Interview with Dr. Sara Hamis

Dr. Sara Hamis is an Assistant Professor in Machine Learning, a Wenner-Gren Fellow, and a Beijer Researcher at Uppsala University. She leads the Artificial Intelligence and Mathematics for Oncology (AIMOn) research group, where her work focuses on developing mathematical, statistical, and computational methods to better understand and predict cancer dynamics. With a PhD in Mathematics from Swansea University and postdoctoral experience in the UK and Finland, Dr. Hamis brings international expertise to the growing field of mathematical oncology.

Isaac

Hi, Dr. Hamis. Thank you for speaking with me today. To start, what initially sparked your interest in mathematics, and what inspired you to apply it to study something as complex as cancer?

Dr. Hamis

Well, first of all, hi, Isaac. Thanks so much for inviting me to this interview. For me, I've always loved mathematics. Most of the games I played as a kid were centered around math. However, biology always perplexed me in school. It wasn’t until university, when we started studying biology through the lens of mathematics, that I truly began to appreciate biology. So, for me, it was math first, biology second, and now I’m just really grateful that I get to use my passion for mathematics to study something as important as cancer.

Isaac

Yes, it seems that many people I've spoken to in the field of mathematical oncology were initially drawn to math first, and then developed an interest in biology later on.

Dr. Hamis

Yes.

Isaac

For students who are new to mathematical oncology, how would you explain what it is and why it’s such an exciting and promising field, especially when it comes to advancing cancer research?

Dr. Hamis

Yes, that's a great question. The way I think about it is that you have cancer research as its own field, and then you have mathematics, which is a separate field. When you combine them, you get this overlapping region, like where my fingers are, and that’s where mathematical oncology exists. It’s the intersection of both cancer research and mathematics, and in this space, something really exciting happens.

One exciting aspect is that when you start to look at cancer research through the lens of mathematics, you uncover patterns about the underlying biology that you can’t just see from the data. But with mathematics, we actually learn a lot about cancer as a system, and that’s really cool.

Another exciting thing is that, because cancer is so complex, we need new mathematics to fully understand it. In turn, cancer itself sparks the development of new mathematical approaches. So, it’s exciting from both directions.

Isaac

Could you speak a bit about the types of data you use in your research models? Do you work with data like images or DNA, for example?

Dr. Hamis

Yes, sure. So now I'm in a really cool position where I analyze a lot of different types of data. We have data on gene expression, for example. We’re looking at single cells and identifying which genes are expressed in each cell, so that’s on a very small scale. But we also analyze larger-scale data.

For instance, we have data on all the people in Sweden who have been diagnosed with cancer, including what medicines they’re taking and what they have in common. So it's really a transition from micro-scale to macro-scale data. In between, we also work with images of cells and tumor growth in mice. We’re covering the whole spectrum, and that’s really exciting.

Isaac

When you conduct research involving the entire population of Sweden and examine various types of cancers, how do you balance using an average model versus an individual patient model?

Dr. Hamis

Yes, that’s a really good question, and there isn’t a single answer for it. A lot of our current work is focused on tackling exactly what you’re asking: How do we capture individual patient behavior from population data? One approach I like to use, quite broadly speaking, is a Bayesian or probabilistic approach, where we include uncertainty in our model. By doing this, we can account for some of the variability between patients while still integrating the broader population data we have.

Isaac

How does Bayesian statistics differ from the regular statistics you might learn in high school?

Dr. Hamis

Yes, with Bayesian statistics in cancer research, you can account for the variability between patients or mice. In contrast, the way we often learn statistics in school is through a frequentist approach.

For example, you might run an experiment, simulate it 1,000 times, and then take the average to determine the outcome. This frequentist approach works well when you have a lot of data, but in cancer research, we often don’t have that luxury. Sometimes we have plenty of data, but other times we don’t. Bayesian statistics, however, can account for this variability.

Another interesting aspect of Bayesian statistics is that you can incorporate prior knowledge into patient models. You start with a prior assumption, include it in your model, then adjust that assumption based on the data you observe.

Isaac

I think I understand. In my study of Bayesian statistics, it seemed like you start with an initial assumption—like, for example, your brain might say the probability of rain on a random day is 10%. But if you see clouds in the sky, you would increase that probability because you're adding new information. Is that how it works? You update your model with new data, which changes your prediction about what will happen?

Dr. Hamis

Yes, exactly. So you start with an assumption, and then you look outside and see that it’s very cloudy. At that point, you can update your knowledge and think, "OK, 10% is no longer a fair estimate." You’re observing a lot of rain clouds, so it makes sense to adjust your belief accordingly.

Isaac

In your research, you use a lot of advanced tools like agent-based models and AI. For students who are new to this, could you explain what some of these tools are and how they help you understand cancer?

Dr. Hamis

Yes. So when we have a lot of data, like with population-level models, sometimes the amount of data is so large that it’s impossible to analyze everything at once with the human eye. AI, or artificial intelligence, is really good at finding patterns in that data. A human simply can't process that much information. So, a lot of what we do with AI is use it to identify patterns in large datasets, and then we interpret those patterns with our human minds.

Isaac

In comparison, when you have very low data versus a lot of data, how does that affect your work? I mean, which one do you prefer more, or does it just depend on the situation?

Dr. Hamis

Yes, so because I trained as a mathematician, I’m more comfortable working with small data systems because that’s what I’m used to. But since I started working in this field, so much has changed. The field of machine learning has evolved, and there’s a lot more imaging data now.

When I started my PhD, the available data was much less than what we have today. So I’ve had to adjust. Now that there's so much data, of course, we need to use it. I’ve transitioned from being a “small data” person to a “big data” person.

Isaac

In your research, you've mentioned that models sometimes miss the mark when complex biological processes come into play. How do you decide which factors to include in your models to make them more accurate? And does having large datasets help with this?

Dr. Hamis

That’s a great question. One of the challenges with mathematical biology or mathematical oncology is that there isn’t a "theory of everything." So every time we approach a research problem, we start by looking at the data we have and the specific research question. Based on those, we design an initial model.

A reasonable way to approach this is to start with a simple model and add complexity as you go along. If you make your model too complex from the start, you might run into issues like overfitting, where including too many irrelevant details can make the model less useful when applied to other systems. So it’s better to keep the model simple at first, then incrementally add more components as necessary.

Another important thing is to work closely with people who have domain knowledge, like medical doctors or experimental biologists, to ensure the model is grounded in real-world expertise.

Isaac

So with large datasets, would overfitting still be a concern? For example, if you try to calculate a model based on an average, does it make sense to use an average when everyone is different in their own way?

Dr. Hamis

Ah, good question. Let me explain. Let’s say we’re trying to determine the appropriate medication dose for a patient. We might consider variables like height, weight, age, and other factors. If we include too many variables in our model, it could become overfitted to a specific person.

For instance, if we create a model that’s too detailed and fit it to one individual, it would work well for that person but might not apply well to others. On the other hand, if we use fewer parameters and base the model on a broader population, it’s less likely to be overfitted. The key is ensuring that we only include the most relevant information in the model, leaving out variables that don’t add meaningful value. This way, we avoid overfitting and create a model that’s more widely applicable.

Isaac

So, if there are too many variables, it might matter a little for one person but not be as meaningful for someone else?

Dr. Hamis

Exactly. Of course, you would need to test the model, and it’s going to vary from case to case. That’s the tricky part. But generally, yes, we try to focus on the most impactful variables and test them to see how well they apply to different patients.

Isaac

You were awarded the LELA Siegel Prize for your work on how cancer cells respond to treatment. Could you explain your paper in simple terms?

Dr. Hamis

Yes, sure. When you study cancer or try to develop a new drug for cancer, one thing that happens is you start by looking at how the drug performs on cells. Let’s say you’re in a lab, and you have a dish with some cells in it. You add the drug and see what happens. This is often the first step—a step we call in vitro. Then you move to testing in a mouse, which is a step called in vivo. 

The in vitro and in vivo systems are very different, and sometimes when you move a drug from the dish to the mouse, a lot of things change. You have to consider many factors.

What my paper was about is how we can take the information we get from in vitro experiments and transfer it to the mouse experiments. It's about how we can move from here to here in a way that makes sense.

Isaac

Going back to the differences between in vitro and in vivo settings—when you move to in vivo, you have to account for additional biological processes. For example, a drug might travel to the liver and become less effective. Could you talk about how you account for those kinds of changes in your modeling?

Dr. Hamis

Yes, for sure. There are many cancer problems solved in mice that don’t work the same way in humans, so it’s definitely an important challenge.

Isaac

Your work also aims to predict how well treatments might work for individual patients. How close are we to using mathematical models to personalize cancer treatment? And what could that mean for patients in practical terms?

Dr. Hamis

Yes, that’s a tricky question because it depends on what you define as a mathematical model. Mathematics is already being used in medicine, but not always in a way that looks like equations. For example, radiotherapy planning or calculating medication dosages is based on mathematical principles, even if it doesn’t always seem like it.

Now, with the rise of large data sets and the ability to gather a lot of personalized information from patients, mathematics will play an increasingly important role in making sense of that data. So, while the role of math in personalized medicine is growing, it's hard to say exactly how far we are from it being widely implemented.

I think math is already involved more than people realize, but as data collection becomes more advanced, math will play a more prominent role. That’s one of the reasons I find this field so exciting.

Isaac

I’ve heard from other mathematical oncology professors that personalizing treatment could make treatments more effective while minimizing harm. I’ve always wondered how that would work.

Many students interested in this field would love to see it happen. But it's hard to predict exactly when or how it will be implemented.

Dr. Hamis

Yes, it's true. But we can already see a push for personalized cancer treatments. And now that we have more data available, it’s becoming more feasible than ever to personalize treatment beyond just things like age or weight.

So I think that’s where we’re headed, and it’s exciting to see.

Isaac

Now, shifting gears to your international work: You're currently at Uppsala University in Sweden, but you've also worked in several other countries. How does international collaboration shape your research?

Dr. Hamis

For me, it’s essential. I did my PhD in Wales, then worked in Scotland, Finland, and even spent some time at a cancer research center in Tampa. In Finland, I worked in both an ecology group and a cancer research group, and those experiences taught me a lot.

I learned that cancer can be thought of in an ecological way, where different cancer cells coexist in a tumor. It was valuable to learn from both ecology and cancer research, and now I’m working with machine learning, which adds another exciting layer.

Working in different environments has allowed me to gather diverse knowledge and spread it, and I think that’s a great thing about international collaboration.

Isaac

It sounds really exciting, especially since a lot of research materials can come from different places. You’ll have research done in Texas, materials from New York, and papers in California—it’s amazing to see it all come together.

Dr. Hamis

Yes, definitely. The mathematical oncology community is very international, and there's a lot of collaboration. It’s a great field to pursue a career in.

Isaac

What advice would you give to high school students interested in using math in medicine?

Dr. Hamis

I’d say, first of all, keep studying math! It’s important for everyone to understand math because it’s everywhere, even in how society works. Whether it's democracy or technology, math plays a crucial role.

There are also a lot of open-source educational resources now. So it's easier than ever to start exploring mathematical biology or mathematical oncology. I’d recommend diving into those online resources and keeping up with math.

And when you get the chance, look into books or courses on mathematical oncology. It’s all about building your foundation in math and applying it to real-world problems like cancer.

Isaac

That’s great advice. I’m always curious about how others got into the field, and your experience shows that it's possible to pursue this path from many different angles.

Dr. Hamis

Yes. One thing I wish was available when I was a student was the book Introducing Mathematical Biology by Alex Best at the University of Sheffield. It’s an interactive, online textbook that helps you work through math problems and includes Python code. It's a great way to connect theory with practical simulations.

For me, those kinds of "toy problems" were really fun because you’d write math on paper, but then the simulations would back it up. It was a cool way to see the math in action, and I highly recommend checking out that book and others like it.

Isaac

Thank you for the recommendation. It's always helpful to hear people’s advice on how to get into the field.

I have one last question: Looking ahead, what’s one big hope or dream you have for the future of mathematical oncology?

Dr. Hamis

Oh, can I share two?

Isaac

Yes, please do!

Dr. Hamis

One is related to education: I hope that mathematical oncology becomes more common as a subject. When I was in Scotland, they offered a course on mathematical ecology, but in the Nordics, it's not as common. I hope that changes, and that we start offering these courses at both university and even high school levels.

The second hope is for better communication between mathematical oncologists and cancer researchers. I want us to keep making mathematical oncology feel relevant and applicable to what cancer researchers are doing.

Isaac

That sounds crucial for the progress of oncology. I agree, it’s important to bridge those gaps. Thank you so much for speaking with me today.

Dr. Hamis

No, thank you! I’d also like to add that what you’re doing with your blog and YouTube channel is contributing to exactly what I’m talking about. Thanks a lot for that, Isaac.

Isaac

Thank you! I hope it makes an impact. It’s still growing, but I’m excited to see where it goes.

Dr. Hamis

It’s great. Thanks again.