The widespread adoption of spatial transcriptomics begins when spatial data changes consequential decisions.
Anyone encountering spatial transcriptomics results for the first time is likely to be impressed. The locations of cancer cells and immune cells become visible within tissue, revealing how distinct biological states coexist within the same tumor. Spatial context lost in conventional bulk or single-cell analyses is restored.
Yet in pharmaceutical R&D meetings, the question that ultimately remains is usually simple:
“What decision should we make differently based on these results?”
“Seeing more” clearly has scientific value. But the business value of “making better decisions” requires separate proof. For spatial transcriptomics to become a tool used repeatedly, it must go beyond revealing new phenomena and reduce uncertainty in choices where failure is costly.
Good Applications Start with a Decision
Spatial transcriptomics can inform a wide range of decisions in the pharmaceutical and biotechnology industries. It can help assess which targets are highly expressed in cancer cells but minimally expressed in normal tissue, which antibodies may penetrate deep into a tumor, and whether an antibody–drug conjugate’s (ADC’s) linker and payload are suited to the tumor microenvironment. In radiopharmaceutical therapy (RPT), it can support the joint assessment of target distribution and potential toxicity to normal organs. For T-cell engagers (TCEs), it can help examine the distance between target cells and T cells, as well as their potential for physical contact.
Ultimately, these applications can extend to pharmacokinetics (PK)—including how far a drug reaches within tissue—and pharmacodynamics (PD), which concerns the effects it produces, as well as toxicity assessment and patient selection. The buyer landscape is correspondingly complex. Research leaders selecting targets, development teams comparing drug candidates, and organizations designing clinical trials each have different questions and budgets.
For major technology companies, another application lies in foundation models that integrate vision, gene expression, and language. By jointly learning from tissue images, cell locations, gene expression, and natural-language knowledge, these models could provide a foundation for biological AI. However, data volume alone does not create a good model. Diversity in training data, clinical context, label quality, and rights to use the data must also be secured.
From a government perspective, important applications may include maps of cancer and immune diseases, three-dimensional human atlases, biosecurity, and national biological data infrastructure. Public projects prioritize data sovereignty and long-term research capacity over the profitability of individual products. However, these initiatives often take years to build. Without robust systems for maintenance and use after a project ends, even vast datasets may fail to translate into practical applications.
The Same Technology Serves Different Markets
Different applications come with different buyers, usage frequencies, existing alternatives, and willingness to pay.
For a one-off mechanistic study, a project-based analysis service may be appropriate. Repeated comparisons of drug candidates or target assessments across multiple cancer types, however, call for a platform that can be reused as new data becomes available. The former depends on expertise and tailored interpretation; the latter requires standardized workflows, speed, comparability, and reproducibility.
Growth in revenue from research services alone is therefore insufficient evidence that mass adoption has begun. The real transition occurs when customers repeatedly make decisions using existing infrastructure, rather than designing a new project from scratch every time a new question arises.
This does not mean every analysis should become a platform. Expert-led, customized research may be better suited to rare cancers, new therapeutic modalities, or early exploratory work, where the questions themselves are not yet well defined. Services and platforms are best understood as a continuum from discovering new questions to translating them into repeatable workflows, rather than as substitutes for one another.
From Data to Virtual Experiments
At Portrai, our experience in the field has likewise taught us that the structure of a decision matters more than the volume of data. Even when the same target is selected, actual drug delivery can vary depending on an antibody’s binding properties, the conditions under which a linker is cleaved, the potency and diffusibility of the payload, and the spatial arrangement of blood vessels and target cells.
To better understand these relationships, we are connecting human-derived spatial transcriptomics data with tissue-level microscopic PK/PD models. The goal is not to declare that a particular drug will succeed. It is to compare how delivery and response within tissue may change when the target, binding affinity, linker, payload, or drug-to-antibody ratio (DAR) is varied while other conditions are held constant—and thereby narrow down which candidates should be prioritized for experimental testing. (Reference: https://link.portrai.io)
This approach also has limitations. A model’s reliability is constrained by the quality of its input data and assumptions. If factors such as blood flow, protein expression, and drug metabolism are not adequately represented, its predictions may diverge from actual in vivo outcomes. Simulation begins as a tool for choosing better experiments. By progressively narrowing the range of conditions that need to be tested, it can help guide us toward better outcomes.
Three Possibilities for the Next Five Years
In an optimistic scenario, spatial data will become embedded in drug candidate design and patient selection, allowing tissue-based virtual experiments to reduce trial and error in drug development.
More realistically, repeated use is likely to emerge first in specific areas rather than across the entire industry—particularly ADCs, RPTs, and TCEs, where target location and the composition of surrounding cells directly affect efficacy and toxicity.
Conversely, if the value of these applications cannot be demonstrated quantitatively, spatial transcriptomics may remain an advanced research tool for a limited group of specialists, even as the volume of data and images continues to grow.
Mass adoption begins when spatial data changes a decision, not when it produces a more impressive image.
Building applications that change decisions requires high-quality tissue from diverse patients, together with clinical context, before algorithms come into play. Organizations that systematically build this foundation will gain a competitive advantage. In the next article, we will explore why tissue and cohorts—the scarcest raw materials in spatial transcriptomics—are about more than simply collecting samples.
