T2DM Urinary Metabolomics Analysis

A metabolomics analysis pipeline investigating urinary biomarkers of Type 2 Diabetes Mellitus (T2DM) using NMR spectroscopy data. This project applies epidemiological and machine learning methods to identify metabolites that are significantly altered in T2DM patients compared to healthy controls.

---

Table of Contents

• Background
• Dataset
• Methods
• Results
• Key Findings
• Requirements
• Visualisations

---

Background

Type 2 Diabetes Mellitus (T2DM) is a metabolic disorder affecting over 400 million people worldwide. Metabolomics — the large-scale study of small molecules in biological samples — offers a powerful approach to understanding the metabolic disruptions associated with the disease and identifying potential urinary biomarkers.

This project uses NMR-based urinary metabolomics to:

• Identify metabolites significantly different between T2DM patients and healthy controls
• Build a machine learning classifier to distinguish the two groups
• Discover robust biomarker candidates confirmed by multiple analytical methods

---

Dataset

• Source: MetaboLights MTBLS1
• Study: A metabolomic comparison of urinary changes in type 2 diabetes in mouse, rat, and human (Salek et al., 2007)
• Technology: NMR spectroscopy (Bruker DRX700, 700 MHz)
• Samples: 132 urine samples
  • 84 healthy controls
  • 48 T2DM patients (diet-controlled, unmedicated)
• Features: 220 NMR spectral bins (metabolite abundance values)

File	Description
s_MTBLS1.txt	Sample metadata (subject IDs, condition, gender)
m_MTBLS1_metabolite_profiling_NMR_spectroscopy_v2_maf.tsv	Metabolite abundance matrix

---

Methods

1. Preprocessing

• Transposed the metabolite matrix (samples × features)
• Log-transformation (log1p) to normalise right-skewed abundance values
• Auto-scaling (z-score standardisation) to place all features on equal footing

2. Exploratory Analysis — PCA

• Principal Component Analysis reduced 220 dimensions to 2
• Visualised group separation between T2DM and Control samples

3. ExWAS — Metabolome-Wide Association Study

• Univariate logistic regression for each of the 220 metabolite features
• Benjamini-Hochberg FDR correction to control for multiple testing
• Effect sizes calculated as mean difference between groups (scaled)

4. Random Forest Classification

• Ensemble classifier with 200 trees and balanced class weights
• 5-fold stratified cross-validation for honest performance evaluation
• Feature importances extracted to identify most predictive metabolites
• Cross-validated ROC curve for unbiased AUC estimation

5. Method Comparison

• Overlapping top features between ExWAS and Random Forest identified
• Dual-method agreement used to prioritise robust biomarker candidates

---

Results

Method	Result
PCA	Partial group separation visible along PC1 (16.8% variance)
ExWAS	87 / 220 metabolites significant after FDR correction (q < 0.05)
Random Forest	Cross-validated AUC = 0.985 ± 0.022
Method overlap	13 metabolites confirmed by both ExWAS and Random Forest

---

Key Findings

Metabolites Higher in T2DM

Metabolite	Biological Relevance
2-oxoisovalerate	Valine/leucine catabolism — altered amino acid metabolism
ethanol	Gut bacteria fermentation marker
unknown_m_8.335	Unidentified aromatic compound — warrants further investigation

Metabolites Lower in T2DM

Metabolite	Biological Relevance
isoleucine	Branched-chain amino acid — well-established T2DM marker
hippurate	Gut microbiome health marker — consistently reduced in T2DM
N-acetylglutamate	Urea cycle intermediate — reflects altered nitrogen metabolism

These findings are consistent with published T2DM metabolomics literature, supporting disruption of amino acid metabolism, gut microbiome function, and energy metabolism.

---

Requirements

pandas
numpy
matplotlib
seaborn
scikit-learn
statsmodels
scipy

Visualisations

PCA of Urinary NMR Profiles

Partial separation between T2DM (red) and Control (blue) samples along PC1.

Metabolites Confirmed by Both Methods

13 metabolites identified as significant by both ExWAS and Random Forest.