Rationale

The cell is difficult to model because it consists of many different components that self-organize spatiotemporally. Computational modeling in general can be understood as a search for models that are sufficiently consistent with input information about the modeled system¹. Given the complexity of the cell, computing an accurate, precise, and complete cell model requires a vast amount of information. This information can in principle be provided by different types of experimental datasets (eg, a cryo-electron tomography map of the cell and a large-scale affinity co-purification dataset), physical principles (eg, statistical mechanics of protein structure and dynamics), statistical preferences (eg, a statistical potential extracted from known protein structures), and prior models of the cell and/or its parts (eg, protein structure models and molecular signaling networks)². Moreover, these experimental datasets and models must by necessity be produced by many scientists with diverse expertise, working iteratively with one another over long periods of time. Such long-term, collaborative, and interdisciplinary work requires experimental datasets and models to be archived, shared among the collaborators, analyzed, used for further modeling, and disseminated to the broader community. In other words, cell modeling is data-centric: advances are driven by the collaborative creation, validation, curation, and exchange of complex data³. As a result, a cyberinfrastructure for comprehensive data management⁴ is critical for a successful cell mapping project. This cyberinfrastructure needs to be efficient, robust, and extensible, including by linking with existing external resources into a federated system⁵, wherever possible. In addition, it must be freely available, so as to encourage its adoption, further development, and use by as many scientists as possible, in turn making it more useful to all.

Goal

Our goal is to develop CellLab, a comprehensive data-centric cyberinfrastructure that facilitates mapping the architecture and function of the cell. Most importantly, CellLab supports a large team of collaborators to store, analyze, use, share, and disseminate many varied data, including experimental datasets and cell models, in multiple iterations of experiment and modeling that are required for cell mapping. CellLab promotes collaboration and reproducibility by integrating data management with various modeling engines and modeling workflows. Thus, CellLab creates a data-centric, socio-technical ecosystem for collaborative science by providing a framework for sharing expertise, experimental data, models, and other resources.

Data-centric approach to cell mapping

CellLab is built using our scientific asset management platform Deriva⁶. Data management abides by the FAIR (Findable, Accessible, Interoperable, and Reusable) principles⁷. To drive the development of CellLab, we use it for modeling glucose-stimulated insulin production and secretion⁸, one of the key functions of the β-cell, in collaboration with the Pancreatic β-Cell Consortium (PBCC). This use case challenges all aspects of CellLab, because insulin production and secretion encompass many of the complexities of the whole cell, including aspects best described using different types of experimental data and models at different scales. Thus, our existing open-source automated Bayesian metamodeling⁹ is the first instantiation of a modeling method in CellLab; metamodeling updates a set of heterogeneous input models, such as spatiotemporal and network models, to make them more consistent with each other. Iterative application of CellLab to the use case results in rapid improvement of CellLab's user-centered design. Additional use cases are under development, including one from the Cancer Cell Map Initiative (CCMI).