PRISM: probabilistic cell-type resolution with integrated spatial modelling for simultaneous ambient RNA decontamination and doublet correction in spatial transcriptomics

PRISM jointly models ambient RNA contamination and cell-segmentation doublets in spatial transcriptomics, achieving Pearson r=1.000 across contamination 0.05–0.50 and UMI 500–5,000 in 40,800 simulated cells.

• Why it matters: Ambient RNA and doublets co-occur in spatial count matrices, but are typically handled separately, limiting accurate cell-type resolution and downstream biological interpretation.
• What they did: PRISM uses an Expectation-Maximization framework to assign probabilistic cell types, call doublets via log-likelihood ratio testing of two-component mixtures, and estimate per-cell contamination using a spatially resolved ambient profile.
• The result: In Visium HD mouse cerebellum from an MLD model (four samples, 125,289 cells), PRISM reports median contamination=0.60, flags 67.4% doublets, removes spurious microglia myelin signal (0.82% to 0.00%), and increases oligodendrocyte myelin purity (2.8% to 5.6%).

AI summary of the abstract · may contain errors · not medical advice · read the paper.