Patient Community Detection Pipeline

This repository provides a modular pipeline for detecting and analyzing patient communities from RDF-based clinical data. It includes tools for feature extraction, graph construction, community detection, consensus analysis, and visualization.

---

Requirements

• Python 3.8+
• rdflib
• pandas
• numpy
• scikit-learn
• networkx
• python-louvain (community)
• matplotlib

You can install the required packages via:

pip install rdflib pandas numpy scikit-learn networkx python-louvain matplotlib

---

Installation

1. Clone this repository:

   git clone https://github.com/your_org/patient-community-detection.git
   cd patient-community-detection
   

2. (Optional) Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate   # Linux/macOS
   .venv\Scripts\activate    # Windows
   

3. Install dependencies:

   pip install rdflib pandas numpy scikit-learn networkx python-louvain matplotlib
   

---

Usage

All scripts assume you have an RDF Turtle file containing patient data, e.g. patients.ttl.

1. community_detection_main.py

Runs the full pipeline (vectors → similarity graph → communities → modularity → save results):

python community_detection_main.py patients.ttl --threshold 0.5 --method louvain

• ttl_path: Path to .ttl file
• --threshold: Similarity threshold for graph edges. Edges between patients with lesser similarity values are ignored.
• --method: louvain

Outputs:

• Console logs for each step
• communities_<timestamp>.txt with community assignments

2. compare_patients.py

Compute pairwise similarity between two specified patients:

python compare_patients.py

Edit compare_patients.py to set:

pacienteA = "...#Case_1"
pacienteB = "...#Case_2"

and run:

python compare_patients.py

3. consensus_communities.py

Builds a co-occurrence matrix over multiple runs to assess stability:

python consensus_communities.py data/patients.ttl --threshold 0.5 --method louvain --runs 20 --output consensus.csv

• --runs: Number of iterations (default: 20)
• --output: Output CSV path

Generates a CSV where each cell (i,j) counts the number of times two patients co-occur in the same community.

4. community/visualization.py

Visualize communities as a network graph:

from community.visualization import visualize_communities
visualize_communities(G, partition, output_path='coms.png', show_edges=False)

• show_edges: Toggle edge drawing for clarity
• output_path: File path for PNG output

---

Module Overview

feature_builder/

• config.py: Namespace URIs for RDF parsing
• vectorizer.py: build_feature_vectors(ttl_path) → extracts patient features and vectorizes into a NumPy matrix
• graph_builder.py: build_similarity_graph(feature_vectors, patient_ids, threshold) → builds a weighted NetworkX graph based on cosine similarity

community/

• community_detector.py: detect_communities(G, method) → Louvain method; evaluate_modularity(G, partition) → computes modularity
• visualization.py: visualize_communities(G, partition, output_path, show_edges, figsize) → renders and saves a graph image

Script	Description
community_detection_main.py	Full pipeline: vectorize → graph → detect → evaluate → save txt
compare_patients.py	Compute and print cosine similarity between two patients
consensus_communities.py	Run multiple community detections and export co-occurrence CSV

---