The problem. No segmentation is perfect. Transcripts from one cell get assigned to its neighbor, and the result is contamination: a cell’s expression profile is polluted by whatever was next to it. This inflates spurious co-expression and blurs cell-type boundaries, and it persists no matter which segmenter you used.

The idea. MisTIC treats missegmentation as something to detect and correct after the fact, rather than a problem you can only solve upstream. The approach identifies transcripts likely to have been misassigned, using spatial and expression context to flag molecules that don’t fit their assigned cell, and corrects the resulting contamination in the cell-by-gene matrix.

Why it matters. This completes the segmentation cluster: Baysor draws the boundaries, Segger tries to draw them faster, and MisTIC accepts that whatever you drew still leaks and cleans up after it. For a facility, that layered view, and a QC step for the QC step, is exactly the reproducibility mindset. Contamination is a quiet source of false biology; naming and correcting it is worth a slot.

Verdict. A 2025 preprint, so I’m holding its specific claims loosely until peer review and independent testing. Conceptually it fills a real gap. I’d want to see how much correction changes real cell-type calls, and whether it ever over-corrects genuine signal, before leaning on it. One to read closely against the manuscript.