For half a century, cancer researchers have had a recurring problem: we keep learning more about cells by removing them from the very neighborhoods that explain their behavior. It is a little like interviewing every guest after a dinner party, but only after blindfolding them, putting them in separate taxis, and asking who was flirting with whom near the cheese plate.
Spatial transcriptomics was invented to fix that awkwardness. Instead of grinding tissue into biological confetti, it lets scientists measure gene activity while keeping cells in place. You get the “what” and the “where.” In cancer, that matters enormously, because tumors are not just lumps of bad cells. They are messy little cities, with immune cells, blood vessels, fibroblasts, exhausted T cells, suspicious myeloid cells, and cancer clones all arguing over the zoning laws.
The new paper by Sun and colleagues introduces Cellist, a computational method for identifying cells in high-resolution spatial transcriptomics data across multiple platforms [1]. That sounds technical because it is technical. But the idea is simple: if you want to understand a tissue map, you first need to draw the borders correctly.
The Cell Border Problem, or Biology’s Bad Coloring Book
Modern spatial transcriptomics can detect RNA molecules at very fine resolution, sometimes even below the size of a cell. Wonderful. Splendid. Also inconvenient.
Those RNA dots do not arrive politely labeled, “Good evening, I belong to cell 47.” They scatter through tissue like crumbs after a departmental seminar. Algorithms must decide which transcripts belong inside which cell boundary. That process is called cell segmentation, and it has become one of the great unglamorous bottlenecks in the field.
Earlier methods often lean heavily on images, such as nuclear stains, or work best on one platform at a time. That is useful, but biology has developed the inconsiderate habit of appearing on many technologies: Stereo-seq, Seq-Scope, seqFISH+, STARmap, Xenium, and whatever machine will appear next year with a heroic acronym and a price tag that makes grant officers blink.
Cellist tries to be more diplomatic. It combines image information with gene-expression information, then uses both to assign transcripts to cells. In mouse brain Stereo-seq data, the authors report better within-cell transcriptomic coherence. In plain English: the algorithm produced cell boundaries whose RNA contents made more biological sense.
Why Cancer People Should Care
The oncology payoff comes when the authors apply Cellist to non-small cell lung cancer samples after neoadjuvant immunotherapy. That is treatment given before surgery, the therapeutic equivalent of softening up the battlefield before the surgeon arrives.
Here, better cell identification helped reveal spatial heterogeneity among tumor clones and pointed to therapy-response-related myeloid subtypes and structures. Myeloid cells are immune-system characters with complicated resumes. Depending on context, they can help attack tumors, suppress immune responses, clean up debris, or generally stand around looking busy while the tumor gets away with things.
This is where spatial context earns its keep. A myeloid cell near a tumor nest may not mean the same thing as a similar-looking myeloid cell parked elsewhere. Location changes the story. In the old days, we measured averages and hoped the tissue would confess. Now we can ask who is standing next to whom, who is excluded, and which neighborhoods seem linked to treatment response.
That is not a small upgrade. It is the difference between knowing a city has police officers and knowing which streets they actually patrol.
A Long Road From In Situ Hybridization
None of this came from nowhere. The ancestors of spatial transcriptomics include in situ hybridization, developed decades ago to see where specific nucleic acid sequences sit in tissue. Later came FISH methods, single-cell RNA sequencing, and then the modern spatial transcriptomics boom, which Nature Methods named its 2020 Method of the Year [2].
The field has been moving fast. Reviews have emphasized that spatial transcriptomics can connect gene expression to tissue architecture in cancer, neuroscience, development, and beyond [2,3]. Other work has warned, quite sensibly, that spatial data only become useful after careful computational handling, because analysis choices can shape what researchers think they see [4]. Anyone who has watched a computer cluster turn a messy dataset into a confident-looking figure knows the feeling. The machine is very sure of itself. The biologist should remain less so.
Cellist enters at exactly that pressure point. It is not proposing a new drug, target, or immune checkpoint. It is improving the measurement layer underneath those discoveries. That may sound modest, but measurement tools often change medicine quietly before they change it loudly. The microscope did not cure cancer either, but I would not want to run a pathology department without one.
What Happens If This Holds Up?
If Cellist proves reproducible across more tissues, labs, and clinical cohorts, it could make spatial transcriptomics more reliable for studying tumor ecosystems. Researchers could compare datasets across platforms with less fear that they are comparing algorithmic quirks. Clinicians might eventually use spatial immune patterns to understand why one lung tumor melts under immunotherapy while its neighbor behaves like it has hired legal counsel.
There are still hurdles. Spatial methods remain expensive, computationally demanding, and full of platform-specific wrinkles. RNA is not protein. Tissue sections are thin slices of three-dimensional chaos. And no segmentation tool, however elegant, can rescue a poor experiment. Biology, as ever, reserves the right to be annoying.
Still, Cellist is the kind of advance that makes the next discoveries less blurry. It helps turn spatial transcriptomics from a beautiful molecular pointillist painting into a map with streets, houses, and addresses. In cancer research, where the neighborhood often explains the crime, that is very welcome indeed.
References
-
Sun D, Zhang L, Han T, Wu Q, Zhang P, Wang C. Accurate, scalable and cross-platform cell identification for high-resolution spatial transcriptomics. Nature Genetics. 2026. DOI: 10.1038/s41588-026-02610-1. PMID: 42162410.
-
Rao A, Barkley D, França GS, Yanai I. Exploring tissue architecture using spatial transcriptomics. Nature. 2021;596:211-220. DOI: 10.1038/s41586-021-03634-9.
-
Longo SK, Guo MG, Ji AL, Khavari PA. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nature Reviews Genetics. 2021;22:627-644. DOI: 10.1038/s41576-021-00370-8. PMCID: PMC9888017.
-
Petukhov V, Xu RJ, Soldatov RA, et al. Cell segmentation in imaging-based spatial transcriptomics. Nature Biotechnology. 2022;40:345-354. DOI: 10.1038/s41587-021-01044-w.
-
Lan Z, Yang Y, Li L, et al. Spatial omics technology potentially promotes the progress of tumor immunotherapy. British Journal of Cancer. 2025;133:421-434. DOI: 10.1038/s41416-025-03075-5.
Disclaimer: The image accompanying this article is for illustrative purposes only and does not depict actual experimental results, data, or biological mechanisms.