A zero-preprocessing enrichment network workbench for turning pathway enrichment results into interactive gene-pathway networks, pathway-overlap projections, diffusion-based candidate rankings, and reproducible export bundles.
The main workflow starts with uploading or loading enrichment data, selecting a preset, mapping columns, and building the gene ↔ pathway network.
The Stats tab summarizes graph size, connected components, top pathways/groups, top genes/items, and edge-weight information.
The Bipartite Diffusion tab ranks pathways and genes/items using random-walk diffusion over the main gene ↔ pathway graph.
The Projection tab builds a pathway ↔ pathway overlap network from shared genes/items.
Projection rankings identify central pathway themes in the pathway-only overlap graph.
Consensus candidates combine bipartite diffusion and projection diffusion signals into a final prioritized pathway list.
The sidebar includes full Network Studio report export and LLM Triage bundle export. The LLM Triage export is a safe handoff bundle and does not run a live public LLM call.
👉 Live App:
https://jc-enrichment-network-studio121-390287836436.northamerica-northeast1.run.app/
Most enrichment tools give you a ranked table of pathways.
JC Enrichment Network Studio helps answer the next question:
Which enriched biological themes are connected, supported by multiple genes, and worth following up?
The app lets users upload enrichment-style results, map or preset columns, build interactive networks, rank follow-up candidates, and export reproducible node/edge/result bundles.
It is designed as a focused enrichment-network workflow, not a general Cytoscape replacement.
- Upload enrichment results or load the demo dataset.
- Select a column mapping preset or manually map columns.
- Build the main gene ↔ pathway bipartite network.
- Build a pathway ↔ pathway projection network.
- Run diffusion ranking on the main network.
- Run diffusion ranking on the projection network.
- Build consensus candidate rankings.
- Export figures, CSVs, report bundles, or LLM Triage input bundles.
The app includes a demo workflow so users can immediately test the full pipeline without preparing a file first.
The demo loads enrichment-style data, expands gene lists when needed, maps columns, and prepares the user to build the network.
Supported input presets include:
- Custom long-format CSV
- Enrichr-style results
- g:Profiler / gprofiler2-style results
- clusterProfiler-style results
- GSEA / MSigDB-style long-format results
This reduces preprocessing friction and helps users move from enrichment output to network exploration faster.
Builds a main enrichment network where:
- genes/items connect to pathways/terms
- adjusted p-values can be transformed into
-log10(adjusted p-value)weights - users can filter by search term, degree, edge weight, max groups, and largest component
- Plotly provides interactive exploration
Runs random-walk / PageRank-style ranking over the main gene-pathway network.
Ranking modes:
- Balanced — combines enrichment evidence and connectivity
- Evidence-weighted — emphasizes strong adjusted p-values
- Connectivity-weighted — emphasizes network hubs and shared structure
Outputs include pathway rankings, gene rankings, and top follow-up candidates.
The candidate table combines:
- diffusion priority
- direct enrichment evidence
- support breadth / degree
These scores are intended for follow-up prioritization, not as new statistical p-values.
Builds a pathway-only graph where pathways are connected when they share genes/items.
Projection edge modes include:
- Jaccard similarity
- shared gene count
- weighted shared support
This helps reveal overlapping biological modules and pathway clusters.
Runs diffusion ranking on the pathway-only projection graph.
This identifies central biological processes based on pathway overlap structure rather than the original gene-pathway bipartite graph alone.
Combines signal from:
- main bipartite diffusion candidates
- projection diffusion candidates
The result is a consensus pathway ranking that highlights terms supported by both direct enrichment-network structure and pathway-overlap structure.
The app supports:
- PNG export through the Plotly UI
- SVG export
- PDF export
- main graph node/edge CSVs
- projection graph node/edge CSVs
- diffusion result CSVs
- candidate ranking CSVs
- consensus candidate CSVs
The app can export a full reproducible report bundle containing:
report_bundle/
run_summary.html
manifest.json
mapped_columns.json
settings.json
input_preview.csv
main_nodes.csv
main_edges.csv
graph_stats.csv
projection_nodes.csv
projection_edges.csv
bipartite_diffusion_results.csv
bipartite_top_candidates.csv
projection_diffusion_results.csv
projection_top_candidates.csv
consensus_candidates.csv
interpretation_notes.md
main_graph.svg
main_graph.pdf
projection_graph.svg
projection_graph.pdf
Static graph image exports may depend on the local/server Plotly-Kaleido environment. CSV outputs and summary files remain the primary reproducible artifacts.
🤖 Optional LLM Triage Export
The public Network Studio app does not run live LLM interpretation and does not spend API credits.
Instead, it provides a safe export-based connection to a companion LLM Triage workflow.
The app can export:
llm_triage_input/
run_summary.json
mapped_columns.json
settings_manifest.json
input_preview.csv
main_nodes.csv
main_edges.csv
projection_nodes.csv
projection_edges.csv
bipartite_diffusion_results.csv
bipartite_top_candidates.csv
projection_diffusion_results.csv
projection_top_candidates.csv
consensus_candidates.csv
interpretation_notes.md
This bundle can later be used with:
local LLM Triage workflows
bring-your-own-key setups
private deployments
paid hosted workflows
consulting/report-generation workflows
This keeps the public demo cost-controlled while still supporting AI-assisted interpretation as an optional companion workflow.
🧪 Input Formats
Custom Long-Format CSV
The simplest input is one row per gene-pathway membership edge:
| gene | term | adjusted_pvalue |
| ----- | -------- | --------------- |
| GeneA | Pathway1 | 0.01 |
| GeneB | Pathway1 | 0.01 |
| GeneB | Pathway2 | 0.04 |
The adjusted p-value column is optional but recommended.
Enrichr-Style CSV
Example:
| Term | Genes | Adjusted.P.value |
| -------------- | --------------- | ---------------- |
| DNA repair | BRCA1;RAD51;ATM | 0.003 |
| MAPK signaling | MAPK1;CREB1 | 0.012 |
The preset expands semicolon-separated genes into long-format edges.
g:Profiler / gprofiler2-Style CSV
Example columns may include
| name | p_value | intersections |
| ----------------------- | ------- | ------------------ |
| nucleosome organization | 0.002 | SUPT16H,H3-3A,HIRA |
The preset expands intersection genes into long-format edges.
GSEA / MSigDB-Style Long CSV
Example:
| term | gene | adjusted_pvalue |
| -------------------- | ------- | --------------- |
| HALLMARK_E2F_TARGETS | SUPT16H | 0.01 |
| HALLMARK_E2F_TARGETS | SUPT5H | 0.01 |
⚠️ Interpretation Caveats
Network Studio does not create new statistical significance values.
Use terms carefully:
Adjusted p-value: statistical enrichment evidence from the input table
Edge weight: transformed evidence, often -log10(adjusted p-value)
Diffusion score: network prioritization score, not a p-value
Follow-up score: practical triage score combining diffusion, evidence, and support breadth
Consensus score: agreement between main bipartite and pathway-projection rankings
Projection network: pathway-overlap graph, not a replacement for the main gene-pathway graph
LLM interpretation, if used through the companion workflow, should be treated as assistive and reviewed by a scientist.
📌 Use Cases
RNA-seq pathway interpretation
scRNA-seq enrichment exploration
ATAC-seq functional annotation follow-up
GWAS enrichment exploration
Functional genomics hypothesis generation
Comparing pathway overlap across enriched terms
Exporting clean node/edge tables for downstream tools like Cytoscape
🧠 Why This Tool?
Standard enrichment analysis answers:
What is statistically enriched?
Network Studio helps answer:
What is connected, biologically coherent, and worth following up?
It is especially useful when enrichment results contain many overlapping pathways and users need a fast way to visualize structure, identify hubs, and prioritize candidate biological themes.
🧱 Tech Stack
Python
Dash
Plotly
NetworkX
pandas
NumPy
SciPy
Docker
Google Cloud Run
⚙️ Local Setup
in bash
git clone https://github.com/jcaperella29/JC-Enrichment-Network-Studio.git
cd JC-Enrichment-Network-Studio/Network_Vis_App
pip install -r requirements.txt
python app.py
Then open:
http://localhost:8050
🐳 Docker Setup
in bash
docker build -t enrichment-network .
docker run -p 8050:8050 enrichment-network
then open same as before
🌐 Deployment
The public app is deployed on Google Cloud Run for containerized execution.
The public version is designed to be cost-controlled: it performs local graph/statistics/export workflows and does not expose an unrestricted hosted LLM interpretation endpoint.
🧭 Positioning
This is not a universal network science platform.
It is a focused enrichment-network studio for scientists who want to:
upload enrichment results
avoid manual Cytoscape preprocessing
build gene-pathway and pathway-overlap networks
rank follow-up biological themes
export reproducible figures and node/edge/result tables
optionally hand off results to a separate LLM Triage workflow