ASCO 2026: The AI and Spatial Profiling Revolution in Cancer Biomarkers

Not Just Whether a Target Exists, but Where It Is in the Tumor: Biomarkers Are No Longer Simply “Positive” or “Negative”

The old biomarker question: Is the target present?

Modern oncology has long since entered an era in which biomarkers change treatment decisions. Markers such as EGFR, ALK, ROS1, and RET in lung cancer; RAS, BRAF, MSI, and HER2 in colorectal cancer; ER, PR, HER2, and Ki-67 in breast cancer; and BRCA and HRD in ovarian cancer are now integral to everyday practice.

However, the traditional biomarker approach has an important limitation: it often summarizes the tumor with a single average value. For example, we call a tumor “HER2 positive,” but often do not measure in detail whether HER2 is the same in every region of the tumor, where positive cells are concentrated, whether immune cells can enter these HER2-high areas, or whether the stroma surrounds these cells.

The new question: Where is the target, which cells are beside it, and in what microenvironment?

This is where artificial intelligence and spatial profiling come in. Digital pathology analyzes the whole-slide image at high resolution. AI models can quantify tumor cells, stroma, lymphocytes, vascular cells, areas of necrosis, immune phenotype, and cellular neighborhoods. The pathology slide can thus become more than a diagnostic document: it may turn into a multilayered biological map capable of predicting treatment response.

An example from ASCO 2026: AI-supported selection in HER2-positive biliary tract cancer

One striking example of this approach at ASCO 2026 was a study in HER2-positive advanced biliary tract cancer. While evaluating a first-line quadruple regimen containing trastuzumab, nivolumab, gemcitabine, and cisplatin, the study examined HER2 expression and immune phenotype together using AI-supported whole-slide analysis.

The multicenter phase Ib/II study included 40 patients . With the treatment combination, the objective response rate was reported as %55, the disease control rate as %95, and median progression-free survival as 10,6 months . The key point, however, was that AI could distinguish which patients might have a higher likelihood of response.

This example shows that AI could go beyond being a simple “tool to assist pathology.” When the amount of target protein, immune phenotype, and tumor microenvironment are analyzed together in the same tissue, patient selection may become more precise.

An immune-cold tumor microenvironment in HER2-high lung cancer

Another analysis presented by Lunit at ASCO 2026 assessed the tumor microenvironment in non-small cell lung cancers with HER2 overexpression. AI-supported spatial analysis of more than two thousand NSCLC whole-slide images reported lower tumor-infiltrating lymphocyte density and a lower proportion of an inflamed immune phenotype in tumors overexpressing HER2.

This finding suggests that HER2-high tumors may have a more “immune-cold” microenvironment in certain contexts. If these results are validated in independent series, it may be necessary to examine not only HER2 level but also the microenvironment map to understand which patients might benefit more meaningfully from combinations of HER2-targeted therapy and immunotherapy.

A spatial signature distinguishing axitinib response in adenoid cystic carcinoma

Another example of AI-supported spatial profiling was presented in adenoid cystic carcinoma. Treatment options for this rare tumor are limited, and predicting response with traditional biomarkers is difficult. AI-based tumor microenvironment analysis reported longer PFS after axitinib in a subgroup with high endothelial-cell and tumor-infiltrating lymphocyte densities: 19,6 months versus 11,1 months.

TLS regions in MSS colorectal cancer: Could a door to immunotherapy open?

Microsatellite-stable metastatic colorectal cancer is generally considered resistant to immune checkpoint inhibitors. However, the “MSS” label does not explain all the details of the tumor microenvironment. One AI-supported tumor microenvironment analysis presented at ASCO 2026 showed that larger regions of tertiary lymphoid structures in MSS metastatic colorectal cancer may be associated with better PFS and OS after immunotherapy.

This area is very important: when we place certain tumors into a single “immunotherapy-resistant” category, we may miss small subgroups whose microenvironment actually has the potential for an immune response. AI-supported spatial analysis may provide an opportunity to define these small subgroups more objectively.

Multimodal AI: When pathology, clinical data, and genomic risk enter the same model

Another important direction at ASCO 2026 was multimodal artificial intelligence. Multimodal AI considers not only the pathology image, but also clinical data, age, stage, tumor size, nodal status, treatment information, and genomic data when needed. The aim is to model the entire disease context rather than a single data source, producing more accurate predictions of risk and treatment benefit.

Studies presented by Artera at ASCO 2026 exemplify this direction. A multimodal AI model assessing chemotherapy benefit and prognosis in node-positive HR-positive breast cancer was presented as Abstract #107 in an oral session. Studies on agreement between multimodal AI and the 21-gene Recurrence Score test in early-stage breast cancer, as well as digital-pathology-based risk stratification in prostate cancer, were also presented.

What could these technologies change in the clinic?

The greatest potential of AI and spatial profiling is to make existing biomarkers more precise. For example, HER2 positivity may no longer be assessed only as “present” or “absent,” but together with the proportion of HER2-high cells, their distribution within the tumor, the presence of neighboring immune cells, and the tumor microenvironment context.

Similarly, immunotherapy decisions may incorporate not only PD-L1 or MSI, but many variables together: whether T cells enter the tumor or remain at its outer edge, TLS presence, the stromal barrier, vascular structure, and macrophage / TIL composition.

Critical reading: Validation is as necessary as enthusiasm

A message for Türkiye: Digital pathology and data infrastructure are strategically important

Molecular oncology, NGS panels, and targeted-treatment decisions are increasingly prominent in Türkiye. However, the era of AI and spatial profiling requires not only molecular testing infrastructure but digital pathology infrastructure as well. High-quality tissue processing, standardized staining, high-resolution scanners, secure data storage, pathology-oncology integration, and clinical outcome data are fundamental components of this system.

Therefore, the concept of a “digital pathology and biomarker board” may become as important as a “molecular tumor board” in the coming period. AI models become meaningful only when combined with the right clinical question, high-quality data, and expert interpretation.

Conclusion

ASCO 2026 showed that artificial intelligence and spatial profiling are becoming increasingly visible in oncology. AI-supported HER2/IHC measurement, tumor microenvironment analysis, TIL/TLS mapping, multimodal risk models, and digital-pathology-based prediction of treatment benefit point to a new era in which biomarkers will be interpreted not simply as “positive” or “negative,” but within a multilayered tumor ecosystem.

These technologies are not yet standard in every field, but the direction is clear. The future question in cancer treatment will not simply be “Is there a target in the tumor?” A more precise question will be: “Where in the tumor is this target, alongside which cells, in what microenvironment, and does this map predict the patient’s treatment response?”