Data Science Core

Seung, Hyunjune

Abstract

Core 2, Data Science Working memory, the ability to temporarily hold multiple pieces of information in mind for manipulation, is central to virtually all cognitive abilities. This multi-component research project aims to comprehensively dissect the neural circuit mechanisms of this ability across multiple brain areas. In doing so, it will generate an extremely large quantity of data, from multiple types of experiments, which will then need to be integrated together. The Data Science Core will support the individual research projects in discovering relationships among behavior, neural activity, and neural connectivity. The Core will create a standardized computational pipeline and human workflow for preprocessing of calcium-imaging data. The pipeline will run either on local computers or in cloud computing services, and users will interact with it through a web browser. The preprocessing will incorporate existing image-processing algorithms, such as Constrained Nonnegative Matrix Factorization and convolutional networks. In addition, the Core will build a data science platform that stores behavior, neural activity, and neural connectivity in a relational database that is queried by the DataJoint language. Diverse analysis tools will be integrated into DataJoint, enabling the robust maintenance of data-processing chains. This data-science platform will facilitate collaborative analysis of datasets by multiple researchers within the project, and make the analyses reproducible and extensible by other researchers. We will develop effective methods for training and otherwise disseminating our computational tools and work flows. Finally, the Core will make raw data, derived data, and analyses available to the public upon publication via the data-science platform, source-code repositories, and web-based visualization tools. To facilitate the conduct of this research, the creation of software tools, and the reuse of the data by others after the primary research has concluded, the project will adopt shared data and metadata formats using the HDF5 implementation of the Neurodata without Borders format. Data will be made public in accord with the FAIR guiding principles ? findable by a DOI and/or URL, accessible through a RESTful web API, and interoperable and reusable due to DataJoint and the Neurodata Without Borders format for data and metadata. These tools will allow the researchers within the project to store, manipulate, and analyze their data efficiently and to share it with other researchers as needed.