Uncovering directionally and temporally variable genes with STAVAG

Spatially variable genes (SVGs) are crucial for understanding spatial heterogeneity in spatial transcriptomics. Yet, SVGs are challenging to align with established biological axes and lack straightforward mechanistic interpretation, thereby limiting their ability to inform downstream experiments or clinical applications. Here, we introduce the concept of directionally variable genes (DVGs) and temporally variable genes (TVGs). Also, we propose a unified framework, STAVAG, that models spatial or temporal information to identify DVGs and TVGs. STAVAG effectively identifies biologically meaningful DVGs for uncovering prognostically relevant gene modules that align with the malignant-immune-stroma structure in the tumor microenvironment, identifying marker genes in three-dimensional brain data, and detecting 12 polarization genes regulating planarian regeneration. Furthermore, by tracking the progression of myocardial infarction and tissue development, STAVAG reveals critical TVGs that can serve as diagnostic biomarkers for the ultra-acute phase of myocardial infarction. STAVAG also identifies both global and local gene modules involved in mouse placentation, as well as tissue-specific and interactive TVG modules that dynamically evolve during mouse embryonic development.