Breakthrough cancer modeling tool more than a decade in the making

By Pete DiPrimio, IU Luddy School of Informatics, Computing and Engineering

 Paul Macklin, professor of Intelligent Systems Engineering and associate dean for Undergraduate Education at the Luddy School of Informatics, Computing, and Engineering, has co-authored a paper in the prestigious journal Cell that oculd lead to potential medical breakthroughs in cancer treatment – particularly cancers of the pancreas and breast.

The article, titled Human interpretable grammar encodes multicellular systems biology models to democratize virtual cell laboratories, describes how the modeling tool PhysiCell, an open source 3-D agent-based simulator able to simulate millions of cells, can turn biological knowledge and data into computer simulations without writing code, thus enabling researchers to quickly begin testing ideas and, hopefully, develop crucial treatments.

Macklin and researchers from Indiana University, John Hopkins University, the University of Maryland and Oregon Health and Science University developed a new way to connect biological expertise, mathematic theory and data in a 10-year effort. It started small in Macklin’s lab and grew to the coalition in the paper. Their work makes advanced biological simulation modeling more intuitive, more collaborative and more accessible throughout a range of major health challenges.

“It takes a long-term commitment to build things and not abandon them,” Macklin said. “Sometimes, it doesn’t look like it will pay off. We made a commitment when I arrived at IU to invest in the engineering and provide long-term support to grow from a fragile single lab code to a robust community resource.

“We kept improving it. We were always doing new science, but we never neglected the tools. We invested in training materials, the software, the community and the usability. That long-term investment paid off. Most software isn’t adaptable enough to try new things like this.”

Macklin called that perseverance “the special sauce.”

“We didn’t think of software maintenance and training people as distractions, but something that enabled our research. Now we’re able to make a leap in the field no one was able to make before.”

Macklin and colleagues Elana Fertig, professor of medicine and epidemiology and director of the Institute for Genomic Sciences at the University of Maryland School of Medicine, Genvieve Stein O’Brien, Bloomberg assistant professor of neuroscience at John Hopkins University, Laura Heiser, associate professor of biomedical engineering at Oregon Health and Science University (OHSU), Lisa Coussens, chairwoman of the OHSU department of Cell, Development and Cancer Biology, and Joe Gray, professor emeritus at OHSU, led a group of 50 researchers that also included IU postdoctoral researcher Heber Rocha (“he’s done incredible work in helping to make this happen,” Macklin said), postdoctoral researcher Jeanette Johnson and assistant professor Daniel Bergman at the University of Maryland.

That team showcased Macklin’s collaboration emphasis.

“It’s always been important to find good people to work with, who believe in working as a team, and who build each other up and support one another,” Macklin said. “We have good partners and good people. We keep adding people who bring things we’re missing. That helps us grow as a team.”

That’s epitomized by Macklin’s work with Fertig, who he called a “wonderful collaborator and a trusted friend. We’ve worked well together. We bring different pieces to the puzzle and completement each other scientifically.”

Macklin is passionate about his research – what he’s done, who he’s done it with, where it could lead. The cornerstones, he said, are teamwork, commitment and dedication. He called the paper an “amazing collaboration.”

Macklin developed the PhysiCell software along with the mathematics and the new model language (the cell behavior hypothesis grammar) in the paper. He said as he grew PhysicCell from a single-lab project to a community, several collaborations came together as a coalition for the paper.

Macklin and Fertig coordinated the project and developed the central techniques, with key expertise and experiments from across the team.

“That is the theme of the paper. Everyone else has contributed amazing pieces that made this a far stronger paper than any of us could have done on our own.”

It began more than a decade ago, when early modeling required hand coding everything, which was time-consuming.

“It got to be so complex and brutal that it was very difficult to move forward,” Macklin said, “and it was hard to reuse prior work.”

He added that meant taking a step back to focus on the process of modeling, and realizing the technical complexity and collaboration were the big bottlenecks to progress.

Macklin and colleagues prototyped the modeling language in the paper and connected with a network of immunologists, cancer biologists and bio-informatics specialists to test and refine it.

As the modeling software improved, more people started using it and, “We wanted to build more interesting models and support more people,” Macklin said. “It took a lot of design and constant refinement to make it easier for people to build impactful models that worked well.”

Finally, after years of effort, it all came together.

“Now, we can build models much faster than we ever did before,” Macklin said. “We can reach a point where we can predict for individual patients what we can do for them.

“We had this software and modeling language and the mathematics worked out. Elana, Laura, Genevieve, Joe, Lisa and the others identified wonderful problems to try it on. “

They took complicated problems in cancer immunology to analyze how immune cells and cancer cells interact. They built models to understand how chemotherapies and immunotherapies might work together on specific patients.

“It’s really a beautiful thing that we all came together and made a much more meaningful story than if we’d been working alone,” Macklin said.

Macklin said their research also opens up “exciting possibilities in virtual clinical trials, patient forecasting, and even digital twins.”

“Part of our ongoing work is to train a new generation of scientists to use these tools. Creating cloud/web-hosted modeling allows anybody to create and explore models without complicated downloads, and even fluidly connecting with supercomputers to run their studies.”

Take pancreatic cancer, which has a high mortality rate in part because it’s not detected early enough. Macklin said that has to be attacked from clinical and theoretical perspectives. That meant building an accurate model of pancreatic cancer to see its growth dynamics and discover what were the best ways in interrupt those dynamics to help a patient. It meant creating a virtual lab to accommodate that.

Now, Macklin said he can meet with a biologist, and in real time create and explore a model that answers compelling questions.

“How do you turn clinical knowledge or biological knowledge into mathematics and simulation code? In the past that was a laborious process that took months or even years. Now, it can be done in real time in seconds or minutes.”

Macklin said a biologist can tell him that a drug or a signaling factor changes behavior in a specific cell type. They can immediately write that into a simulator, test that idea together and “make progress in a real way that’s never been done before. That’s gratifying.”

In the paper, researchers built models to show how oxygen deprivation affects pancreatic and breast cancer, the immune system’s response to cancer and the effectiveness of combination therapies. Models were also used in a simulated clinical trial for pancreatic cancer and a study in how the brain develops.

“It introduces a new cell behavior grammar (a modeling language, and a way to intuitively capture biological knowledge) that lets biologists use their data and expertise to directly create simulation models without writing computer code,” Macklin said. “They can jump straight into exploring their hypotheses with simulations and better understand and plan their experiments. It also provides a new way to connect the emerging field of spatial biology (especially spatial transcriptomics) with mathematical modeling.”

Besides enabling researchers to quickly turn ideas into simulations, the model lets researchers who aren’t trained in coding directly participate in model building, thus eliminating barriers between disciplines. By simulating outcomes first, scientists can refine their strategies and save time and money, critical in these budget-conscious times.

The modeling tool could be used by hospitals, pharma companies and global health organizations to improve therapies and treatment plans.

It can also boost training and education by allowing undergraduates and medical students to explore biology hands-on with fast virtual experiments without having to code from scratch.

Funding came from the Jayne Koskinas Ted Giovanis Foundation for Health and Policy, and the National Institutes of Health.

“People had talked about making models easier to build and share for years,” Macklin said. “We decided, we’ve talked long enough. Let’s do something.

“It’s neat to see where you have a slow burn, then finally everything is in place, the conditions are right and you have ignition. Suddenly, you have something you didn’t have before.

“It’s not perfect, but it’s a solid advance in that direction. We’re already thinking of things we’re going to do next. There’s more work to be done. We’re excited about it.”