CrossEnrich

CrossEnrich compares enrichment outputs across pathway and ontology databases, then adds a standardization and semantic comparison layer so agreement is not limited to exact term-name matches.

Repo layout

• src/crossenrich/ Reusable Python package for standardization, semantic matching, clustering, and validation.
• notebooks/ Project notebooks. notebooks/CrossEnrich_v0.ipynb is the canonical workflow notebook.
• Benchmarks/ Benchmark datasets, weight-search code, and saved evaluation outputs.
• tests/ Unit tests for the reusable package.
• results/ Generated outputs such as matrices, heatmaps, and benchmark summaries.

Current implementation

The notebook notebooks/CrossEnrich_v0.ipynb covers:

• g:Profiler retrieval
• source filtering
• direct term-name Jaccard
• gene-level Jaccard
• Spearman rank correlation
• package-based semantic similarity, clustering, and validation

The package layers are:

• baseline.py Baseline database comparison metrics, including direct term overlap, gene-level overlap, and Spearman rank correlation.
• pipeline.py The package-level workflow wrapper that runs standardization, baseline metrics, semantic similarity, clustering, and validation in one consistent sequence.
• reporting.py User-facing summary tables for source pairs, top consensus clusters, and benchmark/run summaries.
• standardization.py Canonical source mapping, term normalization, optional manual term replacement, tokenization, gene-intersection parsing, parent-term parsing, and source-wise ranking.
• semantic.py Hybrid semantic similarity built from geometric token containment, gene Jaccard, trigram lexical similarity, and SPECTER embedding similarity, followed by clustering and a cluster-level consistency matrix. The current defaults are benchmark-selected from Benchmarks/weight_search_results.csv: token 0.35, gene 0.10, lexical 0.20, semantic 0.35, threshold 0.40.
• validation.py Score-matrix validation, baseline-vs-semantic comparison, and cluster quality summaries.
• visuals.py Plotting helpers for source-agreement heatmaps, source-pair ranking, top consensus cluster charts, and a cluster-level network view.
• network.py Cluster-level enrichment network construction based on shared supporting genes across multi-source clusters.

Suggested workflow

1. Start with a gene list and run enrichment in notebooks/CrossEnrich_v0.ipynb.
2. Filter the results to the selected databases and significant terms.
3. Use the baseline metrics to inspect exact-name, gene-level, and rank-level agreement.
4. Run run_crossenrich_pipeline(...) to build the baseline matrices, semantic similarity matrix, clustered terms, and cluster-consistency matrix with the benchmark-selected defaults. These defaults are the intended starting point for normal use, since they were selected from the benchmark evaluation rather than hand-picked for one example run.
5. Use reporting.py helpers to generate source-pair summaries, top consensus clusters, and run-level summary rows.
6. Use visuals.py helpers to generate polished heatmaps and cluster summary plots.
7. Optionally use the cluster-level network view to visualize how semantic clusters connect through shared genes.
8. Validate the outputs before exporting or reporting them.

Example

import pandas as pd

from crossenrich import (
    build_database_pair_summary,
    extract_top_consensus_clusters,
    run_crossenrich_pipeline,
    save_default_visuals,
)

results = pd.read_csv("your_gprofiler_results.csv")
outputs = run_crossenrich_pipeline(results)
pair_summary = build_database_pair_summary(outputs)
top_clusters = extract_top_consensus_clusters(outputs.clustered_terms, top_n=10)
saved_paths = save_default_visuals(outputs, "results")

CLI example:

crossenrich use-gmt notebooks/HALLMARK_OXIDATIVE_PHOSPHORYLATION.v2026.1.Hs.gmt --gene-set-name HALLMARK_OXIDATIVE_PHOSPHORYLATION
crossenrich all
crossenrich all-visuals
crossenrich run-summary
crossenrich pair-summary
crossenrich cluster-network
crossenrich semantic-similarity-plot
crossenrich selected-source-network --network-sources KEGG REAC
crossenrich all --output-dir custom_results --prefix oxphos_demo
crossenrich status
crossenrich clean-results
crossenrich clean-results --all
crossenrich clear

CLI Glossary

Setup and state

Command	What it does	Notes
crossenrich use-gmt <gmt_path>	Sets a GMT file as the active input, runs enrichment, runs the CrossEnrich pipeline once, and caches the outputs for later fast commands.	Use --gene-set-name when the GMT contains more than one set.
crossenrich use-results <results_file>	Sets an enrichment-results CSV/TSV as the active input, runs the CrossEnrich pipeline once, and caches the outputs.	Use when you already have g:Profiler results.
crossenrich status	Shows the currently active input configuration.	Includes input path, output directory, prefix, and gene set name if applicable.
crossenrich clear	Clears the active input state and cached outputs.	Use this before switching contexts if you want a clean reset.

Main generation commands

These commands use the active cached input created by use-gmt or use-results.

Command	What it saves	Notes
crossenrich all	The full default output bundle: standard summary CSVs and standard PNG visuals.	This is the normal “generate everything” command.
crossenrich all-visuals	All implemented visuals from visuals.py as PNGs.	Includes the semantic similarity plot.
crossenrich run-summary	Only the run summary CSV.	Compact one-row overview of the current run.
crossenrich pair-summary	Only the database-pair summary CSV.	One row per database pair.
crossenrich consensus-table	Only the top consensus clusters CSV.	Shared biological themes across sources.
crossenrich source-specific	Only the source-specific clusters CSV.	Source-unique biological themes.
crossenrich cluster-network-nodes	Only the overall cluster-network node summary CSV.	Table version of the overall network.
crossenrich selected-network-nodes --network-sources <A> <B> [...]	Only the selected-source network node summary CSV.	Requires --network-sources.
crossenrich clustered-terms	Only the clustered terms CSV.	Full clustered term membership table.
crossenrich cluster-consistency-matrix	Only the cluster-consistency matrix CSV.	Semantic cluster overlap across databases.
crossenrich term-jaccard-matrix	Only the direct term-overlap matrix CSV.	Exact term-name overlap.
crossenrich gene-jaccard-matrix	Only the gene-level Jaccard matrix CSV.	Gene-support overlap.
crossenrich spearman-matrix	Only the Spearman matrix CSV.	Rank correlation across sources.
crossenrich semantic-similarity-matrix	Only the semantic similarity matrix CSV.	Raw term-term semantic matrix as data.

Visual generation commands

Command	What it saves	Notes
crossenrich database-agreement-panels	*_database_agreement_panels.png	2x2 panel of direct overlap, gene Jaccard, Spearman, and semantic consistency.
crossenrich source-pair-ranking	*_source_pair_ranking.png	Ranked source-pair semantic agreement plot.
crossenrich consensus-plot	*_top_consensus_clusters.png	Top shared cluster/themes plot.
crossenrich cluster-network	*_cluster_network.png	Overall cluster-level network.
crossenrich semantic-similarity-plot	*_semantic_similarity_plot.png	Semantic similarity heatmap as a PNG.
crossenrich selected-source-network --network-sources <A> <B> [...]	*_selected_source_network.png	Focused pairwise or multi-source cluster network.

Output overrides

These work with generation commands such as all, cluster-network, or semantic-similarity-plot.

Option	What it does
--output-dir <dir>	Save outputs in a different folder instead of the active/default results/ folder.
--prefix <name>	Save outputs with a custom filename prefix instead of the active/default crossenrich.
--network-sources <A> <B> [...]	Required for selected-source network commands and selected network node summaries.

Cleanup

Command	What it does	Notes
crossenrich clean-results	Deletes generated files for the current active prefix in the active output directory.	Safe default cleanup.
crossenrich clean-results --prefix <name>	Deletes generated files for a specific prefix.	Useful when you used a custom prefix.
crossenrich clean-results --output-dir <dir>	Deletes generated files in a different output directory.	Targets a non-default folder.
crossenrich clean-results --all	Deletes everything inside the active output directory.	Use with care.

Run tests with:

python3 -m unittest discover -s tests -v