Structured Expression Dispersion as a Derived Projection of Latent Biological Axes: A Mechanistic Reinterpretation of Survival Signals in TCGA Glioblastoma (v4.0)

This study evaluates whether gene expression dispersion (coefficient of variation, CV) represents a true survival signal or a projection artifact in high-dimensional transcriptomic data. In TCGA glioblastoma, CV stratifies survival; however, falsification through random gene controls, permutation testing, and cross-cancer analysis demonstrates that the signal is not robust or generalizable. Principal component analysis (PCA) substantially outperforms CV in survival prediction, and CV becomes non-significant when PCA is included. Correlation analysis shows that CV partially overlaps with latent structure, indicating that it represents a degraded projection. Enrichment analysis identifies a biologically coherent axis separating metabolic transport processes from extracellular matrix and angiogenic remodeling. These findings demonstrate that dispersion-based survival signals arise from structured biological programs rather than independent statistical mechanisms.