What determines the scalability of spatial transcriptomics experiments is not the best-case outcome, but whether consistent results can be achieved regardless of who performs them.
Researchers who place precious patient tissue on a slide often remain on edge until the experiment is complete. If the tissue section is damaged, preprocessing conditions vary even slightly, or an unexpected issue arises during staining or imaging, they may have to start over.
With ordinary reagents, the experiment can simply be repeated. But a biopsy collected only once at a specific treatment time point or tissue from a rare cancer cannot be obtained again. The cost of failure in a spatial transcriptomics experiment includes not only extremely expensive reagents and equipment fees, but also an irrecoverable clinical opportunity.
The key obstacle to mass adoption, therefore, is not merely the average success rate. It is inter-experiment variability and dependence on skilled operators.
Not a Single Experiment, but an Interconnected Process
Spatial transcriptomics is not completed with a single measurement. It consists of a continuous sequence of steps: tissue selection and sectioning, preprocessing such as fixation and permeabilization, staining, imaging, probe reactions, library preparation, sequencing, and data QC.
Small variations at each stage can be amplified in subsequent stages. Poor tissue quality may be compensated for with greater sequencing depth, or incomplete staining may be corrected using analytical algorithms. This may produce data, but it may not restore biological reliability.
On the surface, this appears to be a problem with the quality of the final output. Underneath, however, lies a structural issue: it is difficult to assess the state of the process at intermediate stages. If a failure is discovered only at the end, it is also difficult to trace which step caused the problem.
Experienced researchers assess tissue conditions and subtle changes based on their accumulated experience. This expertise is extremely valuable, but a process that depends on the tacit knowledge of specific individuals is difficult to scale or transfer to new institutions. If the distribution of results changes whenever the person in charge changes, the work may remain viable as a research project but cannot become an industrial production system. And this is the point at which industrialization is needed.
Automation Is About More Than Reducing Labor Costs
Understanding laboratory automation solely as robots replacing human hands underestimates its value. The essence of automation lies not in repeating the same actions, but in measuring, recording, and controlling variability.
The first benefit is reproducibility. Consistent management of reagent volumes, reaction times, temperatures, and imaging conditions can reduce variation across operators and institutions.
The second is provenance: a traceable record of the conditions and processes through which each result was generated. When a problem occurs, it must be possible to trace its cause so that results can be trusted and compared.
The third is throughput. This means not only processing more slides, but also delivering results within a predictable timeframe.
The final benefit is workforce sustainability. Reducing repetitive, high-pressure tasks allows researchers to focus more on assessing exceptional cases, designing better experiments, and improving scalability.
Automation should not be a technology that eliminates experts, but an operating model that deploys expert judgment where it is needed most.
What We Need Is Not Robots, but Closed-Loop Quality Control
In the short term, QC gates should be established at every major stage of the experiment. Assessing tissue condition, staining quality, signal intensity, and background noise before moving to the next stage can help identify failures earlier. Operational metrics should include not only failure rates, but also repeat rates, batch-to-batch variation, and turnaround time.
In the medium term, standardized SOPs should be combined with robotic automation, along with sensors and software that monitor temperature, humidity, reaction times, and imaging conditions. Recording metadata that is not tied to a specific instrument and including the same control tissue in every batch can help distinguish whether variation originates from the tissue or the process.
In the long term, a closed-loop system is needed in which experimental results feed back into the conditions of subsequent experiments. However, rather than automatically discarding tissue that fails to meet quality standards, a human-in-the-loop structure that flags exceptions for expert review is more realistic.
Automation cannot solve every problem. Fatty or highly fibrotic tissue, tumors containing necrotic regions, and small biopsies may each require different handling. For rare specimens, rigid adherence to standardized procedures may even compromise the information they contain. It is therefore necessary to distinguish which steps should be automated and which should remain subject to expert judgment.
What Portrai Has Learned
As Portrai built a wide range of spatial transcriptomics datasets, we found that many of the anomalies detected during analysis were not independent of the experimental process. Even within the same cancer type, the distributions of detected cells and genes could vary depending on tissue condition, preprocessing, and imaging quality.
Analytical QC alone was therefore insufficient. Pathology images, experimental metadata, and analytical results had to be connected so that anomalies could be traced back to the original process. This is why we view laboratory automation not merely as a data production technology, but as part of a reliable data infrastructure.
The Paradox After Automation Succeeds
Optimistically, within five years, standardized automation could reduce inter-institutional variation and transform spatial transcriptomics experiments from projects performed by a small number of highly skilled specialists into repeatable production processes.
Realistically, experts are likely to remain involved in tissue preparation and exception assessment, while highly repetitive steps such as staining, imaging, and library preparation are automated incrementally. Conversely, if automation increases equipment throughput without integrating QC and provenance, it may produce more batch effects along with more data.
And if automation truly succeeds, a new paradox will emerge. The moment the experimental bottleneck is removed, millions of cells and billions of gene-expression measurements will begin to pour in continuously.
In the next article, we will examine the explosion of spatial transcriptomics data that automation will create and the new computing infrastructure needed to handle it.
