For decades, a stark reality has defined oncology drug development: nine out of every ten therapies that enter clinical trials will fail.1 This staggering attrition rate represents an immense loss of time, capital, and, most importantly, potential hope for patients. A primary driver of this failure is the persistent gap between the laboratory bench and the patient’s bedside. Traditional preclinical models, from 2D cell lines to animal models, have long struggled to capture the profound heterogeneity and biological complexity of human cancer.

At Tempus, we believe the solution is not simply a better model, but a fundamentally better system. We are moving beyond linear, siloed research by building an integrated engine for discovery and validation. This engine combines three pillars: a large library of real-world clinical and molecular data, advanced AI to navigate its complexity, and a suite of patient-derived biological models to test hypotheses in a physiologically relevant context. By creating an iterative loop between these components, we can generate and validate insights grounded in real-world patient biology, supporting a more data-driven approach to discovery.

The persistent challenge of preclinical modeling

The central challenge in preclinical research is creating a model that accurately predicts how a drug will behave in a patient. Traditional models fall short for several key reasons:

• Lack of heterogeneity: 2D cell lines grown on plastic are homogeneous and lose the complex, hierarchical structure of a real tumor.
• Absence of the tumor microenvironment (TME): These models lack the intricate network of immune cells, stromal cells, and vasculature that profoundly influences tumor behavior and drug response.
• Confounding biology: Animal models, while more complex, introduce mouse-specific biology that can lead to misleading results and poor clinical translation.

These limitations mean that discoveries made in the lab often do not hold up in the clinic, contributing directly to the high failure rate of promising new therapies.

The persistent challenge of preclinical modeling

To overcome these challenges, Tempus has developed a proprietary “lab in the loop” platform that integrates our core strengths in data, AI, and biological modeling.

1. Grounding discovery in real-world patient biology: Our process begins with our real-world data (RWD) library, which includes more than 8 million de-identified research records. These records pair clinical data with molecular data, including DNA and whole-transcriptome RNA sequencing, generated with our Tempus xT and xR assays. This multimodal dataset provides a comprehensive view into the complexity of cancer as it exists in patients, serving as the foundation for subsequent discovery work.
2. Navigating complexity with advanced AI: We apply advanced AI and machine learning algorithms across this dataset to help identify patterns, stratify patients, and generate hypotheses. For example, AI can help us identify homogeneous subcohorts within a historically heterogeneous disease, revealing patient groups that may be more responsive to a specific therapeutic strategy.
   
3. Testing hypotheses with physiologically relevant models at scale: The insights generated from our data are then tested in our advanced biological modeling platforms. We have built two complementary systems to address different biological questions:
   • Confounding biology: Animal models, while more complex, introduce mouse-specific biology that can lead to misleading results and poor clinical translation.
   • Patient-Derived Tumor Fragments (PDTFs): For therapies targeting the TME, we use our PDTF assay. In this platform, fresh tumor tissue is dissected into small fragments that preserve the native TME. These fragments are exposed to a drug, and the response across every cell type—from tumor cells to immune and endothelial cells—is analyzed in high detail using single-cell RNA sequencing.

A key feature of our platform is that every biological model is molecularly characterized using the same Tempus assays as our real-world patient data. This shared molecular language helps enable the translation of insights between the wet lab and the dry lab.

The lab in the loop: Translating insights from models to patients

The power of this integrated engine lies in its ability to create a virtuous cycle of discovery and validation. We can derive a hypothesis from RWD, test it at scale in our PDOs, and then use the results to make predictions that are vetted against clinical outcomes in our database.

A clear example of this is our work with the nectin-4-targeting antibody-drug conjugate (ADC), enfortumab vedotin. A pan-cancer screen of our PDOs revealed that while nectin-4 expression correlated with response, it was not a perfect predictor. Many high-expressing models showed a wide range of sensitivities, suggesting other molecular drivers were at play.

Using the Multi-omic profiles of the responding and non-responding organoids, we derived a 30-gene signature that classified models into “high cell killing” and “low cell killing” groups. We then projected this PDO-derived classifier onto our RWD of patients treated with enfortumab vedotin. The results were striking:

• Patients whose tumors were classified as “high cell killing” by our signature showed significantly greater progression-free survival.
• To ensure the signature was predictive and not just prognostic, we applied it to a similar cohort of patients who had never received the drug. In this group, the survival difference disappeared completely.

This demonstrates the potential of our approach: from a screen in patient-derived models, we identified a novel biomarker signature that was associated with clinical outcomes in real-world patients.

Modeling the TME: A frontier for complex therapies

For therapies targeting the complex TME, such as immuno-oncology (IO) agents and angiogenesis inhibitors, our PDTF assay provides critical insights. By analyzing drug response at the single-cell level, we can understand how a therapy impacts not just the tumor, but the entire ecosystem around it.

In one study, we used the PDTF assay to evaluate an inhibitor of the VEGF pathway in endometrial tumors. Single-cell analysis confirmed that the drug had its most robust effect in endothelial cells, downregulating the VEGF pathway and its downstream effectors. We also observed a key biological consequence of angiogenesis inhibition—an increase in hypoxia—providing a functional biomarker of drug activity.

This platform is also uniquely suited for IO research. We employ an efficient pre-screening step to identify immunologically “hot” tumors that are most likely to respond to checkpoint inhibitors before committing to resource-intensive single-cell sequencing. This allows us to focus our analysis on the most informative samples to understand the mechanisms of response and resistance to IO therapies.

A new paradigm for drug development

The “lab in the loop” represents a new paradigm for oncology research. By tightly integrating RWD, AI, and advanced biological models, we are creating a system designed to help accelerate the development of next-generation cancer therapies. This approach allows our partners to:

• Validate targets with confidence using models that reflect human biology.
• Identify predictive biomarkers that can stratify patients and help improve clinical trial success.
• Understand complex mechanisms of action, particularly for therapies targeting the TME.
• Accelerate timelines, moving from molecule to actionable data efficiently.

Note: Content edited for clarity. Please note that the content in this document has been revised for clarity and conciseness. Some language and formatting may have been adjusted to enhance readability while preserving the original meaning and intent of the discussion.

References

1. R&D costs. Knowledge Portal. Accessed August 26, 2025. https://www.knowledgeportalia.org/costs-r-d#:~:text=The%20majority%20of%20the%20studies,a%20summary%20of%20the%20studies

*Internal Tempus database analysis