Peptide Unsupervised Clustering with UMAP and HDBSCAN

This repository implements an unsupervised learning pipeline for clustering peptide and protein sequence data using UMAP for dimensionality reduction and HDBSCAN for density-based clustering.

The goal is to discover latent structure in high-dimensional biological sequence data without requiring predefined labels or cluster counts.

---

Project Overview

Biological sequence data is inherently high-dimensional and difficult to cluster directly.
This project addresses that challenge by:

• Converting peptide sequences into numerical feature representations
• Reducing dimensionality with UMAP to preserve local and global structure
• Applying HDBSCAN to identify dense clusters while labeling noise robustly

This approach is well-suited for exploratory analysis in bioinformatics and computational biology.

---

Pipeline

1. Data Loading

   • Import peptide/protein sequences from FASTA files

2. Feature Engineering

   • Convert sequences into numerical representations suitable for ML

3. Dimensionality Reduction

   • Apply UMAP to project high-dimensional features into 2D/3D space

4. Clustering

   • Use HDBSCAN to identify clusters of varying density
   • Automatically detect noise and outliers

5. Visualization & Analysis

   • Visualize embeddings and cluster assignments
   • Interpret biological structure and separation

---

Tech Stack

• Python
• UMAP
• HDBSCAN
• NumPy, Pandas
• Matplotlib
• Jupyter Notebook

---