Pathology practice has expanded well beyond morphologic diagnosis alone. Pathologists now guide ancillary testing, interpret predictive and prognostic biomarkers, integrate molecular findings, steward limited tissue, and support increasingly time-sensitive treatment decisions. Unfortunately, the workflows underpinning those responsibilities haven’t evolved at the same pace.
For decades, pathology has operated under a sequential, and often limiting, model. Each new question requires another stain, another section, another step. It’s a system that has delivered rigor and reliability, but one that is increasingly misaligned with the realities of modern medicine — where clinicians, researchers, and drug developers are all requiring more from every sample.
Today, that model is beginning to shift. With advances in computational pathology, the H&E slide — long the foundation of diagnosis — is being reimagined as a far richer source of information. Virtual staining is at the forefront of that shift. It doesn’t replace the expertise of the pathologist or the role of established techniques like immunohistochemistry (IHC). Instead, it introduces a new layer: a digital, computational approach that sits above existing workflows, enabling faster, non-destructive, and more scalable biomarker analysis.
The structural limits of physical IHC
IHC remains a gold standard for biomarker detection for good reason. It provides direct, interpretable evidence of protein expression. But it also carries structural limitations. Turnaround times can stretch from days to weeks, particularly as panels grow more complex. Each additional stain consumes tissue, which is an increasingly critical constraint in small biopsies, rare samples, and cases where tissue needs to be preserved for molecular testing. Human interpretation of these stains, while grounded in expertise, can introduce inter- and intra-observer variability that ultimately shapes downstream decisions, from treatment selection to clinical trial eligibility.
The result of these limitations means that clinicians wait, sometimes several days, for results that determine care pathways. Researchers work within the confines of limited or irreplaceable samples and are often forced to prioritize which questions can be asked. Clinical trials slow as patient screening and stratification become bottlenecks.
The challenge is no longer simply establishing whether a marker is present or absent. Increasingly, the value lies in understanding how multiple biomarkers map onto morphology, tumor architecture, immune infiltration, and stromal response within the same tissue section. Physical IHC can provide these answers, but scaling that approach across many markers, samples, and studies often requires additional tissue, slides, time, and cost.
The promise and limits of the first wave of AI
Artificial intelligence (AI) has already begun to reshape pathology, particularly through models that predict biomarker status directly from H&E whole slide images. These approaches have shown real promise, offering a path to faster, more accessible insights. But much of the early work in this area has focused on case- or slide-level prediction: a probability, score, or classification that summarizes the specimen as a whole.
That type of output can be useful, but it is not the same as creating an interpretable biomarker map. Many models encode tissue into abstract feature representations and use those representations to infer biomarker status. While powerful, these approaches can compress heterogeneous biology into a single endpoint. Such a model may suggest that a marker is likely to be present, but it does not show where the signal arises, how it varies across tumor and stroma, or how expression patterns relate to morphology, architecture, and immune context.
From prediction to representation
Virtual staining delivers the necessary granularity, shifting the output from a slide-level prediction to a spatially resolved representation of biomarker expression. Rather than producing a single prediction for a slide, virtual biomarker staining can operate at the individual cell level. Using computational models applied directly to H&E images, it generates spatially resolved maps of biomarker expression across the tissue, effectively recreating staining patterns without the need for additional physical processes. This enables pathologists to identify not only whether a biomarker is present, but where it is expressed, in which cells, and in what context.
This approach aligns more closely with how pathologists already interpret tissue, which is through morphology, spatial relationships, and cellular architecture. By grounding outputs in the structure of the tissue itself, virtual staining preserves interpretability while extending the depth of insight.
When done optimally, virtual staining of biomarkers differentiates itself from earlier AI approaches in key ways:
Outputs can be spatially resolved at the per-cell level, not aggregated at the case level.
Insights remain anchored in underlying tissue morphology.
Virtual biomarkers can be combined in virtual multiplexed panels to predict spatial relationships.
Rather than abstracting away the biology, virtual staining makes biomarker information more visible, localized, and scalable.
Scaling insight, sparing tissue
From a single H&E slide, multiple biomarkers can be derived computationally, which means insight can scale without consuming additional tissue, chemicals, or time. For clinicians, this creates flexibility, enabling earlier triaging decisions and reducing reliance on sequential testing. For researchers and drug developers, it opens access to data that might otherwise be lost due to tissue constraints.
In the context of the tumor microenvironment, virtual staining becomes particularly important. Insights into spatial biology traditionally required complex physical multiplexed assays or extensive IHC panels. Now, spatial relationships between immune cells, tumor cells, and stromal components can be mapped virtually, and analyzed at scale across large cohorts. This can enable earlier and more informed patient triaging and a reduction in the number of cases that ultimately require physical testing. Clinical trials can benefit from more efficient patient stratification.
Virtual staining shifts the operational process as well, compressing turnaround times from days to minutes and decreasing per-biomarker costs. Biomarker analysis becomes scalable across larger populations, not just select cohorts.
A different way to think about the slide
Virtual staining transforms the H&E slide from a starting point for downstream testing to a platform for layered insight, where computation can extract, augment, and scale information without altering the underlying specimen. It doesn’t diminish the role of traditional methods like IHC, but it does change when and how those methods are used, shifting them from default steps to targeted, confirmatory tools. More importantly, it repositions pathology workflows from being constrained by tissue and time to being enabled by data and computation.
Pathology has always balanced art and science, offering interpretation grounded in evidence, and insight shaped by experience. Virtual staining extends that balance. By enabling faster, more scalable biomarker analysis that preserves tissue samples, it addresses some of the most persistent bottlenecks in both clinical care and research. As the field continues to move toward more spatial, quantitative, and integrated models of disease, that added resolution will become increasingly important.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
