Marker-specific pixel attribution for single-cell protein quantification with PixAQ

Abstract

Imaging mass cytometry (IMC) typically quantifies all markers within the same cell mask, although their signals have distinct spatial distributions. This quantification method may omit relevant signals or include signals from neighboring cells. We present PixAQ, which improves single-cell protein quantification in IMC through marker-specific probabilistic pixel attribution. PixAQ combines spatial proximity with anchor-group intensity models to assign pixel-marker observations among neighboring cells and an explicit residual component. Original intensities are averaged over marker-specific supports. Across three cohorts (680 regions), comparisons in 366 patients showed improved agreement with predefined population-marker relationships, with cohort-median patient-paired gains of 1.99 to 4.83 percentage points in expected-marker share over shared-mask quantification. Ablation analyses showed that both spatial and intensity evidence contributed to these gains. Conventional and high-resolution IMC images linked supports to visible local and intracellular signal structures. PixAQ improves single-cell protein quantification by determining which local signals contribute to each marker measurement without changing cell segmentation.