Building the next generation of virtual cells to understand cellular biology

Abstract

Cell science has made significant progress by focusing on understanding individual cellular processes through reductionist approaches. However, the sheer volume of knowledge collected presents challenges in integrating this information across different scales of space and time to comprehend cellular behaviors, as well as making the data and methods more accessible for the community to tackle complex biological questions. This perspective proposes the creation of next-generation virtual cells, which are dynamic 3D models that integrate information from diverse sources, including simulations, biophysical models, image-based models, and evidence-based knowledge graphs. These virtual cells would provide statistically accurate and holistic views of real cells, bridging the gap between theoretical concepts and experimental data, and facilitating productive new collaborations among researchers across related fields.

Keywords: 3D models; Biophysical models; Cell science; Community collaboration; Community modeling; Integrating information; Knowledge graphs; Multiscale modeling; Reproducibility; Simulations; Spatial models; Virtual cells.

Figure 1

Providing data, models, simulation software, and other connected knowledge in online interfaces can make it easy for researchers and students to find, access, use, and extend virtual cells or their contributing components. Example online interfaces already exist: (A) the Cell Feature Explorer enables the interactive exploration of hundreds of thousands of cells at a time on cfe.allencell.org; (B) the Simularium Viewer allows modelers to host their simulations online to provide interactive access to their models with a single URL that plays through the simularium.allencell.org user interface.

Figure 2

Multiple types of useful models published over decades of research are exemplified with the well-known (although still evolving) function and mechanism of the motor protein kinesin. (A) Phenomenological models enable the discovery and characterization of cell structures and processes. Many are generated manually with observation and easy measurements made on raw data such as time series microscope movies, e.g., kymograph models (redrawn after (42)), which are graphical representations of position over time and predictive simplifying equation models. Structural models provide atomic-level detail. (B) Structural models facilitate conceptual and mechanistic hypotheses that combine phenomenological understandings with other types of data such as protein-protein interactomes or kinetic reaction rates. These often begin as mental models, but must be converted to evidence-based conceptual models such as illustrated or animated model figures to enable discourse (43). (C) Simulated models rigorously validate these hypotheses or test and explain detailed mechanisms and unexpected variations that are often too complex for a human to model in their mind (44,45).