Choosing between imaging-based spatial transcriptomics and sequencing-based approaches is the first platform decision, and everything else follows from it. One family gives you the whole transcriptome and needs sequencing capacity plus serious compute. The other gives you subcellular resolution on a fixed panel and occupies an instrument for days. The science is comparable. The operational demands are not.
Key Takeaways
|
How Sequencing-Based Methods Work
Probes hybridize to transcripts in the tissue section, then get captured onto a slide carrying spatially barcoded features. The barcode records where on the slide each molecule came from, and the readout happens on a sequencer rather than on the spatial instrument. Because the probe set covers the transcriptome, you are not choosing targets in advance. Visium and Visium HD from 10x Genomics are the systems most labs evaluate here, and Bruker’s GeoMx profiles user-defined regions of interest rather than a continuous grid.
Operationally, this creates two dependencies outside the spatial instrument: reliable sequencing access, in-house or through a core, and enough compute to handle a very large number of barcoded features. Both are commonly underestimated. The deeper mechanism, including probe chemistry and analyte transfer, is covered on Technology Networks: Exploring the Latest Methods in Spatial Transcriptomics.
How Imaging-Based Methods Work
Transcripts are detected in place, on the instrument, by repeated cycles of fluorescent probe hybridization and imaging. Each target accumulates an optical signature across cycles, and transcripts are then assigned to individual cells by image segmentation. Nothing leaves the instrument for sequencing. The trade-off is that you must select your genes beforehand, because detection is limited to the panel you ordered. Xenium from 10x Genomics, CosMx from Bruker, and MERSCOPE from Vizgen are the systems usually compared.
The operational profile is close to the inverse. No sequencer dependency, but the instrument is tied up for every run, and the output is image data rather than sequence data, which changes your storage requirement and the skills your analyst needs. Segmentation quality also becomes a variable you have to care about, since it decides which transcript belongs to which cell.
Resolution or Whole Transcriptome: Which Trade-off Can You Accept?
Historically, this was a clean split: sequencing-based gave breadth at multicellular resolution, imaging-based gave single-cell resolution on a narrow panel. Newer chemistries have narrowed it from both directions, but the trade-off has not gone away. It is now a question of what you give up, not whether.
| Sequencing-Based | Imaging-Based |
Coverage | Whole transcriptome, no target selection needed | Pre-selected panel, hundreds to a few thousand genes |
Resolution | Set by feature or bin size, and in practice by compute | Individual transcripts, assigned by segmentation |
Readout location | Off instrument, on a sequencer | On instrument, by cyclic imaging |
Instrument occupancy | Slide preparation only, then the sequencer queue | Days per run, scaling with panel size |
You also need | Sequencing access and substantial compute | Bench footprint and image storage |
Cell assignment risk | Bins can mix adjacent cells | Segmentation can misassign cell boundaries |
Strongest for | Discovery, pathway analysis, unknown targets | Validation, defined targets, cell neighborhoods |
Table 1. The two families compared operationally. Figures describe the configurations used in the benchmarking discussed below rather than full product specifications. Confirm current specifications with vendors.
Operational Differences That Decide Purchases
Three practical points settle more decisions than any performance figure, and none features prominently in vendor comparisons.
- Instrument occupancy scales with panel size. In the benchmarking study below, a 377-gene imaging run took around 50 hours on the analyzer. A 5,000-gene run on the same platform took about 7 days. For throughput modeling or a chargeback schedule, that gap matters far more than plex count.
- Headline resolution is often theoretical. The high-definition sequencing-based configuration provides a 2 micron barcoded grid, roughly 11 million features per capture area, which the study’s authors describe as carrying extraordinary computational cost. Bins are therefore typically aggregated to 8 or 16 microns, and that aggregation is what reintroduces cell mixing. Budget the compute or accept the coarser resolution.
- Sample quality gates both families equally. The study found a strong correlation between DV200, the proportion of RNA fragments above 200 nucleotides, and average genes detected per cell on both. A poor block will disappoint whichever family you choose.
On that last point, 10x Genomics’ own Visium FFPE tissue preparation guidance identifies DV200 as the RNA quality determinant to assess before library preparation, and the benchmarking group screened blocks above 34% DV200 before proceeding, a reasonable starting point for your own incoming-sample criteria. One failure from the study is worth noting: a lipid-rich breast sample partially detached during the 95-degree decrosslinking step, folded, and produced a failed library. Fragile and fatty tissues carry real handling risk in sequencing-based workflows, which belongs in your repeat-rate assumptions and in your section-level tracking, as covered in Sample Management in High-Volume Biological Studies.
What Does the Benchmarking Show?
Most published comparisons come from vendors or from providers selling one platform. An exception is a technical comparison of spatial transcriptomics platforms across six cancer types, published open access in Genome Biology in January 2026, which profiled matched formalin-fixed paraffin-embedded serial sections from six tumor types across five commercial platforms spanning both families, registered to a shared coordinate system.
Finding | What It Means Operationally |
The high-definition sequencing platform showed UMI density roughly 3 times lower than the standard version and about 15 times lower than the imaging platform | Finer resolution comes with sparser data per unit area, shifting work onto analysis |
Gene-level correlation between families was strong, above 0.85, and above 0.97 within the sequencing family | The families broadly agree at gene level, so this is not a question of one being wrong |
Among imaging platforms, background signal and low-abundance reliability differed substantially | Within a family, specificity is a real differentiator worth testing on your tissue |
At 8 micron bins, features frequently contained transcripts from 2 to 4 cells; at 16 micron, from 5 to 10 | The resolution you work at is a compute decision, not a specification |
The sequencing platform over-represented epithelial and stromal cells relative to immune cells | Larger cells span more bins and get over-counted, biasing composition estimates |
Whole-transcriptome coverage enabled robust pathway scoring and revealed macrophage subpopulations the panel platform missed | If you need pathway-level or unanticipated biology, panels constrain you |
A single 0.5 x 0.5 mm field of view estimated cancer cell proportion anywhere from 5% to 55%, against 22% for the whole section | Scanned area drives interpretation. Small regions give unreliable composition estimates |
Table 2. Selected findings from Cervilla et al., Genome Biology, 2026. The authors state their data reflect capabilities as of mid-2025, and that the study is a technical reference rather than a platform recommendation.
Within the Imaging Family, a Bigger Panel Is Not an Upgrade The same group compared a 377-gene panel against a 5,000-gene panel on consecutive sections from five tumor types using identical segmentation. The larger panel detected more in total, but nowhere near in proportion to the roughly thirteenfold size difference. On the 225 genes shared between panels, the smaller panel detected more transcripts and more genes per cell, and the larger panel produced around three times fewer transcripts per gene with weaker spatial coherence, attributed to optical crowding. The interpretive consequence is stark. In one colorectal sample, cancer cells classified as MYC-positive were 70% with the smaller panel and 30% with the larger one. Same tumor, same chemistry, same instrument. The larger panel was not simply worse: it improved cell type annotation, enabled deeper immune subtyping, and corrected epithelial cells that the smaller panel misclassified as fibroblasts. Panel size is a trade-off to specify deliberately, not a capability to buy more of. |
Lab Management Certificate
The Lab Management certificate is more than training—it’s a professional advantage.
Gain critical skills and IACET-approved CEUs that make a measurable difference.
Two caveats belong with all of the above. The comparison reflects mid-2025 configurations, and the field moves fast enough that the authors flag one imaging platform having since added whole-transcriptome capability, which erodes the central distinction in this article. And the study was built as a technical reference, not a buying guide, so read it as a map of trade-offs rather than a ranking.
Which Should You Choose for Common Goals?
Work from the question and the constraint, not from the platform. Four situations cover most decisions.
If This Describes You | Lean Toward |
You do not yet know which genes matter, or you need pathway-level analysis | Sequencing-based, for whole-transcriptome coverage |
You have defined targets and need cell-level or subcellular localization | Imaging-based, with the smallest panel that covers your targets |
You have no reliable sequencing access, or the sequencing budget is unsecured | Imaging-based, which keeps the workflow on one instrument |
You have no storage and compute plan, or no analyst with allocated time | Neither yet. Close that gap first, or outsource this project |
Table 3. A starting point rather than a decision. Sample type, tissue area, and budget constrain the answer further.
Note the last row. Both families push work downstream onto people and infrastructure you may not have, one as sparse data across millions of features and the other as large image datasets. Confirm the lab can absorb it using Is Your Lab Ready for Spatial Biology? A Readiness Assessment, and model the real cost using the estimator in How Much Does Spatial Biology Cost? Budgeting for Instruments, Consumables, and Data. Storage and compute sizing is covered in Managing Spatial Biology Data: Storage, Compute, and Infrastructure for Spatial Datasets.
Interested in life sciences?
In practice, the families pair well: profile broadly to find which genes and regions matter, then image those targets at the cell level. The strong gene-level agreement is what makes that handoff defensible. If your path involves both, outsource the discovery step and bring in-house whichever platform your recurring work depends on. System-by-system comparisons are in Spatial Transcriptomics Platforms Compared, the procurement process in Choosing Between Spatial Biology Platforms: A Lab Manager's Buyer's Guide, and the whole-workflow picture in Spatial Biology in the Lab: A Manager's Guide to Evaluating, Implementing, and Scaling Spatial Technologies.
This article was produced under Lab Manager's AI Editorial Guidelines.
