gamdid: generalized additive models for differential distributions in single cell experiments

Single-cell proteomics (SCP) generates protein abundance measurements across hundreds to thousands of individual cells, offering unprecedented resolution to study cellular heterogeneity. However, existing differential abundance (DA) methods are limited to detecting shifts in mean expression, leaving biologically relevant differences in shape undetected. Indeed, the specific power of SCP is to identify differences between individual cells in a population, which are typically only found as shape differences rather than in mean expression. We here therefore present gamdid (generalized additive models for differential distributions), a novel statistical framework and R package for differential distribution (DD) analysis in SCP data. gamdid is based on generalized additive models (GAMs) to flexibly model heterogeneous distributions, perform inference and provide interpretable visualizations. Through semi-synthetic benchmarking on two SCP datasets, gamdid demonstrates conservative false discovery rate control and substantially outperforms competing methods for differences in shape, while achieving comparable performance for mean shifts. A spike-in case study further demonstrates the utility of gamdid and its interpretable visualization. Uniquely among DD methods, gamdid supports omnibus testing across more than two groups, with post-hoc pairwise comparisons via stagewise testing, and is specifically tailored for proteomics abundance data.