Interview with Dr. Jana Gevertz

Dr. Jana Gevertz is a Professor in the Department of Mathematics & Statistics at The College of New Jersey, specializing in mathematical oncology. She develops and validates mathematical models to optimize cancer treatment strategies, focusing on areas such as treatment resistance, virtual populations, and experimental design. She has received national recognition for her contributions to both research and teaching in mathematical biology.

Isaac

Hi everyone, and welcome back to the High School of Mathematical Oncology’s YouTube channel. Today, I’m very excited to introduce our guest, Dr. Jana Gevertz, a professor of mathematics at The College of New Jersey, where she specializes in mathematical biology—focusing especially on cancer modeling and immunotherapy. She recently received the Society for Mathematical Biology’s Distinguished Service Award for her major contributions to the field. 

Dr. Gevertz, thank you so much for joining us today. To start off: what first inspired your interest in mathematical oncology? And were there any pivotal moments in your career over the past 20 years that have shaped your path and the way you approach your work?

Dr. Gevertz

Yeah, there are almost two separate answers to that question, so I’ll start with what got me interested in mathematical oncology to begin with.

When I went to college, I had absolutely no idea what I was going to major in. I know some people go in already having a career in mind—I didn’t. But I knew I really liked math and biology. I wasn’t really aware that there was much of an intersection between the two fields, so I thought I had to choose between them. After going back and forth, I finally landed on mathematics as my major.

In my junior year, I found a class called “Differential Equations in Biology,” and I thought it was cool, so I took it. At the same time, my dad was diagnosed with cancer. He had a colleague twenty years younger than him who was given the same diagnosis and went through the same treatment protocol. My dad ended up being what you would call a “super-responder”—there was no evidence of disease after just one round of chemotherapy, even though it was stage 3. They kept treating him for a bit longer because that was such an unusual response. His colleague, despite going through eight rounds of the same treatment, sadly passed away.

That question—why did two people with the same diagnosis and same treatment respond so differently—just wouldn’t leave my brain. I knew I wasn’t going to answer it by becoming a medical doctor, but I saw mathematics as a way to try to understand the heterogeneity in treatment response. That series of events is what brought me to mathematical oncology.

We’re the lucky ones—my dad’s still here 25 years later. It’s amazing.

In terms of how I approach my work, one major shift I’ve seen in the field since I started is how models have evolved. When I first entered the field, the models were very theoretical. You’d develop a framework, analyze or simulate it, and make broad qualitative predictions. But there wasn’t much connection to real-world data, or to actually using models to influence experiments or treatments.

The field has since moved toward data-driven models. We now try to connect models directly to data and use them to guide treatment strategies. It’s an exciting evolution.

Isaac

Yeah, that’s such a powerful story. It’s inspiring to hear, and I’m glad your dad got better. It’s also interesting how mathematics and biology converged for you personally, and how math in general has gone from being really theoretical to now actually helping guide treatments. I’d like to dig into how you build these mathematical models.

When building a model, how do you decide which aspects to focus on—especially when balancing the need for simplicity with the biological complexity of cancer?

Dr. Gevertz

Great question. Cancer is incredibly complex. If you try to model every component, you’ll end up with thousands of variables and hundreds of thousands of parameters. The model becomes intractable—you can’t analyze it or learn from it. But if you oversimplify, you won’t learn anything meaningful either. It’s always a trade-off between complexity and simplicity.

My general approach—easier said than done—is to make the model as complex as necessary to answer your question, but no more than that.

For example, I’ve worked on several models involving the same immunotherapy drug. In one case, I was interested in finding the optimal way to combine it with another drug. For that, I didn’t need the detailed mechanism of how the drug works—I just needed the outcome. But in another study, I wanted to understand part of the mechanism itself. So we had to add more complexity to the model, because the question demanded it.

So it’s always about tailoring the model’s complexity to the question and to the data you have. If I don’t have spatial data on how a tumor evolves, but I do have data on size over time, I won’t include a spatial component. There’s no point in modeling something I can’t support with data.

Isaac

Yeah, that makes a lot of sense—just focus on what you need to answer the question.

Dr. Gevertz

Of course, the challenge is you don’t always know exactly what you need. But that’s what you try to figure out.

Isaac

Right. It’s still complex, but at least it’s understandable.

Dr. Gevertz

Exactly. That’s the balance we seek.

Isaac

In your work, you’ve mentioned concepts like parameter identifiability and sensitivity analysis. What do those mean in practice, and how do they help you determine whether your model is reliable or needs adjustment?

Dr. Gevertz

Another great question. It comes down to making sure the model reflects something we see in the real world. For example, if I want to predict how a tumor responds to a drug, I need to calibrate the model to match actual data. That means adjusting the model’s parameters—constants within the equations—so that the output fits the data.

Sensitivity analysis asks how much your model’s output changes when you tweak one of those parameter values. Suppose there’s a parameter that controls tumor growth rate—if changing it slightly ruins the model’s fit to the data, it’s a sensitive parameter. Others you can change a lot, and still get a good fit.

Identifiability is related. A parameter is identifiable if your data lets you determine its value with confidence. If you don’t have enough data to do that, it’s not identifiable.

This is important, because even if you can fit your model to data using a range of parameter values, you may not be able to trust its predictions. You might have two different parameter values that both describe the training data well, but lead to very different predictions in a new scenario.

So if a parameter isn’t identifiable, you risk making unreliable predictions when applying the model to other conditions—like combining drugs or changing treatment protocols.

Isaac

That’s really eye-opening—how a tiny change in a parameter can shift the model so much in how well it reflects the real world.

Dr. Gevertz

Right. If it’s a sensitive parameter, at least you know you need to get that value right. And when you do, you can trust your predictions. But the flip side is, some parameters are hard to measure accurately. You don’t want to make the model overly sensitive to something you can’t measure well. These are difficult trade-offs.

Isaac

I suppose cancer cells don’t make it any easier since they constantly adapt to their environment.

Dr. Gevertz

Exactly. Cancer cells adapt, and the disease itself is highly heterogeneous. You might get one result from a biopsy in one tumor region, and a completely different one from another region. That makes it even harder to measure parameters accurately and adds another layer of complexity.

Isaac

Could you explain the difference between drug-induced resistance and pre-existing resistance in cancer cells? Allso, how does phenotypic plasticity add more difficulty to cancer treatment?

Dr. Gevertz

Sure. So the paradigm we used to have for resistance to a drug—whether chemotherapy or any therapy—was that there’s a small population of cells in the tumor that, before you even give the drug, already have properties that make them resistant to it.

What would then naturally happen is: when you give a drug that kills most cancer cells, the tumor shrinks significantly because the sensitive cells die off. But those few resistant cells—already in the tumor—start growing and eventually take over because they’re unaffected by the drug.

In theory, if resistance were only pre-existing, then if you could find a second drug that hits those resistant cells, you could eliminate them too. Maybe you’d need a third drug, but you could eventually hit all the subpopulations. It might become too toxic if you're combining too many drugs, but at least, theoretically, you could overcome the resistance.

Now, we’re gaining a better appreciation for cancer cells being phenotypically plastic, as you mentioned. What that means is: they can change their behavior in response to environmental stresses. Drug exposure is one kind of stress. Others include low oxygen or low nutrient levels.

When exposed to stress, cells don’t necessarily mutate, but they do change which proteins they express. That change alters their behavior. So, a cell that wasn’t initially resistant to the drug might, because of the drug-induced stress, change its behavior and become resistant. In that case, the resistance wasn’t just selected—it was induced by the treatment itself.

So, simply switching to another drug might not work because the resistance is a response to the first drug being introduced. This concept of plasticity—of treatment actually creating resistance—is a much harder challenge when designing effective protocols.

Isaac

It’s wild to think that cancer cells can just reprogram what proteins they express in real time.

Dr. Gevertz

What’s really crazy is that cancer hijacks normal biological mechanisms. During development, your cells need to be plastic so they can adapt to changing conditions. During wound healing, you want cells to respond and change behavior. These are healthy, adaptive systems—but cancer co-opts them for its own benefit.

Isaac

Yeah, it’s like those Kurzgesagt videos I watch on YouTube—they talk about how insanely adaptive cancer can be.

Dr. Gevertz

Yeah, it’s almost like the cells are smart in how they figure it out.

Isaac

So in the clinic, doctors usually only have access to the total tumor size—we have such limited information. How do you figure out what’s actually happening inside the tumor, like the number of drug-resistant vs. drug-sensitive cells, and how they’re adapting?

Dr. Gevertz

That’s a really difficult question. Biotechnology is developing to the point where, if you could apply all available tools to a patient’s data, you’d be able to do things like genetic barcoding to track how different cells evolve over time. We also have ways now to identify rare cell populations with unique behaviors.

The problem is, those tools aren’t available to the average patient. They’re expensive and not part of standard clinical care.

So the real question becomes: what is the minimal amount of data we need to get from a patient in order to build predictive models that are reliable? That’s what we’re actively working on. This is very much a work in progress—no one has completely solved it yet.

But as biotech advances and more types of data become available, math has a role in identifying which tools are most important to apply—so we get just enough information to make good predictions for that individual patient. We’re not there yet, but that’s where we’re headed.

Isaac

That’s a really cool detective story—using whatever data is available and letting biotech evolve alongside modeling.

Dr. Gevertz

Exactly. Costs are coming down as well. But for now, a lot of these techniques are still experimental and used only in research labs—not in clinical settings.

For example, researchers might use xenograft models where a patient’s tumor cells are put into a mouse, or they might work entirely with preclinical data in petri dishes. We’re still in the process of translating that kind of research into direct patient care.

Isaac

Yeah, so it’s a big leap going from the lab to actually helping patients. May I ask about how models improve treatments? How does mathematical optimization help identify the best possible treatment strategy for a patient? And could you share a real example from your research?

Dr. Gevertz

Sure. Let me start with the ideal theory, and then I’ll talk about how it’s implemented in practice.

Imagine you have a patient getting a cocktail of three drugs. You need to decide: when should each drug be given, and how much of each should be given? That’s a question of both timing and dosing. Should they be given together or separately? What’s the optimal schedule?

If you tried to test all those combinations in a clinical trial, it would be impossible—there are just too many. That’s where a predictive mathematical model comes in. If we have a model that accurately describes how patients respond, we can use math to optimize the treatment schedule. The model explores the possibilities, not human experimentation.

Once we’ve identified promising strategies in the model, we bring those ideas back to the clinic for further testing. That’s the ideal approach: use the model to narrow down the best treatment strategies without putting patients at risk.

In practice, it’s still a work in progress. Many mathematical predictions haven’t yet been translated into clinical care. But we’re starting to see clinical trials that incorporate model-based decisions. These usually happen in institutions where mathematicians and clinicians work closely together—like at Moffitt Cancer Center or MD Anderson.

In some of those cases, a patient comes in monthly for imaging, and that data is plugged into a mathematical model. The model then helps decide whether the patient should receive treatment, how much, or whether they should wait. Models like that are beginning to guide dosing and scheduling.

We still have a long way to go before this becomes standard care. For most patients today, mathematical modeling doesn’t play a role in their treatment decisions.

Isaac

It’s amazing how math can help find safer, more effective treatments—and even predict how the patient might respond.

Dr. Gevertz

It also requires building trust between clinicians and models. Say the model suggests not giving a drug this week—the patient might think, “Wait, aren’t you treating me?” Or if it says to reduce the dose, the patient might say, “Why? Don’t I want to kill the cancer?”

So it’s not easy to go from a model’s suggestion to real clinical action, especially with a seriously ill patient in front of you.

Isaac

Yeah, and I guess building that trust really depends on continuing lab experiments and validating the model predictions. How do your mathematical models interact with lab experiments, like single-cell analysis?

Dr. Gevertz

None of my own work has gotten into the single-cell level. I focus more on tissue-level responses—how the tumor as a whole responds to treatment protocols or how resistance evolves across the tumor mass.

That said, there’s a lot of great single-cell data being generated that can inform treatment and resistance dynamics. The researchers who work with that kind of data typically build what we call multiscale models—models that link what’s happening inside an individual cell to what’s happening at the tissue level.

There’s beautiful work in that space, but it’s not the focus of my research. I’m more interested in what we can learn at the macroscopic scale to guide patient treatment.

Isaac

Yeah, that sounds really impressive. So on the scale of a tissue-level model, how would that sort of inform a biological study? How do the two work in combination?

Dr. Gevertz

Yeah. So let me give you an example. We had some data on how two drugs interacted with each other in two different non-small cell lung cancer cell lines. One of the cell lines had a particular mutation that the other didn’t have, and they responded very differently to the same combination therapy.

They had done one standard treatment protocol for each. Once we were able to build a model calibrated to how the bulk tumor volume changed in those two cell lines, we used our underlying math and optimization techniques to ask: for patients with this mutation, what would we recommend as a better treatment strategy? And what about for those without the mutation?

So we used that bulk data—how the tumors changed over time, depending on whether or not they had the mutation—to make different model-based treatment recommendations. But we never looked deeper into the individual cell lines themselves beyond knowing that one had a mutation and the other didn’t.

Isaac

What are virtual clinical trials, and how do they improve cancer treatment or help model virtual patient populations?

Dr. Gevertz

Good question. Let’s start with what a clinical trial is. In a traditional clinical trial, volunteers—some with the disease and some without—are recruited to test the safety and effectiveness of a new therapy. These trials rely on people being willing to participate and receive an experimental drug. They’re also time-consuming and expensive, since you have to recruit participants, treat them, and provide medical care throughout.

There are also issues with who participates. In the U.S., clinical trials are historically overrepresented by white men, even though we know people of different backgrounds and genders can respond very differently to the same treatment. You really want a diverse patient population, but that’s not always achievable in real trials.

That’s where virtual clinical trials come in. If we have enough data from a clinical trial, we can build a mathematical model that captures what we observed in the data. Then, we can run the trial on the model instead of actual patients. Each virtual patient is just a different parameterization of the model.

Because it’s much easier to vary parameters in a model than to recruit diverse patients, you can study the effects of a treatment on a more heterogeneous population—through equations, not actual people.

Isaac

That’s incredible—to hear how virtual patients and trials let you build off pre-existing data. Though I imagine it’s really important to have a solid understanding of the parameters for that.

Dr. Gevertz

Absolutely. One of the ongoing challenges is figuring out what makes a realistic parameterization for a virtual patient. If your parameters are too biased toward the original trial population, you won’t capture people outside that group. But you also don’t want to vary your parameters so much that they no longer represent any real human being.

So there’s a balance—designing a realistic and diverse pool of virtual patients. If you can do that well, then you can effectively run trials on your computer using equations instead of administering drugs to actual people.

Isaac

Yeah, and that’s a lot cheaper, faster, and safer.

Dr. Gevertz

Exactly. You're not exposing people to experimental therapies, so it's a much safer starting point.

Isaac

So as we're wrapping up the interview, I’d like to ask: what foundational skills—in math, coding, biology—should high school or early college students focus on if they’re interested in mathematical oncology?

Dr. Gevertz

That’s a great question. You’re right that this field is very interdisciplinary, so there isn’t just one skill you need.

First, you should have a genuine interest in the biological questions. If you’re not curious about how cancer works or how treatment protocols can be improved, this may not be the field for you. But if those questions excite you, you already have the foundation.

In terms of math and computer science, the main math we use is differential equations. If you’ve taken calculus, you’ve already learned the building blocks—derivatives and integrals. From there, you’d eventually want to take a course specifically in differential equations (sometimes called Calculus IV). That’s a core mathematical foundation for the field.

You’ll also need programming skills. It doesn’t matter too much which language you start with—once you understand the logic of coding, it’s easier to pick up other languages. I use MATLAB, but a lot of people now use Python. If you’ve learned Java in high school, that’s fine too. The key is learning how to structure algorithms and write code that executes a logical sequence of steps.

Isaac

That’s really helpful advice. And yeah, I think you also showed how important interdisciplinary learning is in mathematical oncology. For a lot of our audience, it's a long journey ahead, and research can be tough. There can be failures and setbacks. So, how would you support students through those setbacks, and what advice would you give to someone who wants to use math and science to make a meaningful impact in cancer research?

Dr. Gevertz

That’s such an important question. One of the biggest challenges I see when students enter a field like mathematical oncology is that it feels overwhelming—you feel like you need to know everything. All of math, all of statistics, all of computer science, all of biology.

But no one person is going to master all of that. You have to accept that when you work in an interdisciplinary space, there will be things you know well, and things you’ll need to rely on others for. And that’s okay. You’ll have collaborators and mentors to help with the parts you don’t specialize in.

Also, there’s no one path into this field. People come from all kinds of backgrounds—math, biology, biomedical engineering, computer science. So don’t feel like there’s one correct way in.

As for setbacks in research—this is not like doing homework. In a class, if you’re stuck on a problem, you can come to me and I’ll help because I know the answer—I wrote the problem. But in research, we often don’t know the answer. You might work on something for six months and realize the question wasn’t right, or that it doesn’t have a good answer. And that’s normal.

If the answer is obvious before you start, it’s probably not research. So when things don’t work out, that’s not failure—it’s part of the process. The important thing is to have the tools and mindset to pivot. If we discover we asked the wrong question, how do we find a better one?

So: patience, perseverance, and support—from mentors and collaborators—are essential.

Isaac

Dr. Gevertz, thank you so much for such a thoughtful and insightful conversation. You’ve given us an incredible look at how math can be used to fight cancer—from theoretical modeling to real patient impact—and how students like us can start building the skills to contribute.

For everyone watching, if you're excited about combining math, biology, and coding, I really encourage you to explore this field and check out some of Dr. Gevertz’s work.

Whether you’re interested in programming, biology, or solving equations, as Dr. Gevertz said, there’s a place for you in mathematical oncology. And thanks again, Dr. Gevertz, and thank you all for tuning in!

Dr. Gevertz

Thank you! I appreciate it. It’s great that you’re doing this channel. Reaching students early and letting them know that this field exists and how they can get involved—I would have loved to have had that. So I think it’s wonderful that you’re doing it.

Isaac

Thank you so much.