Summit Therapeutics Inc. (NASDAQ: SMMT) and South Korean medical artificial intelligence developer Lunit Inc. (KOSDAQ: 328130) announced a research agreement on October 23, 2026, to evaluate Lunit SCOPE IO for analysing tumor microenvironment characteristics in Summit’s investigational ivonescimab oncology development programme. The collaboration will use artificial intelligence to examine routinely prepared tissue images and investigate potential biomarkers associated with treatment response, although the companies have not announced clinical results or a validated patient-selection test arising from the agreement.
The research is intended to support a more detailed understanding of the biological environment surrounding cancer cells, particularly the distribution and spatial relationships of immune cells, tumor cells and surrounding tissue. These features may offer information about how tumors interact with immunotherapies and whether biological differences among patients can help explain variations in treatment outcomes.
The timing is significant because Summit is advancing ivonescimab through multiple late-stage clinical programmes, including an ongoing United States regulatory review supported by its Phase 3 HARMONi study. However, the Lunit collaboration is a separate translational research initiative and does not represent an additional regulatory submission, a new ivonescimab efficacy result or a change in the drug’s approval status. Its value will depend on whether computational pathology measurements can be translated into reproducible and clinically useful biomarkers.
How will Lunit SCOPE IO analyse tumor tissue to support Summit Therapeutics’ ivonescimab development?
The collaboration centres on hematoxylin and eosin-stained whole-slide images, a conventional form of tissue preparation widely used in pathology laboratories. Hematoxylin and eosin staining allows pathologists to examine tissue architecture, cellular morphology and relationships between cancer cells and surrounding structures. Digitising these tissue sections enables computational systems to analyse spatial and cellular patterns at a scale that would be difficult to reproduce consistently through manual assessment alone.
Lunit SCOPE IO uses machine learning to characterise aspects of the tumor microenvironment, including immune-cell infiltration and the spatial organisation of different cell populations. Such analysis can distinguish broad immune patterns, including tumors with immune cells distributed throughout tumor tissue and those where immune cells are concentrated outside the tumor compartment. The platform can also quantify selected features associated with the surrounding stromal and vascular environment.
For ivonescimab, the commercial and scientific attraction is the possibility of extracting additional information from tissue material already collected during oncology studies. This could reduce dependence on some supplementary staining procedures for exploratory research and potentially make certain analyses more scalable across large patient cohorts. However, computational pathology does not remove the need for carefully controlled laboratory preparation, image quality assessment and appropriate validation against biological or clinical reference standards.
The agreement does not establish that Lunit SCOPE IO can independently measure all molecular pathways relevant to ivonescimab’s mechanism. Standard hematoxylin and eosin images provide morphological information rather than direct measurements of every protein, receptor or signalling pathway involved in treatment response. Where molecular conclusions are being investigated, complementary laboratory assays may remain necessary.
The companies expect the work to inform exploratory biomarker development and potentially future clinical planning. They have not disclosed the number of tissue samples to be analysed, the specific ivonescimab trials supplying those samples, the exact analytical endpoints or a timetable for reporting findings. These omissions are important when assessing how quickly the collaboration might produce evidence suitable for clinical development decisions.
Why does ivonescimab’s dual PD-1 and VEGF mechanism make tumor microenvironment research relevant?
Ivonescimab is a bispecific antibody designed to combine blockade of programmed cell death protein 1, or PD-1, with inhibition of vascular endothelial growth factor, or VEGF. PD-1 blockade is intended to support antitumor immune responses, while VEGF inhibition targets a pathway associated with angiogenesis and aspects of the tumor microenvironment. Bringing these activities together in one molecule represents a different development approach from administering separate checkpoint inhibitors and anti-angiogenic medicines.
The drug was developed by Akeso and is known as SMT112 in Summit’s licensed territories. Summit holds development and commercialisation rights across major markets including North America, South America, Europe, Japan, the Middle East and Africa, while Akeso has advanced the medicine in China and other territories outside Summit’s licensing arrangement. Ivonescimab received its first marketing authorisation in China in May 2024 but remains investigational and unapproved in the United States and Europe as of October 24, 2026.
The biological rationale for the Lunit partnership follows from the complexity of combining immune checkpoint and angiogenesis-directed activity. VEGF-related signalling can influence vascular architecture, immune-cell trafficking and features of the local tumor environment, while the effectiveness of PD-1 blockade can depend partly on whether functional immune cells are present and able to interact with cancer cells. The contribution of these mechanisms may vary across tumor types, treatment histories and individual patients.
By examining spatial patterns within tumor tissue, Summit could investigate whether particular immune or vascular characteristics correlate with ivonescimab treatment outcomes. Such information might be especially relevant as the programme expands across lung cancer, colorectal cancer and urothelial carcinoma, where the biological and therapeutic contexts differ. A biomarker associated with response in one setting cannot automatically be assumed to perform similarly across these distinct malignancies.
Importantly, the ability to identify immune-cell patterns is not the same as demonstrating that those patterns predict the incremental benefit of ivonescimab. A clinically useful predictive biomarker must do more than identify patients with generally better prognosis or greater responsiveness to immunotherapy. It should provide reliable information about differential treatment benefit and ideally inform a decision that improves patient management.
What existing scientific evidence supports Lunit SCOPE IO, and what remains unvalidated for ivonescimab?
Lunit has previously investigated its SCOPE IO platform in several cancer research settings, providing a foundation for the new collaboration. A retrospective multicentre study published in the Journal for ImmunoTherapy of Cancer in 2024 examined artificial intelligence-defined immune phenotypes in 1,806 patients who had received immune checkpoint inhibitors across more than 27 solid tumor types. The researchers reported associations between an inflamed immune phenotype and more favourable treatment outcomes.
In that analysis, patients classified as having an inflamed immune phenotype recorded an objective response rate of 26.3%, compared with 15.8% among patients without that classification. Median progression-free survival was 5.3 months versus 3.1 months, while median overall survival was 25.3 months versus 13.6 months. These retrospective findings support the potential relevance of image-derived immune patterns, but they do not independently establish that the platform can select patients who will derive greater benefit from ivonescimab.
The distinction between prognostic and predictive information is particularly important when evaluating such findings. A tumor with substantial immune infiltration may be associated with better outcomes among patients receiving checkpoint inhibitors, but confirming treatment-specific predictive value requires an appropriate comparison that can separate the effect of the treatment from other clinical and biological factors. Retrospective association alone is insufficient to demonstrate that using the test to guide therapy improves outcomes.
More recently, Lunit reported additional research at the 2026 World Conference on Lung Cancer examining relationships between tumor microenvironment characteristics and epidermal growth factor receptor mutation subtypes. One analysis used 494 hematoxylin and eosin-stained whole-slide images from EGFR-mutated non-small cell lung cancer cases and identified differences in immune-cell and other cellular characteristics across mutation groups. This work is relevant to translational research in lung cancer, but it should not be treated as clinical validation of a biomarker for ivonescimab.
The new collaboration must therefore move beyond demonstrating that Lunit SCOPE IO can quantify identifiable tissue features. Stronger evidence would involve linking prespecified image-derived measurements with ivonescimab outcomes in appropriately defined patient cohorts, testing the findings in independent datasets and evaluating whether the associations remain robust after accounting for established clinical and molecular variables.
How could the partnership support the wider HARMONi clinical development programme?
Summit’s ivonescimab programme already includes several Phase 3 studies evaluating different treatment combinations and patient populations. Its global HARMONi trial investigated ivonescimab plus platinum-doublet chemotherapy against placebo plus chemotherapy in patients with epidermal growth factor receptor-mutated, locally advanced or metastatic non-squamous non-small cell lung cancer whose disease had progressed following a third-generation EGFR tyrosine kinase inhibitor. The study demonstrated a statistically significant progression-free survival benefit, with primary results subsequently published in The Lancet Oncology.
Overall survival has required more nuanced interpretation. At the prespecified primary analysis, Summit reported median overall survival of 16.8 months with ivonescimab plus chemotherapy and 14.0 months with chemotherapy alone, producing a hazard ratio of 0.79 and a p-value of 0.057. The primary overall survival analysis therefore did not achieve statistical significance, although later follow-up analyses reported nominally significant differences.
At the September 2026 World Conference on Lung Cancer, Summit presented an updated global overall survival hazard ratio of 0.76, with a nominal p-value of 0.0151, using longer follow-up for western patients. These later findings are supportive and clinically relevant, but their nominal significance should not be confused with a successful prespecified primary overall survival analysis. The progression-free survival result and subsequent survival evidence have distinct statistical interpretations that remain relevant to regulatory assessment.
The United States Food and Drug Administration accepted Summit’s ivonescimab biologics license application for review in January 2026, with a target action date of November 14, 2026. This represents a substantially more immediate regulatory milestone than any potential outcome from the Lunit research agreement. The pathology collaboration should not be interpreted as changing the pending application or indicating how the regulator will decide.
The wider development programme includes HARMONi-3, evaluating ivonescimab-based treatment against pembrolizumab-based combinations in first-line metastatic non-small cell lung cancer, and HARMONi-7, evaluating ivonescimab against pembrolizumab in patients with high PD-L1 expression. Additional development extends to colorectal and urothelial cancers, reflecting a strategy to investigate the bispecific antibody across several clinically distinct settings.
These studies could eventually provide opportunities for biomarker investigation, particularly if suitable baseline tumor samples and outcome data are available. Yet the companies have not identified the specific trial populations or tissue collections included in the Lunit agreement. Any future use of the resulting biomarker hypotheses would need to respect the study designs, statistical plans and clinical evidence requirements applicable to the individual programmes.
What validation and data-quality requirements could determine whether AI-derived pathology biomarkers are clinically useful?
One important challenge is whether computational measurements remain reproducible when tissue is collected, prepared, stained and scanned under different conditions. Large multinational oncology programmes typically involve numerous hospitals, pathology laboratories and imaging systems, each of which may introduce variation. A biomarker that performs well in a selected research dataset may show reduced reliability when applied to tissue images from other institutions.
External validation will therefore be important in determining whether any biomarker emerging from the collaboration has broader relevance. The companies would need to assess potential differences associated with image quality, tumor type, specimen source, patient demographics and clinical setting. Where a biomarker is intended to inform treatment selection, analytical reproducibility must also be connected to evidence of meaningful clinical utility.
Another consideration is the timing of tissue collection relative to treatment. A pretreatment biopsy might provide information about baseline immune characteristics, while tissue collected following therapy could reflect biological changes caused by treatment itself. Mixing these sample types without appropriate analytical controls could complicate attempts to distinguish predictive markers from treatment-associated effects.
The research will also need to consider how cellular composition interacts with established oncology measurements, including PD-L1 expression, genomic alterations and other relevant pathological features. Combining these measurements could potentially improve patient stratification, but more complex models create additional validation requirements and opportunities for overfitting. Any improvement over simpler established biomarkers would need to be demonstrated rather than assumed.
Lunit explicitly identifies SCOPE products as research-use-only technologies that are not intended for diagnostic procedures or clinical decisions. Consequently, the current agreement does not authorise the platform to select patients for ivonescimab treatment in routine care. If a defined assay ultimately advances toward a regulated companion diagnostic application, its intended use, analytical performance, clinical validation and applicable regulatory pathway would need separate evaluation.
What commercial value could Summit Therapeutics and Lunit obtain from the AI research agreement?
The partnership reflects increasing interest in using computational pathology as a component of pharmaceutical research rather than solely as a technology for routine diagnostic laboratories. For Summit, the potential value lies in obtaining more detailed information from clinical trial tissue samples and using that information to inform future research or development decisions. For Lunit, the agreement provides another opportunity to apply its oncology research platform in a late-stage drug development environment.
The companies have not disclosed the financial terms of the research agreement, including any upfront payments, service fees, milestone arrangements or potential downstream commercial rights. It would therefore be premature to assign a specific revenue contribution to Lunit or a defined financial benefit to Summit. The immediate development is a research engagement, not a confirmed companion diagnostic commercialisation agreement.
Summit ended the second quarter of 2026 with approximately $690.7 million in cash and investments, according to its July financial disclosure. Its principal near-term financial and strategic considerations remain linked to ivonescimab’s regulatory pathway, ongoing Phase 3 trials and preparations associated with potential commercialisation. A research agreement of undisclosed value is unlikely to displace those programme-level factors in determining the company’s immediate investment outlook.
For Lunit, demonstrating that its image-analysis technology can contribute to biopharmaceutical biomarker research could support further partnerships with oncology developers. However, adoption by pharmaceutical companies will depend on more than algorithmic performance, including integration with trial data systems, specimen availability, contractual arrangements, regulatory expectations and the ability to reproduce findings across different patient populations. Research relationships may generate revenue or future opportunities, but those outcomes cannot be inferred from the announcement alone.
The companies’ latest share-price performance also illustrates the difference between a new research announcement and the broader expectations surrounding publicly traded healthcare businesses. Summit Therapeutics shares closed at approximately $15.46 on October 23, 2026, down about 0.6% for the session but above their October 16 level. Lunit shares closed at 11,300 South Korean won, modestly higher than the previous session, after a volatile trading week.
Neither movement establishes a direct market response to the research agreement. Summit’s valuation remains highly sensitive to ivonescimab’s clinical and regulatory prospects, while Lunit’s investment outlook includes a broader medical artificial intelligence and oncology technology business. The collaboration is best viewed as a potential longer-term research capability rather than immediate evidence of additional treatment efficacy or material commercial revenue.
Can AI tumor microenvironment analysis eventually change ivonescimab patient selection?
The next meaningful evidence from the partnership would be a clearly defined analysis connecting specific Lunit SCOPE IO measurements with treatment response or other clinical outcomes in ivonescimab-treated patients. Ideally, such research would identify a biologically plausible relationship, quantify its statistical strength and evaluate whether the association can be reproduced in independent samples. The analysis would also need to demonstrate how the proposed biomarker performs relative to established clinical and pathological variables.
Subsequent development could involve a prespecified biomarker hypothesis within a clinical trial or an independently validated retrospective analysis of randomised trial samples. Whether either approach would be sufficient for regulatory use would depend on the proposed claim, the strength of the supporting evidence and the intended clinical application. There is currently no announced timeline or confirmed regulatory development plan for an ivonescimab-related diagnostic emerging from this collaboration.
Summit’s immediate clinical programme and the Lunit pathology initiative operate on different timescales. The former includes an active regulatory review and ongoing late-stage efficacy studies, while the latter is intended to deepen scientific understanding and potentially support future research. The real test of the agreement will be whether AI-derived tumor microenvironment features can identify meaningful differences in ivonescimab treatment benefit, rather than simply describe biological variation already visible in tissue specimens.
