TIGP (BIO)—From Rewired Edges to Multilevel Modules: Context and Scale in Transcriptomic Network Analysis

Abstract

Transcriptomic studies often begin by identifying differentially expressed genes, yet important biological signals may also lie in altered gene-gene relationships and their multiscale organization. In this talk, I will present two complementary network strategies for resolving these dimensions.
The first is a phenotype-filtered differential co-expression framework developed to investigate the selective activity of GL24, a 4-(phenylsulfonyl)morpholine derivative, in triple-negative breast cancer cell lines derived from metastatic lesions. The framework retains gene pairs that gain or lose a significant co-expression relationship in both GL24-responsive cell lines while showing no differential co-expression in a non-responsive cell line. By anchoring edge rewiring to the GL24-response phenotype, this design highlights network patterns associated with metabolic alterations, proliferation, and migration or invasion, thereby complementing conventional differential expression analysis.
The second is the Minimum Span Clustering Network (MSCN), an unsupervised, deterministic method that recursively organizes genes into a directly traceable parent-child hierarchy of co-expression modules. Applied to two independent postmortem ASD brain transcriptome cohorts, MSCN recovered multilevel module hierarchies and identified ASD-associated modules enriched for neuronal, synaptic, developmental, and immune processes. Mapping long noncoding RNAs onto MSCN-derived mRNA modules then enabled statistical mediation analysis to prioritize candidate TF-lncRNA-mRNA mediation axes, with a subset receiving concordant support from external transcriptomic data.
Together, these studies illustrate two organizing concepts of transcriptomic network analysis: biological context determines which relationships are relevant, whereas analytical scale determines which structures become visible. The talk will conclude with a discussion of how edge-level and module-level perspectives can be integrated, as well as the statistical and interpretive limits of correlation-based networks.

Download