High-Precision Molecular Profiling and Diagnostic Verification of Breast Cancer Subtypes using PAM50 and Empirical Bayes Modeling

Abstract

High-Precision Molecular Profiling and Diagnostic Verification of Breast Cancer Subtypes using PAM50 and Empirical Bayes Modeling :Punyawatt PitaksuntisookInaccurate molecular subtyping of breast cancer is a critical challenge in clinical oncology because misdiagnosing luminal or basal-like tumours can lead to incorrect chemotherapy selection. In this Independent Study project, I built a computational bioinformatics pipeline to verify sample subtyping and systematically identify key driver genes distinguishing Luminal A and Basal-like breast cancer.Using public microarray data from NCBI GEO (Accession GSE45827, n=155 across 12,499 unique genes), I implemented a PAM50 nearest-centroid correlation classifier to audit sample annotations and correct cohort mismatches. Next, I applied an Empirical Bayes moderated t-test for variance shrinkage, identifying 3,762 differentially expressed genes (FDR < 0.05, |log2FC| > 1.0). Gene Set Enrichment Analysis (GSEA) showed strong upregulation of cell cycle and mitotic spindle machinery in Basal-like tumours, driven by the master regulator FOXM1. Survival modelling using TCGA-BRCA data confirmed that high FOXM1 expression (HR = 2.05, p = 4.3×10⁻¹²) and low AGR3 expression (HR = 0.68, p = 8.1×10⁻⁶) correlate significantly with poor overall survival. The results were also validated against the independent METABRIC dataset (n=1,904).