DisCell: Separating Spillover from Niche Response in Spatial Transcriptomics

DisCell jointly models intrinsic state, niche response, and neighbor spillover in spatial transcriptomics, quantifying robustness and separating spillover-driven from genuine niche programs.

• Why it matters: Spatial counts mix true local biology with diffusion and segmentation spillover, making niche-linked expression patterns ambiguous when using count data alone.
• What they did: DisCell is a generative model with two latent variables plus an explicit spillover term, fitting across increasing spillover and recomputing findings on spillover-free expression.
• The result: In simulated tissue it recovers intrinsic state and niche response and distinguishes spillover-created findings; on ovarian and lung sections it preserves cell-cycle phase more than competing methods and ranks niche findings by robustness.

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