Evaluation of NMF-VAE Integrative Approach for Biclustering and Glioblastoma Biomarker Identification

Glioblastoma (GBM) represents the most aggressive primary brain tumor with poor prognosis. This research develops a novel computational framework that merges the strengths of Non-negative Matrix Factorization (NMF) with Variational Autoencoder (VAE) to improve biclustering performance in GBM gene expression data analysis. Using the GSE4290 dataset, this study analyzes gene expression data from 180 samples (136 tumors and 44 normal controls). The implementation of the NMF-VAE method successfully identified 10 biclusters with coherence values of 0.711 and variance of 0.713, validated through latent space visualization and reconstruction error analysis (15-50 MSE). Differential expression analysis identified three main potential biomarkers: ANXA2, TNFRSF1A, and NAMPT, which demonstrated significant expression changes (fold change 2.5, 2.0, and 3.0) and correlated with tumor cell proliferation, inflammation, and energy metabolism. Visualization of bicluster patterns and gene expression value distributions confirmed the consistency of these biomarkers overexpression in tumor samples. These findings provide new insights into the development of gene expression-based treatment strategies for GBM patients