#33 · Systems biology

Quantifying per-gene predictability of expression from H&E histology images

The dominant thread is linking tissue morphology and transcriptomic data through quantitative statistical frameworks. The preprint introduces the Expression Predictability Score (EPS), a mutual-information metric that quantifies how predictable each gene's expression is from H&E histology images. The published JSNMF model fuses pathology images, gene expression and pathway scores via nonnegative matrix factorization to identify association patterns linked to tumor recurrence. A second published method, PGNEA, addresses gene network enrichment through permutation of gene activity and reports improved speed and power over GbNEA in simulations and in application to a severe COVID-19 gene network. Together these items extend how expression data are interrogated, although PGNEA addresses network enrichment rather than histology directly.