Q&A: How Transcriptomics and AI Are Accelerating Oncology R&D

07/24/2025

Q&A: How Transcriptomics and AI Are Accelerating Oncology R&D

In a recent webinar hosted by Tempus, industry leaders from healthcare and life sciences gathered to explore current and emerging applications of transcriptomic data in oncology.

Authors Justin Guinney, PhD
SVP, Cancer Genomics, Tempus

Rafael Rosengarten, PhD
CEO & Co-Founder, Genialis

Sandip P. Patel, MD
Professor of Medical Oncology, UC San Diego

Justin Guinney, PhD, SVP of Cancer Genomics at Tempus, moderated the session and was joined by Rafael Rosengarten, PhD, CEO & Co-founder of Genialis, and Sandip P. Patel, MD, Professor of Medical Oncology, UC San Diego. The conversation centered on the exploration of transcriptomic data applications that are helping to change both oncology drug development and patient care.

RNA-seq’s utility has increased in oncology, offering a dynamic and functional understanding of cancer that extends beyond traditional genomic analyses. By measuring gene expression and identifying critical gene fusions, RNA-seq provides a more comprehensive view of tumor biology, classification, and progression. This capacity to reveal active biological processes and potential therapeutic vulnerabilities is contributing to advancements in precision cancer diagnostics and the acceleration of drug development.

Justin Guinney, PhD: Rafael, given Genialis’ focus on using RNA to predict patient response to therapy, what do you see as the most significant untapped potential going forward for large-scale RNA sequencing?

Rafael Rosengarten, PhD: The biggest untapped potential, which we’re seeing take off, is clinically relevant RNA profiling tests, likely to be interpreted by AI-driven algorithms. While RNA’s role in detecting structural variants can be therapeutically meaningful — and more information is generally good for understanding disease — in a regulated environment, interpretability is paramount. The FDA wants interpretable answers, not just raw data. This is where AI’s advancements in algorithmic science become valuable. I’m especially excited about robust, predictive ensemble models of complex signatures that profile therapeutically vulnerable tumor biology. Just like with imaging-based AI devices, I anticipate a dramatic increase in FDA-cleared RNA-seq-based tests in the coming decade.

Justin Guinney, PhD: Sandip, from your perspective, when you think about the potential for RNA in driving research, what do you see as the biggest opportunity to impact clinical needs in oncology?

Sandip P. Patel, MD: RNA-seq is used for diagnosing certain actionable fusions, which may be exquisitely sensitive to targeted therapy; some targets, like NRG1, are predominantly diagnosed this way. For research, there are many potential applications. RNA-seq can inform antibody-drug conjugate efficacy, and it’s a very interesting area for immunotherapy, given the lack of good DNA or protein biomarkers; it helps understand immune composition in both tumor and stroma. As Rafael mentioned, we’re focused on the central dogma, and with cancer, you can’t leave any stone unturned. We need to consider DNA, RNA, and protein; the best approach depends on the use case. We’d be remiss not to explore all three aspects of tackling cancer.

“What I love about [Tempus Loop] is that it flips the traditional, linear drug discovery process on its head. Instead of starting with target ID and moving towards patients 20 years later, this begins with large, aggregated real-world patient datasets derived from how they behaved in the real world when exposed to therapeutic intervention.”

– Rafael Rosengarten, PhD, CEO & Co-founder, Genialis

Advancing Research with Tempus Loop

Justin Guinney: One of the latest strategies helping transform oncology research is Tempus Loop, a platform that embodies the concept of a lab in the loop. This proprietary platform integrates real-world data (RWD), patient-derived organoids (PDOs), and AI to support the identification and validation of actionable targets. By leveraging Tempus’ RWD, Tempus Loop aims to address the challenges of traditional drug discovery, such as identifying patient subpopulations with shared clinical and molecular characteristics. Through systems biology and CRISPR screens, the platform provides continuity between RWD and PDOs, allowing for rapid hypothesis testing in models that closely mirror patient biology. This iterative process is intended to support target discovery and validation and may offer a strategy for advancing personalized cancer therapeutics.

Justin Guinney, PhD: Considering Tempus Loop’s approach to developing hypotheses and testing targets in patient-derived organoids, do you see this as a promising strategy? Are there any potential blind spots or limitations for a platform like this?

Sandip P. Patel, MD: I think these multimodal strategies are going to be very important. Often, models are trained on highly selected data that doesn’t reflect real-world use, which we sometimes see in electronic medical records (EMR). This approach, leveraging data while also investigating the biology through organoids or broader datasets, is really smart. We definitely need biological validation for our computational outputs, and to aggregate noisy biological datasets and understand their context with other clinical datasets. This integration can create a virtuous loop, a flywheel effect — combining these datasets helps us better understand human biology to treat disease.

Rafael Rosengarten, PhD: What I love about this concept is that it flips the traditional, linear drug discovery process on its head. Instead of starting with target ID and moving toward patients 20 years later, [Tempus Loop] begins with large, aggregated real-world patient datasets derived from how they behaved in the real world when exposed to therapeutic intervention. We can learn so much by mining this, especially from patients not responding to standard care, to understand their shared therapeutic vulnerabilities. The exciting part about patient-derived organoids is that they let us test hypotheses directly in models much closer to actual patient biology, giving strong signals on a hypothesis’s validity. I think it’s really promising, and we’re excited to collaborate to further validate the insights generated from this approach.

Jason Guinney, PhD: Sandip, given your work in immuno-oncology, how do you approach capturing the right biology in these model systems, like Tempus Loop, and projecting it back and forth with patient information?

Sandip P Patel, MD: One of the issues we have with immunotherapy biomarkers is the subtle, critical differences in immunobiology between mouse models and human models, even humanized mice. That’s why real-world data sets are so meaningful. The flywheel concept can help in two ways: it supports validation of computational observations biologically, and it allows cross-system validation – assessing how well mouse versus human data map to mouse versus human integrations biologically. This type of 2×2 analysis is key because a lot can be lost in translation. The depth of multimodal data within systems like Tempus Loop can be significant, encompassing transcriptomics, radiomics, digital pathology, genomics, and clinical phenotyping. This bench-to-bedside-to-cloud analysis truly helps translate discoveries more efficiently, especially since we’re often developing drugs before the biomarker is fully elucidated, so these approaches may help us better aim for those targets.

“It’s hard to overstate the potential benefits [of foundation models]… An important aspect is training a foundation model on massive, multimodal datasets, like Tempus’, to understand real-world cancer biology as it occurs in patients. Instead of training individual models for specific responses, we’re training a model to understand all molecular and clinical biology interactions. This allows us to probe the foundation model using simpler analytes and learn the best way to treat patients in ways previously impossible.”

– Rafael Rosengarten, PhD, CEO & Co-founder, Genialis

The Role of AI-Driven Algorithms in Oncology R&D

Justin Guinney: It’s clear we’re entering a new era of medicine, which we like to call ‘precision medicine 2.0.’ This is largely driven by the integration of multimodal datasets, moving beyond the traditional reliance on DNA alone. Among the developments in this area are algorithms like Tempus’ Immune Profile Score (IPS) and Genialis’ krasID. IPS is designed to provide a more nuanced prediction of patient responses to immune checkpoint inhibitors, going beyond conventional markers like PD-L1 and TMB. Meanwhile, krasID offers an approach to stratifying KRAS patients by clinical response, using RNA data to predict and monitor drug efficacy. These advancements highlight the potential of AI to refine treatment strategies and help improve patient outcomes, and allow us to explore how these tools, and how AI in general, is being leveraged in clinical practice and drug discovery today.

Justin Guinney, PhD: Sandip, as a practicing lung cancer oncologist, how do you value the deployment and utility of algorithms, especially considering that physicians may not want to withhold therapy if no other good options exist?

Sandip P. Patel, MD: Until we have validated interventions, these datasets are really helpful for breaking ties. For example, in non-small cell lung cancer with PD-L1 greater than 1%, both PD-1 alone and chemo plus PD-1 are reasonable options, and we’re generally working with limited information when we make those decisions. An algorithm could potentially help; if a PD-L1 high patient has low IPS, I might lean more towards a chemotherapy combination, as I’d have less confidence in their response to PD-1 alone. For KRAS inhibitors, we’re seeing combination strategies emerge, and even compound biomarkers like krasID plus IPS may be important, especially for STK11/KEAP1 co-mutated patients where immunotherapy is less effective. There are many possible use cases. Even without prospective validation, these algorithms may be useful for guiding decisions among approved options, as some data is always better than none.

Justin Guinney, PhD: Rafael, with Tempus recently announcing its collaboration with AstraZeneca and Pathos for the development of a multimodal foundation model, and Genialis having its foundation model in RNA and other modalities, how would you characterize the benefits these kinds of foundational models may bring to discovery?

Rafael Rosengarten, PhD: It’s hard to overstate the potential benefits. Despite my caution against overhyping AI, I truly believe in the science. Genialis has been developing these foundational AI tools for 15 years, long before we even had a vocabulary for the technology. With that said, these multimodal models may have a substantive impact. An important aspect is training a foundation model on massive, multimodal datasets, like Tempus’, to understand real-world cancer biology as it occurs in patients. Instead of training individual models for specific responses, we’re training a model to understand all molecular and clinical biology interactions. This allows us to probe the foundation model using simpler analytes and potentially learn new ways to treat patients in ways previously impossible. It’s incredibly fulfilling to see these long-held dreams become a reality and race towards actual practice.

Justin Guinney, PhD: Sandip, what’s your reaction to AI moving to the clinic, particularly regarding the interpretability of these complex, “black-box” models? How do you see AI in the clinic today, and what are your worries or excitement about its future opportunities?

Sandip P. Patel, MD: AI is here and increasingly used clinically for things like radiomics, pathology, and augmenting notes. While use cases are growing, we need to be cautious. We sometimes see issues with model overfitting, lack of generalizability, and insufficient prospective biological validation. A model might perform well on similar data, but clinical settings may require performance across varied, non-linear contexts. I’m hopeful AI can aid in linear, discrete tasks.

Justin Guinney, PhD: There’s potential for RNA to capture more complex biological states like gene overexpression, molecular subtypes, and immune ecotypes. Yet, despite two decades of intense study, few RNA-based tools capturing these elements are in the clinic. What are the key challenges you see in advancing RNA from the lab into the clinic?

Sandip P. Patel, MD: There are a couple of challenges. First, logistical: ensuring sufficient, high-quality material for analysis, which varies by disease context. Second, biological: Does RNA-seq truly inform treatment decision-making or research plans? Its utility may depend on the use case; it may be more helpful for pathway inhibitors or immunotherapy than for proteomic targets.

Finally, and often underappreciated, RNA-seq data is complex. How do you make it interpretable for a busy clinician? Clinicians need clear “yes” or “no” answers, not continuous variables. DNA sequencing is straightforward in this way – you either have a variant or you don’t. So, how do you make these complex RNA-seq datasets interpretable and actionable in a molecular report? That “last mile” for RNA-seq, ensuring a clean decision for clinicians, is a significant, underappreciated aspect, despite promising algorithms like IPS and krasID.

Rafael Rosengarten, PhD: Adding to Sandeep’s points, bulk RNA-seq is now commoditized and deployable globally, and it’s possible to obtain good-quality RNA from archival samples, which has helped clear many logistical hurdles, even though RNA is still a trickier molecule than DNA. Years ago, if you searched for gene expression biomarkers in PubMed, you’d find countless publications, but on the FDA’s website for approved CDx’s, there were hardly any. This gap is closing thanks to improved data harmonization, including efforts to address heterogeneity, bias, and batch effect correction, which are crucial for gene signatures to work clinically.

Another notable change is in the use of machine learning and AI tools. Typical simplistic gene signature scores may not be predictive in real- world settings. However, by using numerous gene signature scores as input to a machine learning algorithm, it is possible to build more sophisticated models that learn pathway interactions and that may prove predictive across independent clinical datasets. The challenge then becomes interpretability, as clinicians often need clear, binary answers. We may be able to provide that green light/red light output, but also offer the ability to peel back layers to show the underlying biological signatures, supporting explainable AI. This level of interpretability is where algorithmic science has advanced, potentially solving problems or doing a much better job than before.

The integration of transcriptomic data and AI-driven algorithms is helping pave the way for advancements in oncology research and development. To learn more about how Tempus is driving innovation and improving patient outcomes through precision medicine, click here or contact us.

*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.

Forward-Looking Statement

This press release contains forward-looking statements within the meaning of Section 27A of the Securities Act of 1933, as amended (the “Securities Act”), and Section 21E of the Securities Exchange Act of 1934, as amended, about Tempus and Tempus’ industry that involve substantial risks and uncertainties. All statements other than statements of historical facts contained in this press release are forward-looking statements, including, but not limited to, statements regarding the future applications of RNA sequencing; the potential for AI, including algorithms and foundation models, to refine treatment strategies and accelerate drug discovery; and the promise of new research methodologies that integrate real-world data with patient-derived models. In some cases, you can identify forward-looking statements because they contain words such as “anticipate,” “believe,” “contemplate,” “continue,” “could,” “estimate,” “expect,” “going to,” “intend,” “may,” “plan,” “potential,” “predict,” “project,” “should,” “target,” “will,” or “would” or the negative of these words or other similar terms or expressions. Tempus cautions you that the foregoing may not include all of the forward-looking statements made in this press release.

You should not rely on forward-looking statements as predictions of future events. Tempus has based the forward-looking statements contained in this press release primarily on its current expectations and projections about future events and trends that it believes may affect Tempus’ business, financial condition, results of operations, and prospects. These forward-looking statements are subject to risks and uncertainties related to: Tempus’ financial performance; the ability to attract and retain customers and partners; managing Tempus’ growth and future expenses; competition and new market entrants; compliance with new laws, regulations and executive actions, including any evolving regulations in the artificial intelligence space; the ability to maintain, protect and enhance Tempus’ intellectual property; the ability to attract and retain qualified team members and key personnel; the ability to repay or refinance outstanding debt, or to access additional financing; future acquisitions, divestitures or investments; the potential adverse impact of climate change, natural disasters, health epidemics, macroeconomic conditions, and war or other armed conflict, as well as risks, uncertainties, and other factors described in the section titled “Risk Factors” in Tempus’ Annual Report on Form 10-K for the year ended December 31, 2024, filed with the Securities and Exchange Commission (“SEC”) on February 24, 2025, as well as in other filings Tempus may make with the SEC in the future. In addition, any forward-looking statements contained in this press release are based on assumptions that Tempus believes to be reasonable as of this date. Tempus undertakes no obligation to update any forward-looking statements to reflect events or circumstances after the date of this press release or to reflect new information or the occurrence of unanticipated events, except as required by law.