For most of the last two decades, transcriptomics meant bulk RNA-seq. You took a piece of tissue, lysed it, and measured average gene expression across whatever cells happened to be present. It was powerful. It was also, in a real sense, a smoothie. You learned what genes were active in the mix — not which cells were expressing what, or where those cells sat in relation to each other.
Single-cell RNA-seq fixed the cell identity problem. You could disaggregate tissue, sequence individual cells, and map the heterogeneity. But you lost spatial context entirely. Cells in suspension look like cells in suspension. You could not know whether a fibroblast sat adjacent to a tumor-infiltrating T cell, or whether the gene expression you measured in a specific macrophage subtype was coming from the stromal margin or the tumor core.
Spatial transcriptomics puts that context back. The current generation of platforms can profile thousands of genes simultaneously at near-single-cell resolution, across an intact tissue section, with each measurement georeferenced to a physical coordinate in the tissue. That is not a small step forward. It changes what questions are even askable.
Why Location Matters for Target Biology
The drug discovery implication is direct. Most targets do not function identically across all contexts. A receptor that is functionally relevant in the tumor microenvironment edge — where hypoxia, immune infiltration, and matrix composition all shift — may behave very differently than the same receptor deeper in the tumor core. If your target identification was based on bulk RNA-seq of whole tumor biopsies, you may have been measuring an average that does not correspond to the biologically relevant cell population at all.
We have seen this pattern in our portfolio diligence. Companies that used spatial readouts during target validation caught cell-type specificity that bulk data would have missed. In one case, a target that looked clean in bulk transcriptomics turned out to be expressed primarily in a stromal cell subtype that happens to co-localize with the cell type of interest — not in the therapeutic target cell at all. That distinction mattered enormously for whether the program had a credible mechanism.
The question is no longer just "what is expressed" — it is "expressed where, by which cell, in what neighborhood." Spatial data makes that question answerable for the first time at scale.
Where the Technology Actually Is Right Now
There are a few distinct approaches worth separating out. Sequencing-based spatial methods capture transcriptome-wide or near-transcriptome-wide readouts, typically at 10-55 micrometer resolution depending on the platform. Imaging-based methods — in situ sequencing and highly multiplexed fluorescence approaches — achieve subcellular resolution but are limited to panels of a few hundred to a few thousand genes.
Neither is definitively better. The right tool depends on the biology. If you need transcriptome-wide unbiased discovery, you use a sequencing approach and accept the resolution tradeoff. If you need true single-cell spatial resolution with known targets, you use an imaging approach. Most serious programs use both.
Throughput has been the historical constraint. Processing spatial sections is time-consuming and expensive — a significant obstacle to running the kind of large-sample validation needed before a target gets committed to a program. That is changing. Instrument throughput has increased roughly 4-fold in the last eighteen months across major platforms, and cost per slide has dropped enough to make 50-100 sample studies feasible in an industrial drug discovery context, not just in academic labs running n=5.
The Diligence Question We Ask
When we evaluate drug discovery companies that claim spatial genomics is central to their platform, we ask a specific question: where in the development workflow does the spatial readout actually inform a decision that would otherwise have been made differently?
Target selection is the obvious answer, but not the only one. Spatial data is increasingly used in patient stratification — identifying tumor microenvironment archetypes that predict response to a given therapy better than bulk transcriptomic signatures can. It is also appearing in translational biomarker work, where you need to understand which tissue compartment a blood-based biomarker is actually tracking.
What we do not find compelling is spatial data used as a visualization tool or a marketing claim — "we profiled our lead target in spatial" — without a clear statement of what decision it changed. The technology is expensive enough, and the data complex enough, that if you cannot say concretely how it affected your program strategy, it probably did not.
Where the Hype Gets Ahead of the Data
A few areas deserve skepticism. The claims about spatial transcriptomics unlocking new target classes are often premature. The tools are powerful, but identifying a spatially restricted gene expression pattern is not the same as validating a druggable target. The path from observation to therapeutic hypothesis still requires functional validation, genetic evidence of disease relevance, and an assessment of whether the target is actually accessible to a drug modality. Spatial data can sharpen the starting hypothesis; it does not compress the rest of the process.
There is also an emerging data analysis problem that is underappreciated. Spatial datasets are large, compositionally complex, and require computational methods that are still maturing. The risk of false discovery — finding apparent spatial patterns that do not replicate — is real, particularly in studies without adequate statistical power or in datasets where tissue processing artifacts mimic biological signals. The best companies we have seen treat spatial data the way serious clinical researchers treat exploratory genomics: with pre-specified hypotheses, independent replication cohorts, and honest acknowledgment of what the data cannot answer.
The technology is genuinely powerful. The biology it reveals is genuinely important. The discipline required to use it rigorously is not universal. That gap — between the technology's capability and the rigor with which it is applied — is where we do our diligence work.
