AI Biomarker Discovery Is Rewriting Cancer Treatment

Finding the right cancer treatment for the right patient has always been medicine’s hardest matching problem. Oncologists order biopsies, run molecular panels, wait weeks for results, and still face uncertainty about whether a drug will work. In 2026, artificial intelligence is changing that equation by discovering biomarkers that human researchers would never find on their own.

The shift is not theoretical. Major pharmaceutical companies are signing deals, regulatory agencies are granting approvals, and AI-identified biomarkers are entering clinical workflows right now.

What Are Biomarkers and Why Do They Matter

A biomarker is any measurable indicator that tells doctors something useful about a patient’s disease. In oncology, biomarkers help answer critical questions: Will this tumor respond to immunotherapy? Is the cancer likely to recur after surgery? Should we try a targeted therapy or stick with chemotherapy?

Traditional biomarker discovery relies on hypothesis-driven research. Scientists suspect a protein or genetic mutation plays a role, then spend years testing that theory. The process is slow, expensive, and limited by human pattern recognition. Most cancer patients still receive treatments selected through trial and error rather than precise molecular matching.

AI flips this approach. Instead of starting with a hypothesis, machine learning models analyze thousands of tissue samples, genomic profiles, and clinical outcomes simultaneously. They find correlations that no human team could spot across datasets of that scale.

Imagene AI and Daiichi Sankyo Push Multimodal Analysis Forward

In April 2026, Imagene AI announced a collaboration with Daiichi Sankyo that illustrates where the field is heading. The partnership focuses on antibody drug conjugate (ADC) development programs, one of the hottest areas in oncology.

Daiichi Sankyo will use Imagene’s OI Suite, a multimodal platform built on their CanvOI foundation model. The system analyzes hematoxylin and eosin (H&E) stained slides and immunohistochemistry (IHC) images alongside molecular profiles and longitudinal clinical outcomes. By combining these data streams, the AI generates predictions about which patients are most likely to respond to specific ADC treatments.

The word “multimodal” is key here. Previous AI pathology tools typically analyzed one data type at a time. A model might examine a tissue slide or parse genomic data, but rarely both together. Imagene’s approach mirrors how experienced pathologists actually think, integrating visual tissue patterns with molecular context to form a complete picture.

The practical goal is straightforward: identify biomarker candidates earlier in drug development, strengthen patient stratification in clinical trials, and reduce the number of patients who receive ineffective treatments.

Lunit and CellCarta Build the Infrastructure Layer

While Imagene focuses on biomarker discovery, other companies are building the infrastructure that makes AI-powered diagnostics scalable. In March 2026, Lunit and CellCarta announced a strategic collaboration to accelerate AI-enabled digital pathology across translational research, clinical trials, and companion diagnostic programs.

Lunit brings computer vision models trained on millions of pathology slides. CellCarta contributes its global network of laboratories and regulatory expertise. Together, they aim to create standardized workflows that pharmaceutical companies can plug into their drug development pipelines.

This infrastructure work matters because the best AI model in the world is useless without reliable data pipelines, validated staining protocols, and regulatory-grade quality controls. The gap between a research prototype and a clinically deployable diagnostic tool is enormous, and partnerships like this one exist to close it.

Roche and AstraZeneca Score a Regulatory First

Perhaps the most significant milestone for AI companion diagnostics came when Roche received FDA Breakthrough Device Designation for the VENTANA TROP2 RxDx Device. This computational pathology system earned the first-ever Breakthrough Device Designation granted for an AI-driven companion diagnostic.

The device targets non-small cell lung cancer (NSCLC). It uses immunohistochemistry staining combined with digital pathology scanners and an AI algorithm to process whole-slide images and generate a quantitative TROP2 score. That score helps identify patients most likely to benefit from datopotamab deruxtecan, an ADC co-developed by Daiichi Sankyo and AstraZeneca.

The AI algorithm incorporates AstraZeneca’s proprietary quantitative continuous scoring (QCS) platform. Unlike traditional binary positive/negative tests, QCS produces a continuous score that captures the full spectrum of TROP2 expression. This granularity allows oncologists to make more nuanced treatment decisions rather than relying on crude cutoffs.

The device is expected to launch commercially in 2026, making it one of the first AI-powered companion diagnostics to reach routine clinical use.

The Record-Breaking CDx Approval Count

The Roche device is not an isolated event. The total number of FDA-approved companion diagnostic tests hit a record 188 in early 2026, reflecting sustained regulatory momentum. Each approved CDx test represents a treatment-biomarker pair that has been validated through rigorous clinical evidence.

AI is accelerating this pipeline in two ways. First, it speeds up biomarker identification during early drug development, giving companies CDx candidates to validate sooner. Second, it enables companion diagnostics that would be impossible without computational analysis, such as tests that integrate spatial tissue architecture with protein expression patterns across an entire tumor section.

The companion diagnostics market is projected to grow substantially through the decade, driven by the increasing number of targeted therapies entering clinical development and the parallel maturation of AI pathology platforms.

How AI Biomarker Discovery Actually Works

Understanding the technical process helps explain why AI biomarkers differ from traditional ones.

A typical AI biomarker discovery pipeline starts with a large dataset of digitized pathology slides paired with clinical outcomes. The AI model, usually a vision transformer or convolutional neural network trained on millions of image patches, learns to associate visual tissue features with patient responses to treatment.

The model might discover that a specific spatial arrangement of immune cells around tumor nests predicts immunotherapy response better than any single molecular marker. Or it might find that the texture of the tumor stroma, invisible to human assessment at scale, correlates with drug resistance.

These discoveries are then validated through retrospective analysis of independent patient cohorts and prospective clinical studies. The validation step is critical because AI models can find spurious correlations in training data that do not hold up in new patient populations.

Foundation models like Imagene’s CanvOI represent the latest architectural evolution. These models are pre-trained on massive pathology datasets and then fine-tuned for specific biomarker tasks. This transfer learning approach means new biomarker models can be built faster with smaller task-specific datasets, reducing the data bottleneck that previously limited AI pathology research.

Challenges That Remain

Despite the momentum, several obstacles stand between current AI biomarker tools and widespread clinical adoption.

Data standardization remains the biggest technical challenge. Different hospitals use different scanners, staining protocols, and tissue preparation methods. An AI model trained on slides from one institution may perform poorly on slides from another. Cross-site validation and domain adaptation techniques are improving but not yet solved.

Regulatory pathways are still evolving. The FDA has approved individual AI-driven diagnostics, but there is no standardized framework for validating AI biomarker discovery platforms as a category. Each new biomarker-drug pair requires its own regulatory submission, which limits the speed at which AI discoveries reach patients.

Clinical integration presents practical hurdles. Most pathology labs are not yet equipped for routine digital pathology workflows. Scanning all slides, storing terabytes of image data, and running AI inference requires infrastructure investments that many hospitals have not made.

Interpretability concerns persist. When an AI model identifies a novel biomarker based on tissue morphology patterns, clinicians want to understand why that pattern predicts treatment response. Black-box predictions face resistance from oncologists who need to explain treatment rationale to patients and tumor boards.

What This Means for Patients

The immediate impact of AI biomarker discovery falls on cancer patients who currently receive treatments that will not work for them. Across oncology, response rates for many therapies hover between 20% and 40%. That means the majority of patients endure side effects from drugs that do not help them.

Better biomarkers mean better patient selection. If an AI-powered companion diagnostic can identify the 35% of patients who will respond to a specific ADC, the remaining 65% can be directed toward alternative treatments sooner. This is not just an efficiency improvement. For patients with advanced cancer, every week spent on an ineffective therapy is a week lost.

The convergence of multimodal AI platforms, regulatory acceptance, and pharmaceutical investment suggests that 2026 may be remembered as the year AI biomarker discovery moved from research curiosity to clinical reality. The partnerships announced this spring are not pilot programs. They are production deployments aimed at influencing real treatment decisions for real patients.

For anyone following precision medicine and AI healthcare, the companion diagnostics space deserves close attention. The technology is mature enough to deliver results, and the business case is strong enough to attract sustained investment from the largest pharmaceutical companies in the world.