Pinpointing the Right Patients for ADC Therapies

Imagene AI and Daiichi Sankyo team up to embed AI-driven biomarker discovery earlier in ADC oncology drug development

13 Apr 2026

The promise of precision oncology has always rested on a difficult problem: identifying which patients will benefit from a drug before the drug is given. Biomarker discovery, the science of finding biological signals that predict treatment response, has traditionally lagged behind drug development rather than guiding it. A new partnership between Imagene AI and Daiichi Sankyo, announced on April 9th, suggests that gap may be narrowing.

At the centre of the collaboration is Imagene's OI Suite, built on a foundation model called CanvOI. The platform combines histopathology imaging, molecular data, and longitudinal patient records into a single analytical system. Backing it is a data lake of more than 3.5 million tissue samples linked to clinical outcomes, a resource intended to give the model enough breadth to work across varied patient populations.

The focus is Daiichi Sankyo's antibody-drug conjugate pipeline. ADC therapies are among oncology's more precise instruments, attaching toxic payloads directly to cancer cells via targeted antibodies. That precision, however, cuts both ways: the drugs' effectiveness and their toxicity profile depend heavily on the biology of individual tumours, making patient selection critical. Imagene will apply response prediction models and its Composite Continuous Scoring methodology to that selection problem. The approach measures target protein expression on a continuous scale rather than sorting patients into blunt categories, resolving gradations that conventional immunohistochemistry scoring tends to flatten.

The commercial logic is clear enough. Clinical trial failure rates in oncology remain stubbornly high, and a significant share of that failure traces back to poor patient stratification. Embedding biomarker analysis earlier in programme design, rather than treating it as a validation step near the end, could reduce costly late-stage attrition.

Whether the platform delivers on that logic at scale remains an open question. AI tools that perform well on curated datasets do not always transfer cleanly to the heterogeneous reality of clinical practice. Regulatory acceptance of continuous scoring approaches, rather than established categorical thresholds, is also unproven territory.

For now, the partnership reflects a broader industry shift: biomarker strategy is moving from the margins of drug development toward its centre. Whether that repositioning produces better drugs, or merely more confident predictions, will only become clear when the data from Daiichi Sankyo's trials come in.