Web Application · 📄 Cite · ⚙️ Getting Started
Gene set enrichment analysis translates differential expression into biological meaning, yet every conventional tool collapses the Gene Ontology's hierarchical structure into flat ranked lists. MondrianMap restores that hierarchy. It organizes enrichment results into 13 semantically principled layers derived from the GOALS framework and renders them as color-encoded rectangular maps where:
- Block area encodes statistical significance (−log₁₀ adjusted p-value)
- Color encodes effect direction (red = upregulated, blue = downregulated)
- Spatial proximity preserves semantic relatedness (GoBERT + UMAP)
- Layer navigation enables multi-resolution traversal from molecular mechanism to system-level theme
MondrianMap integrates three NIH Common Fund Data Ecosystem (CFDE) databases — LINCS L1000, GTEx Aging Signatures, and MoTrPAC — as pre-indexed, searchable repositories, alongside support for custom gene list upload.
Figure 1 — CFDE datasets or user-uploaded gene lists undergo enrichment analysis (Fisher's exact test, FDR ≤ 0.05). Enriched GO-BP terms are assigned to 13 GOALS semantic layers and embedded via GoBERT + UMAP for spatial layout. The resulting Mondrian maps encode significance (area), direction (color), and semantic proximity (position) within a navigable, layer-resolved interface.
Figure 2 — Interactive interface. (a) Gene set input (custom or CFDE case studies). (b) Layer-resolved Mondrian map canvas with zoomable drill-down. (c) Enrichment results panel with term statistics. (d–e) Dynamic filters for gene count, significance, and Jaccard crosstalk thresholds. (f–g) AI hypothesis generation module producing layer-grounded biological narratives.
Figure 3 — Layer-resolved Mondrian maps discriminate cancer driver mechanisms. (a–b) TP53 and KRAS knockouts at GOALS Layer 8: the same immune recruitment processes (neutrophil chemotaxis, granulocyte chemotaxis) appear as blue blocks in TP53 and red blocks in KRAS — a directional inversion visible at a glance. (c–e) Multi-resolution navigation within TP53 across Layers 4, 6, and 13 reveals three progressively broader biological narratives from a single enrichment. (f–g) Two independent TP53 replicates at Layer 7 confirm visual reproducibility.
Two additional case studies — GTEx tissue-aging signatures and MoTrPAC exercise temporal dynamics — are presented in the accompanying manuscript.
No installation required. Visit mondrianmap.smartdrugdiscovery.org to analyze CFDE databases or upload custom gene lists.
Frontend
git clone https://github.com/aimed-lab/mondrian-web.git
cd mondrian-web
npm install
npm run devPython Pipeline
cd python
pip install -r requirements.txt
python process_pipeline.py --helppython python/ingest_database.py <path_to_gmt> \
--id <db_id> \
--name <display_name> \
--label-type <category_label> \
--description <description>| Flag | Description |
|---|---|
--id |
Short database identifier (e.g., LINCS) |
--name |
Display name in the web application dropdown |
--label-type |
Category label (e.g., Drug Perturbation) |
--shard-size |
Target shard size in MB (default: 5) |
--single-dir |
Treat all entries as unidirectional (no Up/Down pairing) |
If you use MondrianMap in your research, please cite:
MondrianMap: Hierarchical Enrichment Visualization for Multi-Resolution Biological Discovery Al Abir, F., Yue, Z., Saghapour, E., Hossain, M.D., Sembay, Z., Zhang, S., & Chen, J.Y. (2026). (Under Review)
Mondrian Abstraction and Language Model Embeddings for Differential Pathway Analysis Al Abir, F. & Chen, J.Y. (2024). IEEE International Conference on Bioinformatics and Biomedicine (BIBM). DOI: 10.1101/2024.04.11.589093
GOALS: Gene Ontology Analysis with Layered Shells for Enhanced Functional Insight and Visualization Yue, Z., Welner, R.S., Willey, C.D., Amin, R., Li, Q., Chen, H., & Chen, J.Y. (2025). DOI: 10.1101/2025.04.22.650095
Fuad Al Abir, Zongliang Yue, Ehsan Saghapour, Md Delower Hossain, Zhandos Sembay, Sixue Zhang, Jake Y. Chen
Correspondence: jakechen@uab.edu
MondrianMap is open-source. Source code is available at github.com/aimed-lab/mondrian-web.