LeiDAN: A Go Implementation of Leiden Clustering and Weighted KNN for Drug Association Networks

This repository provides a pure Go implementation of the Leiden community detection algorithm together with a weighted K-nearest neighbors (KNN) graph builder designed for large-scale transcriptomic data. The full pipeline has been applied to LINCS L1000 Level 5 signatures to reproduce the Drug Association Networks (DANs) described in:

Musa et al., “A Drug-Centric View of Drug Development: Drug Association Networks,” Scientific Reports (2019).
https://www.nature.com/articles/s41598-019-44291-3

The project includes both the high-performance backend for constructing and clustering similarity graphs and an interactive RShiny dashboard that visualizes community evolution, enrichment, and metadata structure across Leiden levels.

Group members:

1. Jeff Winchell
2. Ajay Prabhakar
3. Nethan Ramachandran

Link to Video Demo

🚀 Features

Algorithmic + Graph Features

• Pure Go Leiden clustering

  • Fast local moving
  • Refinement step
  • Multi-level aggregation
  • Modularity-based optimization

• Weighted KNN graph construction

  • Cosine-similarity–based
  • Efficient for large, sparse data
  • Integrates naturally with CSR structure

• Custom CSR Graph

  • Memory-efficient adjacency matrix
  • Fast neighbor iteration
  • Suitable for large LINCS-scale graphs

Data-Driven Features

• Reconstruction of Drug Association Networks (DANs)
• Integration of ATC, MoA, and TAS metadata
• Discovery of functional drug similarity beyond predefined labels

---

📊 Interactive RShiny Dashboard

This project includes a fully implemented RShiny dashboard for visual exploration of the clustering results.
Key capabilities:

• Community label evolution viewer

  • Track how individual nodes move across Leiden levels
  • Visualize stability and transitions between communities

• TAS, ATC, MoA, and metadata enrichment panels

  • Automatically computes enrichment for final communities
  • Provides interpretable summaries and visualizations

• Multiple quality functions

  • Compare Reichardt Bornholdt Potts Model (RBPM) vs vanilla modularity
  • Explore different resolution (γ) settings

• Optimization utilities

  • Visualize how clustering quality varies across γ
  • Aid in selecting resolution parameters

---

📁 Repository Structure

├── code       <- go code
├── dashboard  <- rshiny dashboard
├── data       <- put your data here
└── notebooks  <- python notebook used to preprocess data

How To Run Go Code

cd leiden

go build  

./leiden dataset prefix outputDir/ gridSearch k(optional) gamma(optional)

• dataset: input dataset in .csv format (use data/mcf7_subset_pca.csv)
• prefix: tag you want for graphs and images
• outputDir: filepath that the algorithm results go to
• gridSearch: true or false, if user wants to find optimal hyperparameters
• k: # of nearest neighbors for KNN
• gamma: resolution hyperparameter of Leiden algorithm

How To Run R Shiny Dashboard

In R terminal:

setwd("<projectdir>/dashboard/")

install.packages('renv'); renv::restore() 

shiny::runApp("<projectdir>/dashboard")