Identifying Depression Subtypes Using Gene Expression

A Multi-Phase Analytical Framework for Classification, Subtyping, and Systems-Level Discovery

Author: Vybhav Chaturvedi
Roll No.: 25BM6JP60

---

Project Overview

This repository contains a complete analytical workflow for understanding molecular heterogeneity in Major Depressive Disorder (MDD) using postmortem amygdala gene expression (GSE54564 dataset).

The project consists of three sequential phases, each implemented in a dedicated Jupyter Notebook:

1. Phase 1 – Baseline Supervised Classification
2. Phase 2 – Nested Validation & Advanced ML
3. Phase 3 – Unsupervised Subtype Discovery & Network Biology

The workflow demonstrates that binary classification fails due to diagnostic heterogeneity, revealing a biologically meaningful Neuro-Immune-Vascular subtype within MDD.

---

Repository Structure

C:.
│   01_Phase1_Baseline_Classification.ipynb
│   02_Phase2_Advanced_Classification.ipynb
│   03_Phase3_Unsupervised Analysis.ipynb
│   requirements.txt
│
├───data
│       GSE54564_non-normalized.txt
│       GSE54564_series_matrix.txt
│       NCBI_Depression.bgx
│
├───output
│   ├───Phase_1
│   ├───Phase_2
│   └───Phase_3
│
└───output/cytoscape_session

---

Installation

Requires Python 3.9+.

pip install -r requirements.txt

Or using conda:

conda create -n depression_env python=3.9
conda activate depression_env
pip install -r requirements.txt

---

How to Run the Project

1️⃣ Phase 1 — Baseline Classification

01_Phase1_Baseline_Classification.ipynb

• Loads raw data
• Performs preprocessing
• Trains 7 ML models
• Generates EDA + ROC + confusion matrices
• Saves outputs in output/Phase_1/

---

2️⃣ Phase 2 — Nested LOO-CV & Advanced ML

02_Phase2_Advanced_Classification.ipynb

• Implements Nested Leave-One-Out CV
• Evaluates 6 advanced pipelines
• Computes unbiased performance estimates
• Saves results to output/Phase_2/

---

3️⃣ Phase 3 — Unsupervised Discovery & Systems Biology

03_Phase3_Unsupervised Analysis.ipynb

• Consensus clustering (500 bootstraps)
• Subtype identification
• Cell-type deconvolution
• WGCNA network construction
• Cytoscape exports in output/cytoscape_session/

---

Reproducibility

Run all three notebooks in order.

---

Key Findings

• Classification fails (AUC ~ 0.49), proving diagnostic heterogeneity
• 4 molecular subtypes discovered
• Subtype S3 enriched for MDD (86%)
• Inhibitory neuron depletion (d = –0.77)
• Inflammatory module (M2) upregulated
• Neurovascular module (M8) downregulated
• MED8 emerges as a key regulatory hub

---

Contact

For replication or questions:
Vybhav Chaturvedi

• ISI Kolkata (25BM6JP60)
• IIT Kharagpur
• IIM Calcutta
• TIET