Open Collections

Reproducible Virtual Tissue SimulationsâOpportunities and Challenges

Banff International Research Station for Mathematical Innovation and Discovery

A wide variety of multiscale models address complex biological dynamics, ranging from the organization of the cytoskeleton inside a cell to the organization of cells into tissues in development. However, as these models and their simulations become more complicated, their reproduction (confirmation of a scientific result using independent methods) for the purpose of scientific validation and the reuse of their components become more problematic. An increasing number of software frameworks aim to simplify the simulation of such models (Cytoscape, Physicell, VCell, CompuCell3D, \dots) and these frameworks often make replication (confirmation of a scientific result using the same means) more practical. However, reproducibility remains very difficult because the specification of a model in one framework cannot be executed in another framework. This difficulty occurs because the model specification does not preserve the underlying biological hypotheses and concepts. As a result, a researcher trying to reproduce a result in computational biology first needs to try to extract the underlying biological model, then translate it into a new mathematical and computational instantiation. The effort required at each step is large and error-prone and publications often lack enough information to allow unambiguous disassembly into the biological concepts need for reassembly and reproduction. The Systems Biology Markup Language (SBML) community has made significant progress in defining standards for specifying biochemical network dynamics models based on the underlying biological concepts while also describing their specific mathematical implementation. This approach couples biological descriptors with mathematical implementation in a way that allows the two to be separated. SBMLâs preservation of the underlying conceptual model in the mathematical model and its simulation in turn supports understanding, reproducibility, portability, knowledge extraction, reuse and extension. However, ODE based networks are much simpler than the spatially-defined models of virtual tissues. We consider some of the difficulties in building reproducible virtual tissue models and discuss possible approaches to enhance reproducibility of virtual tissues, including: 1) Standards for the conceptual description of cell and tissue-level objects and processes. 2) Standards for the parametrization of key biological processes (e.g., mean velocity and velocity and angular persistence times for cell motility). 3) Standards for the static specification of spatial structure in tissues for simulation initialization and outputs (like MultiCellDS or VCellâs efforts to specify spatial configuration of cells and components). 4) Approaches to allow the componentization of submodels and their interconnection (model APIs, e.g., to combine an SBML based cell cycle model with a spatial cell model of division). 5) Approaches to allow the interconnection and interoperability of software components (software APIs, e.g., to allow the use of a Physicell diffusion equation solver with a CompuCell3D cell dynamics solver). 6) Standards for the visualization of simulations. 7) Standards for the specification of input, output and tolerance data sets for the validation of models. 8) Standards for the specifications of virtual populations, parameter exploration, sensitivity analysis and optimization. Clearly, we cannot pursue all of these goals at once, but we can perhaps begin to identify communities of the willing to begin work on some and prioritize the development of others.

Mathematics; Biology and other natural sciences; Partial differential equations; Mathematical biology

video/mp4

Author affiliation: Professor

2019-12-18

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/72838

Unreviewed

http://creativecommons.org/licenses/by-nc-nd/4.0/