How costalab/scaccordion works
scACCorDiON is a Python toolkit for comparing and clustering directed, weighted cell–cell communication graphs with optimal transport. The main workflow takes per-sample interaction tables into Accordion, which builds graph representations, derives graph-based costs, and computes pairwise Wasserstein distances; those distances can then be evaluated against sample labels. The package also exposes k-barycenter clustering for interpretable cluster representatives, graph-distance implementations, and bundled PDAC/example-data loaders. The source excerpts show Accordion’s internal calls; README claims about broader analysis and barycenter capabilities are documented, but their complete orchestration is not shown.
Main entry points: Accordion.py Accordion · datasets.py Dataset Loaders · utils.py Graph Matrices · distances.py Graph Costs
Components 10
Graph Analysis
- Accordion
scaccordion/tools/Accordion.py - Graph Matrices
scaccordion/tools/utils.py
Distance Methods
- Graph Costs
scaccordion/tools/distances.py - Directed GWOT
scaccordion/tools/GWOT.py - Graph OT
scaccordion/tools/GOT.py
Clustering
- K Barycenters
scaccordion/tools/KBarycenters.py - Label Evaluation
scaccordion/tools/utils.py
Data and Access
- Dataset Loaders
scaccordion/tools/datasets.py - Bundled Data
scaccordion/data
Other components
- Analyst
Connections 9
- Analyst → Accordion: provides tables
scaccordion/tools/Accordion.py - Accordion → Graph Matrices: builds matrices
scaccordion/tools/Accordion.py - Accordion → Graph Costs: computes costs
scaccordion/tools/Accordion.py - Accordion → Label Evaluation: evaluates distances
scaccordion/tools/Accordion.py - Analyst → K Barycenters: provides inputs
README.md - Analyst → Directed GWOT: compares graphs
- Analyst → Graph OT: compares graphs
- Analyst → Dataset Loaders: requests data
scaccordion/tools/datasets.py - Dataset Loaders → Bundled Data: reads bundled data
scaccordion/tools/datasets.py