Drs. Jeanette Johnson, Elana J. Fertig, and Daniel Bergman review mathematical models and genomic data to simulate cancer cell growth. [University of Maryland School of Medicine]Drs. Jeanette Johnson, Elana J. Fertig, and Daniel Bergman review mathematical models and genomic data to simulate cancer cell growth. [University of Maryland School of Medicine]

Researchers at Indiana University and at University of Maryland School of Medicine (UMSOM) Institute for Genome Sciences (IGS), have developed a method to predict cell activity in tissues over time, similarly to how weather forecast models can predict developing storms.

The newly developed software combines genomics technologies with computational modeling to predict cell changes in behavior, such as communication between cells that could cause cancer cells to flourish.

What makes this research unique is the use of a plain-language “hypothesis grammar” that uses common language as a bridge between biological systems and computational models, and simulates how cells act in tissue. Reporting on their framework in Cell, the researchers outline virtual experiments exploring how cancer responds to the cells in its environment and how the brain forms layers in development. The study is the result of a multi-lab project, stemming from research at the interface of software development involving key collaborations between bench and clinical team science researchers.

The investigators suggest their work could eventually lead to the development of computer programs to aid selection of the best treatment for cancer patients by creating a “digital twin” of the patient. “As much as this new ‘grammar’ enables communication between biology and code, it also enables communication between scientists from different disciplines to leverage this modeling paradigm in their research,” said Daniel Bergman, PhD, a scientist at IGS and Assistant Professor of Pharmacology and Physiology at UMSOM. Bergman is co-lead author of the team’s published paper, which is titled “Human interpretable grammar encodes multicellular systems biology models to democratize virtual cell laboratories.” In their paper the authors concluded that their approach, “… bridges biological, clinical, and systems biology research for mathematical modeling at scale, allowing the community to predict emergent multicellular behavior.”

Cells interact to form ecosystems that evolve as dynamical systems, the authors explained. Recent single-cell and spatial multiomics technologies may quantify individual cell characteristics, but predicting their evolution requires mathematical modeling. “Generating temporally resolved multicellular predictions remains an open computational challenge,” the team further noted. And while bioinformatics techniques and machine learning can predict cellular trajectories and phenotypic changes in individual cell types, the team continued, they can’t account for more complex changes with time across multicellular networks. “More advanced computational tools are needed to fill the gaps between measurement times and leverage biological knowledge and mechanism to forecast unseen emergent behaviors in multicellular systems de novo.”

Jeanette Johnson, PhD, a postdoctoral fellow at the IGS, further commented, “Although standard biomedical research has made immeasurable strides in characterizing cellular ecosystems with genomics technologies, the result is still a single snapshot in time—rather than showing how diseases, like cancer, can arise from communication between the cells. Cancer is controlled or enabled by the immune system, which is highly individualized; this complexity makes it difficult to make predictions from human cancer data to a specific patient.”

Paul Macklin, PhD, Professor of Intelligence Systems Engineering at Indiana University led a team of researchers who developed the grammar to describe cell behavior. This grammar uses natural language statements (cell rules) to create mathematical models, allowing scientists to use simple English language sentences to build digital representations of multicellular biological systems, for example, developing computational models for diseases as complex as cancer. “This enables systematic integration of biological knowledge and multiomics data to generate in silico models, enabling virtual “thought experiments” that test and expand our understanding of multicellular systems and generate new testable hypotheses,” the authors stated. “Previously, custom hand-written code and a high level of technical knowledge were required to implement even basic models. Our hypothesis grammar can encode complex cellular behaviors and responses to signals in single lines of human-readable text.”

Bergman and colleagues at IGS combined this grammar with genomic data from real patient samples to study breast and pancreatic cancer, with technologies such as spatial transcriptomics. In breast cancer, the IGS team modeled an effect where the immune system cannot curtail tumor cell growth and instead promotes invasion and cancer spread. They adapted this computational modeling framework to simulate a real-world immunotherapy clinical trial of pancreatic cancer.

Using genomics data from untreated tissue samples of pancreatic cancer, the model predicted that each virtual “patient” had a different response to the immunotherapy treatment—showcasing the importance of cellular ecosystems for precision oncology. For example, pancreatic cancer is a difficult cancer to treat, in part, because it is often surrounded by a dense structure of non-cancerous cells called fibroblasts. The team used new spatial genomics technology to further demonstrate the ways fibroblasts communicate with tumor cells. The program allowed the scientists to follow the growth and progression of pancreatic tumors to invasion from real patient tissue.

The new grammar is open source so that all scientists can benefit from it. “By making this tool accessible to the scientific community, we are providing a path forward to standardize such models and make them generally accepted,” said Bergman.  To demonstrate this generalizability, Genevieve Stein-O’Brien, PhD, the Terkowitz Family Rising Professor of Neuroscience and Neurology at Johns Hopkins School of Medicine (JHSOM) led researchers in using this approach in a neuroscience example in which the program simulated the creation of layers as the brain develops.

“The new conceptual framing (a grammar) for specifying cell behavior hypotheses introduced in this study can systemize and facilitate our thinking of how cells interact to drive tissue ecosystems,” the authors wrote in summary. “We demonstrated a variety of models extending from carcinogenesis and immune response to tumor growth and demonstrating broader extensions to neurodevelopment.”

Johnson added, “What makes these models so exciting to me as someone who studies immunology is that they can be informed, initialized, and built upon using both laboratory and human genomics data. Immune cells are amazing and follow rules of behavior that can be programmed into one of these models. So, for instance, we can take data and treat it as a snapshot of what the human immune system is doing, and this framework gives us a sandbox to freely investigate our hypotheses of what’s happening there over time without extra costs or risk to patients.”

Co-lead author Elana J. Fertig, PhD, Director of IGS, Associate Director of Quantitative Sciences for the Greenebaum Comprehensive Center, and Professor of Medicine and Epidemiology at UMSOM, added, “Ever since my transitioning from my training in weather prediction at the University of Maryland, College Park into computation, I have believed that we could apply the same principles to work across biological systems to make predictive models in cancer. I am struck by how many rules of biology we don’t yet know. Adapting this approach to genomics technologies gives us a virtual cell laboratory in which we can conduct experiments to test the implications of cellular rules entirely in silico.”

Fertig calls the research “a tapestry of team science” with additional validation of the computational models coming from clinical collaborators at Johns Hopkins University and Oregon Health Sciences University. “With this work from IGS, we have a new framework for biological research since researchers can now create computerized simulations of their bench experiments and clinical trials and even start predicting the effects of therapies on patients,” said Mark T. Gladwin, MD, Vice President for Medical Affairs at the University of Maryland, Baltimore, and the John Z. and Akiko K. Bowers Distinguished Professor and UMSOM Dean. “This has important applications to enable digital twins and virtual clinical trials in cancer and beyond. We look forward to future work extending this computational modeling of cancer to the clinic.”