ELLA: modelling subcellular spatial variation of gene expression

Spatial transcriptomic technologies are becoming increasingly high-resolution, enabling precise mapping of mRNA localization within cells. We introduce a computational framework that models subcellular gene expression using a unified cellular coordinate system and a nonhomogeneous Poisson process, capturing spatial variation while maintaining strong statistical control. Through simulations and analyses of four diverse spatial transcriptomic datasets, the method identifies genes with distinct localization patterns and links them to molecular features. For example, long noncoding RNAs or long protein-coding mRNAs are more likely to be nuclear-enriched, whereas transcripts encoding ribosomal or membrane-associated proteins are more likely to be localized to the cytoplasm or cell periphery. Dynamic localization is also observed across cell-cycle phases.