Q&A: Navigating tumor complexity to build confidence in new targets

11/18/2025

Q&A: Navigating tumor complexity to build confidence in new targets

Justin Guinney, PhD, SVP of Cancer Genomics at Tempus, discusses the Tempus Loop platform and explains how it uses multimodal data and patient-derived organoids to accelerate early-stage discovery and validate novel targets with greater confidence.

Authors Justin Guinney, PhD
Senior Vice President,
Cancer Genomics

What is the core challenge in oncology drug development that Tempus Loop was designed to address?

Justin Guinney, PhD: The principal challenge we’re up against is the stubborn fact that human disease and biology are mind-numbingly complex. That complexity is really what leads to the high cost and, frankly, the high attrition rates we see in drug development. But that doesn’t mean there aren’t opportunities for optimization. We believe a key way to tackle this is to get a deeper understanding of cancer’s heterogeneity. Our approach, however, is to find the homogeneity within the cancer. We posit that if you can focus on those more homogeneous cell populations, you’re just more likely to find the true cancer drivers—the true dependencies—and those are what are more likely to lead to truly effective drugs.

To do this effectively, we really needed three foundational pillars: first, a deep understanding of patient populations from rich data; second, better biological models that actually recapitulate the disease biology we see in the data; and third, of course, an AI engine—that’s the substrate that allows us to make sense of all this complex, multimodal data. Tempus Loop was built on these pillars to improve the likelihood of successfully advancing an asset from early to late-stage development.

What is Tempus Loop, and how does it function?

Justin Guinney, PhD: Tempus Loop is our AI-enabled, lab-in-the-loop platform for target discovery and validation. The platform embraces the widely recognized “lab-in-the-loop” concept, which focuses on rapidly developing hypotheses and then testing them in relevant model systems.

It integrates three foundational building blocks that Tempus has been developing for over a decade:

A large data repository: This has over 8 million de-identified patient records, and for many of those, we have rich, multimodal data—genomics, transcriptomics, and essential clinical information—which is key for dissecting disease complexity.
A large organoid library: Over 1,000 patient-derived tumor organoids serve as the basis for testing our hypotheses.
A sophisticated AI engine: We leverage Tempus’ AI backbone to analyze all that complex data and extract targets we believe are robust.

The platform propagates this iterative cycle. We start by identifying a de-identified patient cohort with high unmet need using our real-world data. Then, we apply AI to characterize those patients, which may lead to the discovery of novel disease subtypes. Next, we bring in systems biology and in silico tools to identify and prioritize potential targets within those new subtypes. But the critical step is what comes next: we test our target hypotheses in relevant tumor models. We select PDO models that are matched to the patient subtype of interest, and then we run rapid assays like CRISPR knockout or drug screens to see how the viability or other phenotypes are affected. Importantly, we learn from those experiments, and new hypotheses naturally emerge. That then brings us right back to our real-world data where we further refine our hypotheses, effectively closing the loop. Combining wet- and dry-lab workflows in a single platform optimizes handoffs and surfaces insights faster, so the end-to-end process accelerates preclinical discovery.

“The ultimate goal is to build a robust, evidence-based case that a specific gene is a bona fide target worthy of advancing into a drug development program. By integrating the deep data analysis with the functional validation, we give our customers the confidence they need to advance an asset into the next stage.”

– Justin Guinney, PhD, SVP, Cancer Genomics, Tempus

How does Tempus Loop use AI to identify new disease subtypes from real-world data?

Justin Guinney, PhD: At a high level, Tempus Loop leverages AI to identify patterns within the data cohort. We recently completed a project with a large pharmaceutical company focused on non-small cell lung cancer (NSCLC). We started with a very large funnel—over 120,000 lung cancer patients—and applied our AI algorithms to a cohort of 8,000 lung adenocarcinoma samples. The algorithms identified patterns as new subtypes of disease. More specifically, Tempus Loop was able to uncover six distinct molecular subtypes, C1 through C6.

Once we identified these subtypes, we drilled in to understand them biologically. We analyzed transcriptional signatures, genomic variants, and clinical associations. Through this systems biology approach, we found that the C5 subtype was associated with a much worse overall survival and was depleted of EGFR mutations. That was a huge flag to our teams because those patients don’t have access to certain targeted therapies, so it’s a subtype with a clear, significant unmet need. That ability to deconstruct a highly heterogeneous disease into more homogeneous, clinically relevant subgroups is a foundational, crucial step in our process.

Could you walk us through how a potential target is validated using the platform?

Justin Guinney, PhD:Sticking with the NSCLC example, after we identified the C5 subtype as an area of high unmet need, we used a range of computational and in silico approaches—protein-protein networks, gene regulatory networks, and various statistical models—to nominate a few potential therapeutic targets specific to that C5 biology. The next crucial step was validation; we needed to test the targets in a relevant model system. We built a classifier to project the C5 molecular signature onto our library of over 1,000 PDOs. That let us make an intelligent selection, picking a few PDOs that closely mirrored the C5 patient phenotype to serve as our test models, while also selecting other subtypes, like C1, to act as our controls. Our hypothesis was simple: these nominated targets are more likely to be effective for the C5 subtype, but not for the others.

We then ran a CRISPR knockout experiment on those targets in the selected PDOs. The results were exactly what we hoped for: knocking out the targets significantly reduced cell viability in the C5-matched PDOs, but it had little to no impact on the C1 control PDOs. That experiment helped to validate our initial hypothesis. It also gave us high confidence that the Tempus Loop platform can not only identify those promising targets, but also validate them in a biologically relevant context that holds up.

What is the final output of a Tempus Loop project for a life sciences partner?

Justin Guinney, PhD: The entire process culminates in what we call a target profile report. It’s not a single data point; it’s a comprehensive report that aggregates many different forms of evidence. As I said, target discovery is fraught with many perils. It’s challenging, and there’s no silver bullet. It requires building a lot of partial forms of evidence that, when aggregated together, point you in the right direction.

The report includes the functional validation data from the PDOs, but also spatial profiling data. That’s really valuable because it tells you if a target is expressed right within the tumor cells, or if it might be found in the stroma or the microenvironment. It’s this combination of data points that enables pharmaceutical companies to think much more intelligently about the right mechanism of action and the appropriate way to build a drug program around it.

The ultimate goal is to build a robust, evidence-based case that a specific gene is a bona fide target worthy of advancing into a drug development program. By integrating the deep data analysis with the functional validation, we help give our customers the confidence they need to advance an asset into the next stage.

To learn more about how Tempus’ AI-enabled platform supports target discovery and validation, explore our solutions for life sciences. For in-depth demonstrations of our proprietary applications, contact us here.