Mar 17, 2026 10:57 AM
The Evolution of Precision Medicine: A Conversation with Tom Turi from Flagship Biosciences
Participants: Tom Turi, PhD (CEO, Flagship Biosciences) Trent Carrier, PhD, MBA (CEO, Carolina Molecular), Jason Amsbaugh, MBA (CEO, Samba Scientific),
Overview
The transition from descriptive biology to quantitative precision medicine has been more than a technological evolution; it has been a complete paradigm shift in how we identify targets and treat patients. To explore this changing landscape, we recently sat down with Tom Turi, CSO of Flagship Biosciences and Jason Amsbaugh, CEO of Samba Scientific. Tom’s career offers a rare, front-row seat to this history—from pioneering genomic approaches at Pfizer during the early days of the Human Genome Project to leading AI-powered spatial biology today. In this conversation, we discuss the “mountains behind mountains” of biological data and why the future of clinical diagnostics might be found within the pixels of a standard H&E slide.
The Flagship Biosciences Mission
Jason Amsbaugh: I’d like to start by talking about Flagship Biosciences, where you currently serve as CSO. What is the story behind the company?
Tom Turi:
Flagship has been around for over 15 years. It began as a purely computational company focused on image analysis. During the development of their early image analysis algorithms, the team realized that the primary obstacle wasn’t the code itself, but the quality of the images that were being analyzed. Clients were sending in images from their laboratories that were highly variable in terms of staining quality and consistency and image capture, which confounded the algorithm development.
We realized that to truly improve algorithm development, they needed to standardize the entire image production pipeline. This meant controlling the process from the ground up—starting with the slides, performing staining and immunohistochemistry (IHC) in-house, and then moving through to the scanning onsite. Taking this process in-house led to the growth and development of our own internal laboratory, which is now a major part of our business. Today, we analyze tissues of all sorts, including tumor, neuro, and rare disease specimens. We are entirely disease-agnostic and are able to provide a high level of standardized data that powers our advanced computational models.
Jason Amsbaugh: What differentiates Flagship today, especially now that so many AI-based imaging companies have entered the space?
Tom Turi:
Competitors move into this space from different angles. You have technology-first companies that develop a tool and then realize they need an algorithm solution. Then you have algorithm-based companies that simply partner with external laboratories.
Flagship stands out as being one of the few end-to-end solution providers 100% focused on biopharma. We have been at this the longest as a combined service provider—owning both the wet lab and the image analysis algorithms. Like me, most of our scientific staff comes directly from biopharma backgrounds; we have lived in the drug development space and applied these biomarkers ourselves. We think like drug developers rather than just software engineers. We combine that with an operational mindset that prioritizes agility and high quality, allowing us to develop bespoke assays for clients who have very specific endpoints. We try to be consultative, because we understand the strategic decision-making process behind biomarker selection through the development of a companion diagnostic. Additionally, we can provide insights whether a companion diagnostic is cost-effective for a specific program, or if it’s even necessary.
The Early Days of Pharma Genomics
Jason Amsbaugh: You were one of the first to implement genomic approaches for drug development at Pfizer back in the 1990s. What did those early stages of pharma’s involvement look like?
Tom Turi:
That was back in 1994, during the very early stages of the Human Genome Project. We had just signed a relationship with Incyte Genomics. This period was predated by a massive investment by SmithKline Beecham (now GSK) with Human Genome Sciences, which was an exclusive deal. Pfizer was second; we completed the first deal with Incyte, and eventually, about 20 other large and mid-sized pharma companies followed our lead.
At that time, we were performing what might be considered early spatial genomics by sequencing cDNA libraries out of individual tissue types, such as the liver or kidney. By 1997, we began conducting single-cell sequencing analysis based on Jim Eberwine’s early publications. While those initial experiments didn’t yield much information due to low yield, they represented the very first spatial-like transcriptomic studies. We evolved from cDNA library sequencing to microarray technology, leading efforts to combine these readouts with computational approaches and integrated in vitro and in vivo technologies for target evaluation.
Trent Carrier: What was the experience of being a first mover in this space like? Did the complexity of biology reveal more obstacles or opportunities than you originally anticipated?
Tom Turi:
You always learn in this field. Often, we think we are smarter than we actually are until biology teaches us a lesson. Initially, we had tunnel vision regarding target identification. I could pull out a GPCR based on only three amino acids in a database. However, we were fooled many times. Early on, we thought there would be over 100,000 genes in the human genome, but we were humbled when it turned out to be approximately 25,000. Biology taught us that even when you think you know a lot, there is always a next level of inquiry: we know the gene exists, but where is it expressed, when does it appear, and how much of the protein is there?
The Philosophy of Innovation
Jason Amsbaugh: There appear to be two schools of thought around scientific innovation. Do you think that identifying new scientific problems drive development of new tools, or do new research tools and technology lead to the pursuit of previously unidentified problems?
Tom Turi:
You typically focus on the specific problem you are trying to solve. If I have a problem, I will build a tool to solve it. However, many things are developed along the way that have no immediate utility and are put on the shelf. Years later, you stumble across that technology and realize you finally have an application for it.
Microarray technology, for instance, originated from photolithography in the computer industry. Affymetrix borrowed that technology, and we figured out how to apply the chemistry of nucleic acid building blocks on top of it. Science often involves this kind of cross-pollination where a discovery from “left field” suddenly becomes the missing piece of a biological puzzle.
In the past, biology was primarily descriptive. Today, it has completely flipped; biology is a quantitative science. We ask how much of a marker is there, who else is present, and what combination of factors are at play. This shift makes the science much more precise because you are dealing with a discrete, quantitative entity.
The Evolution into Spatial Biology & Diagnostics
Jason Amsbaugh: When did you first start seeing the merger of genomic-based readouts with tissue-based spatial imaging?
Tom Turi:
It actually goes back to embryology 40 or 50 years ago. We should give more credit to the Drosophila researchers of the 1980s who used Lac-Z staining and fusions to show beautiful patterns in embryos, such as the fushi tarazu or “tiger stripes.” That was spatial biology in its finest form, even if we could only look at one gene at a time.
We have now reinvented that paradigm for oncology. For years, we ignored the interplay between the immune system and cancer cells. With checkpoint inhibitors like CTLA-4 and PD-L1, we saw distinct responders and non-responders. Spatial analysis allowed us to see if T-cells were being excluded or if they were invading the tumor. This relationship has profound clinical meaning, helping us understand why a tumor is “hot,” “cold,” or a “desert.” These clinical signals were the trigger point for the current explosion in spatial histology and genomics.
Jason Amsbaugh: How does that translate from discovery into a clinical diagnostic? Early PD-L1 antibodies were criticized for not being predictive enough — is that still the case?
Tom Turi: PD-L1 remains our best biomarker, but it is not completely informative for identifying the right patient. We are stuck in a paradigm where some drugs, like Keytruda, require a companion diagnostic while others, like Opdivo, use a complementary one. We still haven’t quite cracked the nut on a better, single biomarker, which is why the industry is migrating toward expression signatures and multiplex sets of markers. These signatures, such as MammaPrint for breast cancer, are useful tools, but they haven’t yet caused a broad paradigm shift in diagnosis.
Jason Amsbaugh: In the present world of multi-omics, are people effectively combining signatures from different modalities at the translational level to create more predictive diagnostics?
Tom Turi:
They are. Initially, these are discovery tools used to guide internal development programs. We are seeing sponsors move toward multiplex, multi-omic markers that combine protein-based and RNA-based biomarkers. This creates an assay that looks at genes and proteins simultaneously, requiring specialized approaches for co-registration on slides and sophisticated image analysis to ensure the readouts are accurate.
Trent Carrier: Do you see this multi-omics approach moving into the clinic? It seems that to get the FDA to agree on an endpoint, you have to strip away the complexity.
Tom Turi:
Complexity is not your friend in the clinic. Clinicians are used to looking at single markers like HDL or HER2. Medical technologists require simple, “push-button” systems that spit out a single number. When developing these technologies, you have to think with the end in mind: who will run it and where?
The FDA is open to innovation, but they require extreme rigor and the ability to explain every part of the process. They are uncomfortable with “black box” AI algorithms. However, recent guidance allowing panel-based companion diagnostics to go through the 510(k) process rather than the PMA process is a massive game changer for the industry.
Jason Amsbaugh: It’s interesting that NGS itself hasn’t evolved to the point where a patient tube goes in, a button is pushed, and a result is shared.
Tom Turi: That is because it is a system rather than just an assay. It requires the instrument, the reagents, and a standardized extraction process. Until we standardize the workflows, there will be churn. We have NGS-based companion diagnostics, initially they were offered though limited distribution models like Foundation One, where one company owns the entire system. More recently, IVD Cancer Gene Panels have come to market. However, we are still struggling to find the right set of signals—whether point mutations, structural variants, or methylations—to guide treatment.
The Future & Emerging Tech
Jason Amsbaugh: When you look back over the progress made in the past 30 years, is the current state of precision medicine where you thought it would be?
Tom Turi:
In 1994, we didn’t even talk about “precision medicine.” My “aha” moment came at Pfizer when we saw tumors literally melt off a patient’s body in a melanoma trial for a CTLA-4 inhibitor. Because the response rate was only 20%, we almost threw the drug away because we didn’t know how to identify who would respond. Thankfully, companies like BMS stuck with it and got Yervoy approved. Today, we have over 40 approved companion diagnostics. We are still in the learning phase, but we are chipping away at the iceberg one little sledgehammer at a time.
Jason Amsbaugh: Closing question: what is the most exciting emerging technology in the lab right now?
Tom Turi: Spatial transcriptomics and spatial proteomics will be revolutionary for identifying candidate biomarkers. Combining protein and nucleic acid-based biomarkers is key. I am also excited about the information we are finding straight off of H&E stains. Being able to predict mutational status or treatment response from a standard H&E slide would be a game changer, allowing for “virtual genomics” from a single picture. This will accelerate development and treatment by making high-level insights more accessible.
Conclusion
As Tom noted, the path toward true precision medicine is an ongoing process of “chipping at the iceberg one sledgehammer at a time.” The transition from discovery-phase complexity to clinical-ready simplicity is where the most impactful innovation happens today. At Carolina Molecular, we are proud to serve as a technology facilitator for partners like Flagship Biosciences, helping to turn high-level insights into standardized, actionable workflows. Whether navigating shifting regulatory guidance or pioneering “virtual genomics” from standard tissue samples, our goal remains consistent: getting the right treatment to the right patient at the right time. We invite you to join us as we continue to move these exciting technologies forward.