How Can Functional Modeling, AI Accelerate Translational Breast Cancer Research?

Translating laboratory discoveries into clinical benefit for patients with breast cancer increasingly depends on combining genomic profiling with newer tools, including patient-derived functional models and artificial intelligence (AI).

CancerNetwork® spoke with VK Gadi, MD, PhD, deputy director of the University of Illinois Cancer Center and professor and director of medical oncology at the University of Illinois College of Medicine Chicago, about the state of translational oncology research at the 2026 Big Ten Cancer Research Consortium (CRC) Summit.

Gadi began by explaining where functional drug testing with patient-derived organoids stands today and how it complements genomic profiling. He then described the window-of-opportunity phase 2 BRE-04 trial (NCT04891068) of preoperative low-dose azacitidine (Vidaza) in high-risk early-stage breast cancer.¹ The trial is investigating whether epigenetic priming can make immunologically cold tumors more visible to the immune system.

Gadi also discussed how clinicians might prioritize targeting PIK3CA and ESR1 mutations in hormone receptor (HR)–positive metastatic disease. He covered the bottlenecks in moving laboratory discoveries into the clinic and how the Big Ten Cancer Research Consortium can help close that gap. In conclusion, he identified AI as the translational advance he expects to have the greatest effect on breast cancer over the next 5 years.

CancerNetwork: Where does functional drug testing stand today, and how close is it to routine use in guiding treatment?

Gadi: One of the things people have forgotten is the origin of our focus on genomics and how it would educate us in making treatment decisions. In the old days, we would take cancers, do our best to sequence them, find mistakes, and hope that some of those mistakes were actionable. Then we would start treating patients based on that.

The early experiments from those days showed that maybe 5% to 10% of patients could benefit that way, but 90% of patients’ cancers were not addressed.³ There was far more complexity than those methodologies could deliver at the time. They have improved, we have many more tools, and that number has inched up. There is still a gap, though, between what a cancer is doing in a human and what you see when you look at it from a genomics perspective. This is where functional modeling comes in.

One of the things you are seeing now at the [National Cancer Institute (NCI)] is an emphasis on this.² The current NCI director is also a researcher in this space, so there is a movement toward taking living systems and understanding how a cancer works while it is still alive. Taking that knowledge and applying it to patients is a complex phenomenon.

The easiest way I explain it is with an anecdote. It is very rare to find somebody on the street using a map to figure out where to go next, and that is essentially what genomics does. It lays out the map. It tells you which roads are blocked and which are open, and you are left trying to guess what the traffic pattern is going to be.

Today, that is not how we travel at all. We open an app on a phone that tells us what the weather is. It tells us that the street we thought we would take now has an accident, so we should go another way. That is the kind of map we need of cancer, and that is essentially what functional modeling provides.

That offers 2 benefits. First, we can make real-time assessments for the patient sitting in front of us, so functional modeling becomes a diagnostic tool for selecting the next treatment that patient should get. I was one of the founders of one of the first companies to do this, but there are now many companies in that space doing this for different kinds of cancers.

Second, we can take genomic information and validate it in these living models. That may give us a truer hit, which we can then pour resources into to see whether it should be the next new drug. The target might be shared not just within breast cancer or lung cancer but across tumor types, so you may identify patients whose cancers have a vulnerability in that pathway. That has been the hope for functional modeling, stretching back 10 to 15 years now.

How do you see functional testing fitting alongside genomic profiling? Are they complementary or competing approaches?

Functional testing and genomics are complementary, and I caution people against relying on one over the other. Genomics can clearly inform functional testing about the right next thing to do, and it also works the other way around.

At scale, functional testing is hard. It is what I call a low-throughput strategy. You have to get a tumor, grow it in a culture dish, and then expose it to 100 drugs or what have you. There are limits to how quickly, efficiently, and accurately we can do that. Some people can, but it is a low-throughput exercise.

Genomics is the other end of the spectrum. The cost of goods for this work has been enormously reduced; we are talking dollars per patient. The bioinformatics is hard and challenging, but AI is going to help us with that as well.

You have a high-throughput way of getting a lot of knowledge, which may not have all the insights you need, and a lower-throughput way that may give you maximal knowledge. We just have to integrate the 2. Then we will see patterns that make sense and can make treatment decisions based on them.

Your window-of-opportunity trial of low-dose azacitidine before surgery in high-risk early breast cancer is investigating epigenetic priming. What is epigenetic priming, and what is the trial testing?

That trial does a couple of different things. First, to answer what epigenetic priming is: we know that when you alter methylation and histone acetylation in cancers, or in any cells, you open and close the chromatin in a way that turns genes on and off.

One of the most heavily suppressed elements in our genome is ancient retroviral genomes. Some huge percentage of our genome is what we call junk DNA. It has all been silenced for a reason: if those genes get turned on, they are highly inflammatory.

Drugs like azacitidine, a demethylating agent, essentially open the chromatin, and you then inadvertently express these ancient retroviral genomes. They make funny-looking proteins, and funny-looking proteins should be very interesting to the immune system.

This trial essentially tests that hypothesis in otherwise immunologically cold tumors, which are tumors that are just not interesting to the immune system. Can we [assess] them a little by opening the genome in places where it should not be open? We presented preliminary data at [the American Association for Cancer Research (AACR) Annual Meeting], and we have shown that this is exactly what is happening, at least in some patients. Now the goal is to show that we can generate biomarkers for this type of strategy to see whether it could advance to clinical use.

That does 2 things. First, it tells me whether a hypothesis can be true in a unique study design, where we use the window of opportunity before patients get standard-of-care surgery to ask what we can learn from the cancer before we take it out. This is a low dose of treatment, and we are getting biologic markers. Once we prove the hypothesis this way, we can design larger studies and ask the question more formally: Is this a drug we should be incorporating into standard care for early-stage breast cancer?

Second, this is a platform study, and it is modular. Once we find the answer for one drug, I can take it out, put in another drug, and ask a similar question, because all the biomarkers are stable. We can run the study iteratively with different drugs coming in and out and validate drugs for future study in larger trials.

As ESR1 and PIK3CA co-mutations become increasingly common in HR-positive metastatic disease, how should clinicians approach testing, timing, and treatment sequencing?

This is a question about prioritizing which mutations to target. We know that PIK3CA is what we call a truncal mutation. It is present at diagnosis in about 30% to 40% of all breast cancers, and in some of those cancers it is an actual driver. In other words, it is leading to resistance to our therapies and promoting the cancer's growth.

ESR1 is different. It is an acquired mutation that results from selection pressure from the treatments we have been giving. If I give antiestrogen treatment over time, a cancer cell that wants to survive may develop an ESR1 mutation.

These mutations can occur independently, but they can certainly occur on top of each other. In the metastatic setting, we see these mutations in about 10% to 15% of [patients with] HR-positive, HER2-negative disease. We also see them in HR-positive, HER2-positive [groups], and we see PIK3CA mutations in triple-negative [groups]. These mutations are not unique to any one cancer.

In the HR-positive, HER2-negative setting, we have agents that have been studied individually to target these mutations, and a clinical trial has validated that combining them can also be effective. We have FDA approvals in the first line where we can do that. Those drugs are not ideal, because they have complications, and some patients may not know they have the mutation until later. In that case, these drugs can also be used in combination in later lines of therapy.

We are still exploring this therapeutic space and figuring out the prioritization. If you think the ESR1 mutation is the driver because it just emerged, it makes abundant sense to go after it. If you think PIK3CA is the reason the cancer has been resistant to your therapies the whole time, you should go after that, until we have more data showing that we can safely combine these therapies.

What is the biggest bottleneck in moving a laboratory discovery into a clinical trial, and what can consortia such as the Big Ten CRC do to facilitate that?

One goal of many academic institutions is to take internally developed ideas and assets and give them patient impact. Those assets might be a diagnostic tool, a drug, an app, or an AI algorithm. We call this process translation: getting something from one step to the next, and the next after that. Most universities do not have a sufficient ecosystem to do this, and some communities do not either.

By ecosystem, I mean not just the smart people who came up with the idea. It also includes the chemists who can make the drug, the biologic modelers who can figure out how to make the biologics, and the computational experts. At the terminal end, it includes knowing how to start a company or license the technology to one, how those negotiations go, and how to bring in venture capital and early-stage investors. That entire ecosystem needs to exist.

At our university, we have found those people, and we have a strong track record of doing this. At our cancer center, we formed essentially a mini-Shark Tank. We identify innovators in our ecosystem and find out what step they are at. If they have the interest and appetite to move their technology down the translational pipeline, we are there every step of the way, telling them what the next step is and connecting them with the expertise to get past it.

Not every idea is a winner. That does not mean an idea is untrue or scientifically unimportant; it just means that in its current iteration, it is not ready for patient impact. That is not a judgment, just a fact. Sometimes we have to give the tough feedback that an idea is not ready, ask the innovators to refine a few things, and then revisit it iteratively.

This is how we are doing it, and we are modeling this approach for all the Big Ten schools. They have the pieces, and hopefully more of them can organize around these themes. Some are already doing so and have a tremendous track record of taking early discoveries further.

References

1. BRE-04: window of opportunity trial of preoperative low dose azacitidine in high-risk early stage breast cancer. ClinicalTrials.gov. Updated June 23, 2026. Accessed October 22, 2026. https://tinyurl.com/4wn878jw
2. Letai A. Could functional precision medicine be the new precision oncology? National Cancer Institute. August 6, 2026. Accessed October 22, 2026. https://tinyurl.com/yebcrmtr
3. Marquart J, Chen EY, Prasad V. Estimation of the percentage of US patients with cancer who benefit from genome-driven oncology. JAMA Oncol. 2018;4(8):1093-1098. doi:10.1001/jamaoncol.2018.1660