Density-Aware Spatial Randomization (DenSR) generates realistic null models for object-based colocalization analysis

Quantitative colocalization analysis in fluorescence microscopy is widely used to study molecular interactions among proteins, RNAs, and other cellular components. 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 molecular association from incidental overlap arising in crowded or spatially heterogeneous environments. Randomized null models are commonly used to estimate colocalization expected by chance, but they often fail to preserve the heterogeneous spatial density of biomolecules, leading to underestimation of random colocalization and reduced sensitivity to small effect sizes. Here, we introduce Density-aware Spatial Randomization (DenSR), an in silico framework that generates spatially realistic null models by preserving both local clustering and global spatial organization. Applying DenSR to protein and RNA datasets with heterogeneous spatial organization demonstrates that uniform randomization substantially underestimates background proximity, whereas density-aware null models provide more accurate expectations and prevent overestimation of colocalization due to underestimated background proximity. In contrast, analysis of sparse transcript datasets shows that different randomization strategies converge. Together, these results establish DenSR as a general approach for improving the estimation of colocalization expected by chance across diverse spatial distributions.