Multiplex immunofluorescence (IF) can provide rich spatial and phenotypic information for characterizing the tumor microenvironment – but it comes with practical limitations. Dependent on specialized instrumentation and expertise, the technique is too costly and labor-intensive for many labs. Downstream image analysis often requires substantial computational pathology support, and the technique can be affected by technical variability between runs. Together, these factors can limit scalability and broader adoption.
To counter these barriers, computer models are being developed to generate biomarker staining virtually. We connected with Christopher Jackson, Chief Scientific Officer at ViewsML, to learn about one such system, presented at USCAP 2026.
How might virtual staining address barriers associated with multiplex IF?
Our approach uses AI to generate virtual biomarker staining from routine H&E slides. Because the model is trained using clinical-grade IHC labels paired to H&E images, it's a more scalable way to extract biomarker-related information without requiring the full wet lab workflow for each new case. This can reduce cost, shorten turnaround time, save precious tissue, and expand access to tumor microenvironment profiling, including on archived H&E material.
Briefly, how does virtual multiplex IF work?
We use a paired H&E and IHC image training strategy to develop virtual biomarker models. Upon deployment, we can then predict per-cell biomarker spatial and expression levels all from a single H&E whole-slide image. Multiple biomarker-specific models can then be applied to the same H&E image, and the outputs can be combined into a composite virtual multiplex-style image for visualization of different tissue compartments and cell populations.
What did you investigate in the study presented at USCAP 2026?
We evaluated a virtual biomarker panel in non-small cell lung cancer (NSCLC) using markers representing tumor, stromal, vascular, and immune compartment biomarkers – CD3, CD31, CD45, CD68, SMA, and pan-cytokeratin. We assessed both the qualitative appearance of the virtual staining patterns and quantitative per-cell performance, since reliable cell-level signal is important for future downstream analyses.
What were the key results?
The models demonstrated strong per-cell discrimination and correlation, with per-cell AUCs ranging from 0.90 to 0.93 and per-cell Pearson correlations exceeding 0.70, where 1.0 would represent perfect performance for both metrics. In addition, board-certified anatomic pathologists at ViewsML reviewed the virtual stains and found that they highlighted the expected cell populations and tissue compartments. Taken together, these results support the feasibility of generating biologically meaningful virtual biomarker images from routine H&E whole-slide images.
How did the system perform in comparison to multiplex IF from routine H&E whole-slide images?
We compared the virtual outputs against paired, same-slide physical IHC-based reference labels. In that setting, the models achieved per-cell AUCs greater than 0.90, reflecting strong accuracy in identifying positive versus negative cells, and Pearson correlations greater than 0.70, indicating agreement in predicted expression intensity. Importantly, because these predictions are made at the individual cell level within intact tissue architecture, they preserve the spatial context of biomarker expression. The resulting staining patterns were also consistent with the expected underlying biology.
What are the potential implications of this work for NSCLC diagnostics?
We are building the world's first virtual biomarker library at scale that confers spatial and expression predictions at the per-cell level. By enabling virtual biomarker staining for tissue characterization directly from routine H&E slides, this approach could make tumor microenvironment analysis more practical, scalable, and accessible in NSCLC.
Potential applications include translational research, biomarker discovery, and clinical trials, where spatial characterization of tumor and immune biology across large cohorts could be performed while simultaneously preserving tissue, saving cost and time. In the longer term, this may help lay the groundwork for future precision oncology workflows built on routine histology, utilizing virtual biomarker staining assays as a screening tool prior to further downstream physical testing.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
