Abstract

Subcellular omics technologies now allow us to obtain insights into steady-state localisation and re-localisation of biomolecules in high-throughput. However, robust analysis of these experiments can be slow and challenging. Here, we show that existing approaches to differential localisation fall into two classes, marker-dependent and marker-free, that fail for fundamentally different statistical reasons, with failure modes that are not completely overlapping. We exploit this observation in SANDLE (statistical analysis of differential localisation experiments), which combines a marker-dependent generative model of subcellular niches with a marker-free regression test for changes in fractionation profiles. This dual strategy provides better control of false positives, up to 100-fold reduction in analysis time, and accommodates more complex experimental designs than existing methods. We demonstrate SANDLE's versatility across a wide range of subcellular omics experiments, including drug-treatment responses, cross-species and life-cycle comparisons, post-translationally modified proteoforms, and RNA re-localisation. Co-analysing transcriptome and proteome dynamics during the unfolded protein response, we further reveal lncRNAs whose steady-state localisation is condition-dependent. By addressing key methodological limitations, SANDLE enables broad applications of subcellular omics from fundamental biology to clinical research.