Colocalization analysis in fluorescence microscopy is a widely used method to study molecular interactions, such as those between proteins or RNAs. Object-based approaches identify discrete molecular features and quantify distances between neighboring centroids to infer colocalization. A key challenge in this approach is distinguishing true biological association from colocalization arising from random spatial overlap of fluorescence signals, which is typically addressed by randomized null models. However, existing randomization strategies often fail to preserve the local spatial density of biomolecules, especially in crowded subcellular regions. This can lead to an underestimation of colocalization by chance and a loss of sensitivity to small effect sizes. In this study, we developed an in-silico framework for density-aware spatial randomization (DenSR) that preserves both regional molecular density and native spatial distributions. To validate this approach, we applied it to previously published datasets, and protein pairs with well characterized low or high degree of colocalization, allowing us to assess whether DenSR can distinguish true biological interactions from incidental spatial overlap. Our results show that DenSR enhances the specificity and sensitivity of object-based colocalization analysis across various datasets.
Voelker et al. (Sun,) studied this question.