Bayesian Network Analysis of Treatment-Gene Expression Relationships in Glioblastoma Multiforme

Authors

• Olivia Williamson
• Natalie Goulett
• Gabriel Odom

Overview

This repository contains R scripts and associated data for a research project analyzing gene expression and treatment effects in patients with Glioblastoma Multiforme (GBM). Using two datasets from the Gene Expression Omnibus (GEO) database, we constructed a Bayesian network to explore relationships between three chemotherapy treatments (Regorafenib, Lomustine, and Selinexor) and gene expression within these patients.

The project pipeline, as outlined in Comparing Three Chemotherapy Treatments' Effects on Gene Expression in Patients with Glioblastoma Multiforme using the Hill Climbing Algorithm, by Williamson, O. and Goulett, N. (2024), involves data acquisition, preprocessing, normalization, merging, batch effect removal, and Bayesian network analysis. The Bayesian network was built using the bnlearn package with the Hill Climbing algorithm.

Scripts

Each R script corresponds to a specific step in the analysis pipeline. The scripts are designed to be run sequentially.

1. 1_DownloadData.R

• Purpose: Download raw count data from GEO for two datasets: GSE186332 (raw counts, normalized later) and GSE154041 (pre-normalized with TMM).
• Authors: Olivia Williamson & Natalie Goulett
• Details: Raw counts and metadata are downloaded and stored locally. Gene names are converted to Hugo gene symbols for consistency across datasets.

2. 2_AddGeneSymbols.R

• Purpose: Convert gene identifiers to consistent Hugo gene symbols across both datasets.
• Authors: Olivia Williamson, Natalie Goulett, Gabriel Odom
• Details: Utilizes the hgnc package to map gene symbols, preparing datasets for subsequent analysis.

3. 3_NormalizationDiscretization.R

• Purpose: Normalize gene expression data and discretize it into categories of over-, non-, and under-expressed genes.
• Authors: Olivia Williamson, Natalie Goulett, Gabriel Odom
• Details: The TMM method is applied to raw counts, then counts per million (CPM) are calculated and scaled. Data is discretized based on thresholds.

4. 4_MergeRemoveBatchEffects.R

• Purpose: Merge datasets and remove batch effects to ensure compatibility for network analysis.
• Authors: Olivia Williamson, Natalie Goulett, Gabriel Odom
• Details: Uses limma to remove batch effects between GSE154041 and GSE186332, ensuring harmonized data before Bayesian network analysis.

5. 5_GetMatchedPhenoData.R

• Purpose: Match and merge phenotype data for both datasets.
• Authors: Olivia Williamson, Natalie Goulett, Gabriel Odom
• Details: Extracts treatment and sample information, transforming treatment data into binary columns representing each treatment group for further analysis.

6. 6_BuildBN.R

• Purpose: Build the Bayesian network to analyze gene expression interactions and treatment effects.
• Authors: Olivia Williamson, Natalie Goulett, Gabriel Odom
• Details: Uses the bnlearn package to build a Bayesian network with the Hill Climbing algorithm. Results are visualized and analyzed for significant gene-treatment interactions.

Datasets

The datasets used in this project:

• GSE154041: Contains 74 samples from patients treated with Regorafenib or Lomustine.
• GSE186332: Contains 57 samples from a phase-II Selinexor trial with four treatment groups.

Both datasets were acquired from NCBI's GEO database.

Analysis Summary

• Normalization: Raw counts were normalized with TMM, followed by transformation to CPM.
• Merging: Gene symbols were standardized, and batch effects removed.
• Bayesian Network Construction: The Hill Climbing algorithm was used to identify genetic and treatment interactions, highlighting key pathways in GBM progression and treatment response.

Results and Discussion

The resulting Bayesian network identifies several significant gene-treatment interactions, including the influence of Lomustine on the NGFR gene. Conditional probability tables provide insights into these relationships.

Dependencies

This project requires the following R packages:

• BiocManager, GEOquery, Biobase, biomaRt, tidyverse, hgnc, edgeR, limma, pathwayPCA, bnlearn, Rgraphviz

Install packages using:

# Install Bioconductor packages
BiocManager::install(c("GEOquery", "Biobase", "biomaRt", "edgeR", "pathwayPCA", "bnlearn", "Rgraphviz"))

# Install CRAN packages
install.packages(c("tidyverse", "hgnc", "limma"))