How Computing Can Battle Cancer: A Conversation with Dr. Eric Stahlberg

For three decades, Eric Stahlberg has worked at the intersection of computing and cancer, following a single question: how can the most powerful computational tools help solve one of medicine’s most complex problems? After helping lead the collaboration between the National Cancer Institute and the U.S. Department of Energy to bring high-performance computing into oncology at scale, he joined The University of Texas MD Anderson Cancer Center, where he now focuses on translating data science into patient impact.  

Stahlberg also founded and co-chairs the Computational Approaches for Cancer Workshop at SC, a long-running forum for the growing convergence of HPC, AI, and cancer research. SC26 spoke with Stahlberg about the path that brought him there and what it takes to unite computing and medicine around shared goals. 

SC26:  You started as a computer science major and have said that a chemistry professor steered you toward computational chemistry. What did that moment look like, and how did it change your trajectory?

Eric Stahlberg: It really was an evolution while attending Wartburg College, a small liberal arts college in my hometown of Waverly, Iowa. As a computer science major, I chose to take chemistry as my science general education requirement. Dr. Zemke was the instructor, and a classmate and I were both CS majors in his class. He gave us the opportunity to work on small projects using the new Apple IIe computers in the CS department. That was my first introduction to computational chemistry, implementing formulas and displaying results graphically on the computer.

My interest kept growing, and I added both chemistry and math as majors. Dr. Zemke gave me another opportunity to work on an independent study project with MOPAC, a computational chemistry software package of the time. It was a chance to apply the math I was learning and the computer science skills I was developing to actually study chemistry. That combination became a foundation I have followed throughout my career. 

SC26:  In 2001, you made what you’ve called “the jump to bioinformatics” at the Ohio Supercomputer Center. What was driving that?

Stahlberg: That’s an interesting little story. I first became familiar with Chemical Abstracts Service in the mid-80s, working as an undergraduate student programmer, and returned there in the late 90s after a postdoc at Argonne and several years at Cray Research and Oxford Molecular Group. Working on their flagship SciFinder product, it really began to hit home how important the biological context was becoming for computational chemistry. The molecules were no longer just abstract structures; they existed in the context of materials and biology. 

When the opportunity emerged to help lead a new bioinformatics effort at the Ohio Supercomputer Center, I jumped at it, figuring it was a way to learn more about that biological context. I wanted to learn the part I did not yet have. 

SC26:  What was the friction at that time between computing and life sciences research?

Stahlberg: Computing in life sciences was used heavily for sequence analysis, predominantly BLAST searches, with some applications in large-scale biomolecular modeling and simulation. The real limitation was an absence of individuals able to cross between subject domains, including HPC, life sciences, and sequence analysis. During my time at OSC, a group of us began to build this cross-over community, eventually establishing the Ohio Collaborative Conference on Bioinformatics. This was a five-year series of mini-conferences that brought together plant biologists, medical scientists, computer scientists, and others, all with an interest in bioinformatics across the state of Ohio.  While at OSC, I had the opportunity to continue envisioning and working on problems that would challenge the center’s large computers. One of those problems was to develop multiscale models of biological systems that would integrate and simulate cellular dynamics using computational chemistry methods such as molecular dynamics. We never got those proposals funded, but the conversations mattered and were really important for shaping later efforts.

SC26:  Walk us through how the NCI-DOE collaboration came together. What were the real challenges in getting two such different institutional cultures to collaborate?

Stahlberg: The NCI-DOE collaboration came together in a way that could not have been anticipated, and I want to acknowledge the many colleagues who helped make it real. 

I was at the Frederick National Laboratory for Cancer Research, directing a bioinformatics core, when it became clear that the level of computing available through DOE exascale systems was several orders of magnitude beyond what was being applied to cancer. I was fortunate to have a personal network that gave me a chance to act on that.

The opening came around 2013, when Warren Kibbe, then heading the NCI Center for Biomedical Informatics and Information Technology, supported two years of funding to explore future directions for HPC at NCI. When I officially started that effort, I reached out to DOE colleagues with a simple idea: modeling cancer. That is when things started to roll.The early challenge was helping people on both sides recognize their overlapping interests. The respective agency missions were not as different as they looked on the surface. The DOE was already asking how to approach non-engineering, data-intensive computing, and there was a massive, growing body of cancer data that required exactly that. Bringing individuals together from across the community was the basis for the Frontiers of Predictive Oncology and Computing meetings. As people began to see common ground, the teams started moving forward together.

Eric Stahlberg (right), with Dr. Ethan Dmitrovsky (left), former head of Frederick National Laboratory for Cancer Research, and Dr. Dwight Nissley (center), head of the Cancer Research Technology Program at Frederick National Laboratory, at the Summit supercomputer at Oak Ridge National Laboratory. Summit was among the DOE systems central to the NCI-DOE collaboration to bring high-performance computing to cancer research at scale.

SC26:  In 2015, you helped found the Computational Approaches for Cancer Workshop. What has changed most about the conversation between HPC and cancer researchers over those 11 years?

Stahlberg: Early on, it was commonly seen as somewhat far-fetched. This whole idea of, “What are we going to do with HPC and cancer?” But the seeds were there. Many of the early conversations at CAFCW focused on what the NCI and DOE were doing in computing for cancer: modeling and drug discovery, natural language processing of written pathology reports, and multiscale simulation, which helped raise interest in a different future for HPC and cancer. 

The perception was that cancer was too complex: too many factors to realistically bring together into something you could compute or model. That is where AI and deep learning really helped bridge the gap. I still remember the early discussions with the DOE about using deep learning as the central part of the NCI-DOE collaboration, and the development of CANDLE, the core deep learning software built at the DOE and used across the NCI-DOE pilot projects. With enough observational data, one could train a model to give us at least some guidance or insight into what might be happening. The accessibility of AI technologies has really democratized HPC for cancer. Early on, the potential large-scale uses of AI were difficult to envision beyond image processing. Now it is ubiquitous. 

When we started the workshop, I was thinking, “If we can get three years for that workshop, that’ll be success.” It was about year four or five (enduring through the pandemic even), it became clear this was actually going to work. We were seeing more people coming in, AI technologies were becoming broadly available, and the Cancer Moonshot put real money into cancer researchers exploring new approaches. Of late, ChatGPT has become another inflection point. You can look at publication counts to see when these technologies emerged and observe the changes in how the community increasingly engaged with computing.

Computing is a way to bring people and pieces together. One can work in isolation on different pieces, and computing enables one to bring the pieces together, clarify interactions, and explore the combined modeling needed to understand how complex systems actually behave. People begin to realize that computing is not just a tool for one field. It’s a means of integration across all. And that’s what makes it so powerful in cancer, a problem that requires many disciplines working together at once to advance.

SC26: In December 2024, you moved from a federal laboratory into UT MD Anderson, inside the hospital for the first time. What is different about working where the patients are?

Stahlberg: Much of my motivation was to shift from almost exclusively the cancer research domain, where the past decade has produced a large number of promising computational approaches to improve patient outcomes. In making the shift to UT MD Anderson, particularly with the Institute for Data Science in Oncology, there is a focus on translation for use in cancer care.  Reaching patient impact has been the driving aim for me from the beginning: to shorten and accelerate the path from an important cancer research insight to reach actual patient impact.

Inside the hospital, there is a clear focus on patient care and safety. Within the Institute for Data Science and Oncology, a central question is how to translate new data science capabilities into better patient experiences, advanced research, improved operations, and approaches that support clinical decision-making to improve cancer care. There is a level of care and stewardship that is so important to appreciate when introducing new approaches. Another new dimension is simply direct access to clinicians, having conversations about the problems they actually have and how computing and data science might address them. That opens a very different envelope of opportunities than what I had appreciated in the research-focused environment.

---

CAFCW participants at SC23, left to right: Eric Stahlberg, Sally Ellingson, Lynn Borkon, Patricia Kovatch, James Lillard, Orly Alter, Oleksandr Narykov, and Kevin Leung. Stahlberg, Ellingson, Borkon, Kovatch, and Lillard served as CAFCW co-organizers; Alter, Narykov, and Leung were presenters.

SC26: Tell us about what the UT MD Anderson Institute for Data Science and Oncology is working to build, and what the path actually looks like from a research capability to clinical use. 

Stahlberg: The Institute for Data Science in Oncology was created at UT MD Anderson by my colleagues, Drs. David Jaffray and Caroline Chung, who recognized early that data science would have a significant impact across cancer, from clinical care to operations to research. Their vision established IDSO with the aim of integrating the power of team data science with UT MD Anderson’s scientific and clinical expertise to transform research and patient care. It is not just about technology, but also about people, culture, education, data science teams, and the caring approach for which UT MD Anderson is known. 

One function of IDSO is to serve as a hub for data science in cancer: fostering communities, driving innovation through projects, and building collaborations inside and outside UT MD Anderson to advance data science applications for clinical and real-world use. Focus areas in medical imaging, cell and genomic analysis, computational modeling, decision analytics, and quality, safety, and access help to bring communities together around opportunities for impact using team data science.

There is no one-size-fits-all path from research to clinical use. A common path involves vetting the new capability for robustness, demonstrating its viability using broader retrospective data, identifying the path to adoption through an observational trial with clinicians, and, with those stages complete and regulatory requirements met, moving into a clinical trial before broad deployment. Safety is paramount at every stage. 

SC26:  You have been bringing students to Supercomputing for years. What do you tell early-career scientists about the value of being in that room?

Stahlberg: I would use the term ‘students’ loosely, recognizing that my efforts have been to bring individuals and new communities into supercomputing to appreciate what is truly possible with today’s computing capabilities. What one sees on the trade show floor is that computational science is not an isolated field or an isolated group; it is a global effort. As I would tell those visiting SC for the first time, it’s an opportunity to see the computational science world on steroids, with potential collaborators from across the globe, the diversity of applications, and examples of technology that far exceed what’s typically found in the classroom or workplace.  

What I tell early-career scientists about the importance of being in the room is this: be open, open to new ideas, new ways of thinking, and to building new connections that may help down the road. Second, I share the idea of never stopping learning, dreaming, and challenging yourself to do more and leave your comfort zone. Lastly, I simply ask that early-career scientists never lose sight of the importance of creating opportunities to help others. I have found over my career that working to open doors for others pays a tremendous return in doors that are opened for you.

SC26:  You have described teaching as one of your deepest sources of fulfillment. What do you want early-career scientists at SC26 to understand about working at the intersection of computing and medicine?

Stahlberg: Teaching is indeed very fulfilling, helping and enabling others to grow and succeed. I also believe that, as one works to teach others, you learn even more yourself. In working at the intersection of computing and medicine, it’s important to understand that HPC actually helps you unite a network, not just the technical network, but a social, personal network, because you have that HPC common ground that you can then cross-apply. The things I’ve been able to help influence in bringing HPC to cancer have come because I’ve had the benefit of a wonderful network of colleagues developed over many years.  

And that leads to what I would advise anyone early in their career: don’t be afraid to cross disciplines because the insights you have in your field may be entirely new perspectives to someone in a different field. The computational perspective you bring can open new opportunities to advance the frontiers of medicine. Never stop learning, and be willing to learn from others as you work on challenging problems together.

As a computational scientist, the following has helped guide me through the last many years. If you can observe something, you can model it. It’s then a matter of how good the model is. As  George Box is commonly quoted,  “All models are wrong, but some models are useful.” There is more to the message that I have taken to heart. Take the imperfect models we have, identify where they both have value and have limits, and work together to improve them and move the field forward. That’s the mindset I’d leave with anyone starting out at this intersection.

Attend SC26 Workshops

HPC Unites in Chicago

Registration Opens 8 July 2026

Full information regarding registration categories and fees are added in the weeks leading up to the opening of registration. Join us as the HPC community unites in Chicago this fall.

SC26 Registration