🧬 Multivariate Analysis of Obesity-Related Gene Expression

Master in Bioinformatics – R statistics project Group 8 Date: January 2026

---

📌 Project Overview

This repository contains the complete multivariate statistical analysis performed for Activity 3 of the Master in Bioinformatics. The objective of this work is to explore patterns of gene expression associated with obesity-related phenotypes using unsupervised learning techniques.

The analysis integrates:

• Data preprocessing and quality control
• Normalization assessment
• Principal Component Analysis (PCA)
• Clustering of genes and individuals
• Visualization of gene contributions
• Biological interpretation in relation to BMI (IMC)

All analyses were performed in R, following reproducible and transparent data-science practices.

---

🧠 Biological Context

The dataset contains expression values of genes previously associated with:

• Energy balance
• Appetite regulation
• Hypothalamic signaling
• Obesity risk (e.g. LEP, LEPR, MC4R, POMC, FTO, ADCY3, PCSK1)

In addition, anthropometric variables such as BMI (IMC) were used to explore how molecular patterns relate to phenotypic stratification.

---

📂 Repository Structure

.
├── data/
│   └── data_simpson.csv
│
├── scripts/
│   └── analysis_activity3.R
│
├── results/
│   ├── PCA_plots/
│   ├── clustering/
│   ├── heatmaps/
│   └── figures/
│
├── report/
│   ├── actv3_g8.html
│   └── actv3_g8.docx
│
└── README.md

---

🔬 Analysis Workflow

1️⃣ Data Preparation and Cleaning

• Selection of gene-expression variables
• Removal of missing values (NA)
• Verification of zero-variance genes

✔ No missing values or zero-variance genes were detected

---

2️⃣ Normality Assessment

A Shapiro–Wilk test was applied to each gene:

• Null hypothesis: data follow a normal distribution
• Result: all genes showed non-normal distributions (p < 0.05)

📌 This justified the use of:

• Standardization
• Multivariate and unsupervised methods

---

3️⃣ Principal Component Analysis (PCA)

PCA was performed using scaled and centered gene-expression values.

Variance explained:

Component	Variance (%)
PC1	12.5%
PC2	8.6%
PC3	6.5%

Together, the first components capture the dominant biological variability.

---

4️⃣ PCA Interpretation

🔹 Variable (Gene) Space

• Correlation circle used to visualize gene loadings
• Genes with longer vectors contribute more strongly
• Direction indicates correlation or opposition between genes

Notable contributors:

• PC1: POMC, ADCY3, PCSK1, LEPR
• PC2: NTRK2, CADM2, TMEM18

This reveals distinct biological axes related to appetite signaling and neuronal regulation.

---

5️⃣ Gene Clustering (k = 3)

Genes were clustered based on PCA coordinates:

• Cluster 1: Neuronal and synaptic regulation genes
• Cluster 2: Appetite and energy balance regulators
• Cluster 3: Hormonal and metabolic signaling genes

This grouping highlights functional modularity within obesity-related pathways.

---

6️⃣ Individual (Patient) Clustering (k = 3)

Clustering of individuals revealed:

• Two main dense groups near the PCA origin
• One clearly separated cluster with extreme PC1 values

This suggests the presence of molecular sub-phenotypes within the cohort.

---

7️⃣ Contribution Analysis

Barplots of variable contribution identified genes driving each axis:

• PC1: dominated by metabolic and hormonal genes
• PC2: enriched in neuronal development and signaling genes

Red dashed lines indicate the expected average contribution.

Genes above this threshold are biologically informative.

---

8️⃣ Heatmap of Gene Expression

• Scaled expression values
• Visual comparison across individuals
• Clear heterogeneity across samples

The heatmap supports PCA and clustering results, confirming structured variation rather than random noise.

---

9️⃣ BMI (IMC) Projection

Individuals were categorized as:

• Normal weight
• Overweight
• Obesity

When projected onto PCA space:

• Normal and overweight individuals largely overlap
• Obese individuals display greater dispersion

📌 This indicates:

Obesity is associated with increased molecular heterogeneity rather than a single transcriptional profile.

---

📊 Main Conclusions

• Gene-expression profiles show structured multivariate patterns
• PCA successfully reduces dimensionality while preserving biological meaning
• Obesity-related genes cluster into functional modules
• Individuals with obesity show greater transcriptomic variability
• Molecular phenotypes cannot be explained by BMI alone

---

🧪 Tools & Libraries

• R
• tidyverse
• factoextra
• pheatmap
• gtsummary
• gt
• broom

---

📘 Files Included

• actv3_g8.html – full reproducible report
• actv3_g8.docx – formatted academic submission
• All figures generated directly from code

---

✍️ Authors

Ana P Tedim

aartetxeizaguirre---

AlbaEliasRod

Alejandro Pascuál Hernandez

Alejandra Martin

📜 License

This repository is intended for academic and educational use only.

---

✨ This project demonstrates the application of unsupervised learning to biological systems, highlighting how multivariate statistics can reveal hidden structure in complex genomic data.