BostonGene and Kyoto University have established a strategic research partnership that will apply BostonGene’s multimodal artificial intelligence platform to the NOBEL-ioPDT Phase II trial in advanced gastrointestinal cancers. The investigator-initiated study is evaluating immune checkpoint inhibitors combined with photodynamic therapy in patients with unresectable advanced or recurrent oesophageal and gastric cancer.
The announcement, made on July 22, 2026, introduces an extensive biomarker-discovery component rather than a new therapeutic candidate or a fresh clinical trial. BostonGene plans to integrate genomic, transcriptomic, immune and clinical data collected during the study to identify biological signatures associated with treatment response, resistance and potentially durable clinical benefit.
That distinction matters. No efficacy results, safety findings or validated biomarkers were disclosed with the partnership. The immediate value lies in creating a translational research layer around an existing Phase II programme, with the longer-term objective of determining whether molecular and immune characteristics can identify patients more likely to benefit from the experimental combination.
The collaboration also deepens an existing research relationship between BostonGene and Kyoto University. The organisations previously announced a separate biomarker project involving immune checkpoint inhibition and chemoradiotherapy in oesophageal squamous cell carcinoma. The latest initiative focuses on photodynamic therapy and covers advanced oesophageal and gastric cancers, making it a distinct research programme with different therapeutic and biomarker questions.
What will BostonGene analyse in the Phase II NOBEL-ioPDT gastrointestinal cancer trial?
The NOBEL-ioPDT study, registered as jRCT2051220176, is a multicentre, investigator-initiated Phase II trial led by Professor Manabu Muto of Kyoto University. Kyoto University’s public research information describes a planned enrolment of 80 patients, with case accumulation beginning in January 2023.
The programme includes patients preparing to receive first-line standard treatment containing an immune checkpoint inhibitor, as well as certain patients whose oesophageal or gastric cancer has progressed following immune checkpoint inhibitor monotherapy. The trial is assessing the safety and efficacy of adding photodynamic therapy to immune checkpoint inhibitor-based treatment, with response rate identified as a primary outcome for the respective study cohorts.
BostonGene will analyse molecular and clinical information produced through the trial. Its platform is intended to combine tumour genomics, gene-expression data, immune-system characteristics and clinical outcomes, allowing researchers to investigate relationships that may be missed when biomarkers are evaluated individually.
The analysis could examine whether particular tumour pathways, immune-cell populations, microenvironmental states or gene-expression patterns are more common among responders. It could also help researchers distinguish biological features associated with primary resistance from changes that emerge during or after treatment.
This remains a biomarker-discovery exercise. The partnership announcement did not identify a validated signature, provide performance measurements or state that a BostonGene-derived test will be used prospectively to assign treatment. Any signature emerging from the programme would require further validation before it could support routine clinical decisions.
Why are researchers combining photodynamic therapy with immune checkpoint inhibition?
Photodynamic therapy uses a light-activated photosensitising agent to damage targeted tumour tissue. Once the agent has accumulated in or around the tumour, light of an appropriate wavelength is delivered to the treatment area, triggering a photochemical reaction that can destroy cancer cells and affect local tumour vasculature.
The therapeutic rationale extends beyond direct local destruction. Photodynamic therapy may cause forms of tumour-cell damage capable of releasing antigens and inflammatory signals. Researchers are investigating whether this process can increase immune-cell infiltration or make an immunologically inactive tumour environment more responsive to checkpoint blockade.
Immune checkpoint inhibitors work by releasing inhibitory signals that restrain antitumour immune activity. However, removing that restraint may be insufficient when a tumour lacks meaningful immune-cell infiltration, contains strongly immunosuppressive features or has developed other mechanisms of resistance. Photodynamic therapy is therefore being studied as a possible way to alter the local tumour environment while immune checkpoint inhibition supports a broader antitumour response.
Preclinical research has supplied a biological rationale for this combination, including observations of increased immune-cell activity after photodynamic therapy. Such findings do not establish clinical benefit in advanced oesophageal or gastric cancer. The Phase II study must determine whether the proposed interaction produces responses of sufficient depth and duration without creating unacceptable combined toxicity.
The distinction between local and systemic effects is especially important in advanced disease. Photodynamic therapy is delivered to accessible tumour tissue, whereas unresectable or recurrent gastrointestinal cancer can involve disease beyond the illuminated lesion. Researchers will therefore need to establish whether local treatment contributes to meaningful disease control at untreated sites, rather than merely shrinking the directly treated tumour.
Can multiomic AI move the biomarker search beyond single immune-response signals?
Immune checkpoint inhibitor response is difficult to predict with one marker. Measurements such as programmed death-ligand 1 expression can be informative in certain indications, but their performance varies by cancer type, assay, treatment regimen and clinical setting. Tumour mutational burden, microsatellite instability and immune-cell infiltration also capture only parts of a much larger biological system.
BostonGene’s multiomic approach is designed to analyse these relationships at several levels. Genomic data can identify mutations and pathway alterations, transcriptomic information can reveal active biological programmes, and immune profiling can characterise immune-cell composition and functional states. Clinical data then provide the outcome context needed to investigate whether any of those features correlate with response or resistance.
This broader view could be particularly relevant for a treatment that combines systemic immunotherapy with a locally delivered procedure. The response may depend not only on the tumour’s baseline immune state, but also on whether photodynamic therapy induces sufficient local inflammation, antigen presentation and immune-cell recruitment.
Artificial intelligence can help evaluate large numbers of interactions across relatively complex datasets. It cannot compensate automatically for small sample sizes, missing specimens, differences between clinical sites or imbalances among trial cohorts. With a planned enrolment of 80 patients spread across different cancers and treatment settings, the number of evaluable patients within each molecular subgroup may be limited.
That creates a familiar precision-oncology problem. A model can identify an apparently strong pattern in a development cohort while performing much less effectively in an independent population. The most valuable output may therefore be a shortlist of biologically plausible signatures for further testing, rather than a treatment-selection tool ready for immediate clinical deployment.
What must happen before any discovered biomarker can guide gastrointestinal cancer treatment?
A candidate biomarker must first show analytical reliability. The underlying assays and computational pipeline need to produce consistent results across sample batches, collection sites and processing conditions. This is particularly important when tissue quality, tumour content and prior treatment can influence molecular measurements.
The biomarker must then undergo clinical validation in data that were not used to discover it. Researchers will need to determine whether it predicts benefit from the ICI-PDT combination specifically, rather than merely identifying patients with a generally favourable prognosis.
Prospective clinical utility represents a higher threshold. A signature may correlate with response yet still fail to improve treatment decisions when used in practice. Demonstrating utility would require evidence that biomarker-guided selection meaningfully changes outcomes, reduces exposure to ineffective treatment or improves the design of subsequent clinical trials.
Regulatory considerations would depend on how the resulting signature is used. An exploratory research finding has a different status from a clinical assay used to select treatment. Any future diagnostic application would require a clearly defined intended use, validated analytical performance and evidence supporting the relationship between the test result and the proposed clinical action.
No regulatory submission, companion diagnostic programme or commercial biomarker product was announced. The current partnership should therefore be viewed as an evidence-generation project supporting future development decisions.
What will determine whether the BostonGene partnership delivers commercial value?
BostonGene is privately held, so the announcement does not create a directly measurable public-market catalyst. Financial terms were not disclosed, and the collaboration should not be interpreted as near-term product revenue or commercial adoption of a new diagnostic.
Its strategic value rests on platform validation. If BostonGene can generate reproducible biomarkers from a clinically complex combination trial, the work could strengthen its position with pharmaceutical developers seeking patient-selection strategies for immunotherapy combinations. The ability to distinguish responders from non-responders could help sponsors refine enrolment criteria, prioritise indications and design more efficient follow-on studies.
The partnership also adds to BostonGene’s presence in Japan through BostonGene Japan, its joint venture with NEC Corporation and Japan Industrial Partners. Academic collaborations can help the company access clinically characterised samples and establish relationships within Japan’s oncology research ecosystem. Those advantages become commercially meaningful only if the resulting analyses influence development programmes or lead to validated testing services.
For Kyoto University and the NOBEL-ioPDT investigators, the biomarker layer could explain why patients with apparently similar clinical characteristics experience different outcomes. It may also provide evidence for adjusting treatment sequence, identifying resistance mechanisms or designing studies focused on molecularly defined populations.
Which results will show whether the ICI-PDT strategy is ready to advance?
The first test remains the Phase II clinical outcome. Investigators must establish whether the combination produces a credible response rate and acceptable safety profile in the individual oesophageal and gastric cancer cohorts. Duration of response, progression patterns and activity at lesions not directly treated with photodynamic therapy will also be important when judging whether the approach offers more than local tumour control.
Safety assessment must account for both components. Immune checkpoint inhibitors can cause immune-mediated toxicities, while photodynamic therapy involves a photosensitising drug, specialised light delivery and procedure-related risks. The combination must remain feasible across multiple centres, not only at institutions with extensive photodynamic therapy expertise.
The biomarker programme will face a separate test. BostonGene and Kyoto University will need to show that any proposed molecular signature is biologically coherent, statistically credible and reproducible in an independent dataset. Transparent reporting of sample availability, assay methods, missing data and model validation will be essential for interpreting the findings.
Until those results emerge, the partnership is best understood as a sophisticated attempt to learn from an ongoing Phase II trial, not evidence that AI-selected ICI-PDT treatment improves patient outcomes. Its importance will ultimately depend on whether the collaborators can convert complex multiomic observations into a validated rule that performs reliably beyond the original 80-patient study.
