Answers are grounded in this paper and cite their sources.
Highly multiplexed imaging techniques are an important part of the biological analysis toolkit. However, the resulting images can suffer from considerable noise and spillover. Because accurately correcting spillover normally requires additional wet-lab experiments to determine how the channels interact with each other, here we present SpillOT, a purely computational zero-shot denoising and spillover correction tool that operates in an unsupervised manner, learning solely from the input image without the need for additional experiments or representative training datasets. SpillOT uses optimal transport theory to identify similar patches across channels and retain only the true signal. We demonstrate that SpillOT surpasses existing zero-shot image restoration techniques and markedly improves the reliability and precision of downstream analyses on Imaging Mass Cytometry, Tissue Cyclic immunofluorescence and Opal multiplexed immunofluorescence datasets. We also show that using SpillOT, we can find previously undetectable immune infiltration patterns in diabetes data. We release SpillOT as both a standalone script and a Fiji plugin.
No takes yet. Share an insight, caveat, or question.
Lequyer et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: