The rapid ascent of antibody-drug conjugates (ADCs) has transformed the oncology landscape, offering a high-precision alternative to traditional, blunt-force chemotherapy. With roughly 15 FDA-approved products now on the market and the sector projected to reach a staggering $57 billion valuation by 2032, ADCs represent the frontier of modern cancer treatment. Yet, beneath the industry’s optimism lies a persistent, confounding clinical reality: many tumors that appear "target-positive" in standard laboratory assays fail to respond to these sophisticated drugs.
To solve this, Gilead Sciences and AI-spatial biology specialist Nucleai have formed a strategic alliance to decode the biological discrepancies that cause ADC resistance. By merging Gilead’s expansive clinical trial datasets with Nucleai’s advanced computational pathology, the partnership seeks to move beyond the binary "positive or negative" staining of traditional pathology, looking instead at the spatial architecture of the tumor microenvironment to determine why some patients thrive while others derive no benefit.
The Promise and Peril of "Targeted Chemotherapy"
For years, the ADC value proposition was sold to clinicians as the "GPS-guided" missile of oncology. The concept is elegant in its simplicity: attach a potent cytotoxic payload to an antibody that recognizes a specific protein on the surface of a cancer cell. The antibody navigates through the bloodstream, binds to the tumor, and releases its toxic cargo only where it is needed, theoretically sparing healthy tissue and eliminating the systemic toxicity associated with conventional chemotherapy.
However, as Dr. Ken Bloom, head of pathology at Nucleai, observes, clinical reality is far more complex. "Initially, what we were told was that ADCs are basically targeted chemotherapy," Bloom notes. "I’m going to put a GPS signal on my beachhead, and it’s going to zone in just to the tumor cells and release its payload."
In practice, this "beachhead" is often shifting. A tumor might test positive for a target protein on a biopsy slide, but the drug fails because the target is either altered, structurally inaccessible, or masked by the tumor’s own defenses. Cancer cells are notoriously adaptive; they employ drug-efflux pumps to expel payloads, or utilize extracellular proteases within the microenvironment to cleave the ADC’s linker prematurely, releasing the toxic payload in the wrong location.
A Chronology of the Gilead-Nucleai Collaboration
The partnership between Gilead and Nucleai did not emerge overnight; it is the product of a multi-year evolution in how biopharma companies handle translational data.
- Early Phase (Pre-2024): The relationship began with focused, preclinical research. Nucleai provided specialized multiplex immunofluorescence (mIF) analysis, allowing Gilead to observe how ADCs interacted with complex, non-homogeneous tumor tissue in controlled settings.
- Expansion (2024-2025): Recognizing the limitations of preclinical data, the companies scaled their collaboration. They moved toward the retrospective analysis of whole-slide images (WSI) taken from Gilead’s diverse portfolio of clinical trials across multiple oncology indications.
- The August 2026 Disclosure: The formal announcement on August 11, 2026, codified this work, revealing that Nucleai has been training AI models on H&E (hematoxylin and eosin) and IHC (immunohistochemistry) slides to correlate spatial tissue features with actual patient outcomes.
- The Horizon: Both organizations are currently preparing to share their findings. Dr. Bloom indicates that peer-reviewed publications and clinical presentations are slated for late 2026 or early 2027, as the team works to validate several "candidate biomarkers" that appear more predictive than current standard-of-care assays.
Unmasking the "Epitope Problem"
One of the most significant barriers to ADC efficacy is a long-standing "hidden secret" in pathology: the difference between where a diagnostic test looks and where the drug actually binds.
Most conventional IHC assays used to determine patient eligibility are designed to detect the intracellular domain of a protein—the portion inside the cell membrane. This is done because tissue processing is more stable for internal proteins, ensuring consistent results across labs. However, ADCs are designed to bind to the extracellular domain—the portion exposed to the bloodstream.
A prime example is the HER2 receptor. The standard 4B5 diagnostic antibody detects the intracellular domain, meaning it can flag a tumor as "HER2-positive" even if the cell is shedding the extracellular portion of the receptor (forming p95HER2 fragments). Because the drug (such as trastuzumab) requires the extracellular domain to dock, a patient might be classified as a candidate for the therapy despite the tumor lacking the very "hook" the drug needs to attach itself.
This "inside-outside" mismatch is a primary focus of the Gilead-Nucleai partnership. By using AI to analyze spatial relationships, the team is attempting to identify markers that specifically confirm the presence of the accessible extracellular epitope, rather than just the total protein volume.
Computational Pathology: Beyond the Human Eye
The transition from "digital" to "computational" pathology represents a paradigm shift. For decades, digital pathology meant simply scanning a glass slide into a computer for easier storage or remote viewing. Computational pathology, however, uses machine learning to extract data that the human eye cannot quantify.
The Power of the Neighborhood
"Now we care about the neighborhood that a cell is in," says Dr. Bloom. By analyzing the density of tumor-infiltrating lymphocytes, the spatial distribution of extracellular matrix components, and the proximity of cancer cells to the vasculature, Nucleai’s AI can provide a "spatial signature" of a tumor. This allows researchers to understand not just if a target is present, but how the tumor environment facilitates or hinders the ADC’s access to that target.
Standardizing Accuracy
Beyond drug development, computational pathology acts as a "floor-raiser" for clinical diagnostics. Human pathologists are subject to fatigue, cognitive bias, and variations in training. By integrating AI "overreaders," laboratories can ensure that assessments remain consistent. According to internal benchmarks cited by Nucleai, AI-assisted workflows catch human errors and provide a degree of uniformity that was previously unattainable in decentralized pathology labs.
Official Perspectives: The Path Forward
Gilead Sciences views this collaboration as a critical component of its broader translational medicine strategy. Meghna Das Thakur, senior director of oncology biomarkers at Gilead, emphasized that the partnership is not merely about using AI to look at pictures, but about integrating deep scientific and clinical expertise. "Combining [Nucleai’s] specialized expertise in AI-driven spatial biology and tissue analytics with Gilead’s clinical expertise allows us to generate insights more efficiently," she noted.
However, industry leaders remain cautious about the path to clinical implementation. Dr. Bloom acknowledges that the pharmaceutical industry has yet to establish a standardized, regulatory-compliant "highway" for AI-based biomarkers. "There isn’t a pathway that’s been forged clearly yet," he admits. "But the good news is that there are several leaders out there attempting to forge that first path. What you’re going to see, as soon as the first one hits, is a wave that follows."
Implications for Future Cancer Care
The implications of the Gilead-Nucleai work are profound. If researchers can successfully move beyond simple IHC scoring to more complex, spatial-based biomarker panels, the clinical impact would be two-fold:
- Reduced Treatment Failure: By filtering out patients who are unlikely to respond due to epitope inaccessibility, clinicians can avoid exposing them to the unnecessary toxicities of ADCs and move them toward more effective therapies faster.
- Expanded Patient Access: Conversely, a more nuanced understanding of the tumor microenvironment might reveal that some patients—currently classified as "low-expressors" or "negative"—are actually ideal candidates for ADCs based on the specific spatial architecture of their tumors.
Ultimately, the goal is to bridge the gap between the slide and the bedside. While AI provides the analytical engine, it does not replace the physician. As Dr. Bloom frequently tells his students, "A fool with a tool is still a fool." The future of oncology lies in the hybrid model: the clinical intuition of the trained pathologist augmented by the high-resolution, spatial-analytical power of AI. As this collaboration continues to mature, the industry waits to see if these "candidate biomarkers" will be the key to unlocking the full potential of the $57 billion ADC market.
About the Author
