Neuroimaging research is increasing in volume and scope, needing "big data" methods for discovery. A number of important resources already exist for neuroimaging, including data repositories, crowd-sourcing knowledge bases, standardized ontologies and terminologies, and meta-analytic repositories. However, while data and code-sharing efforts are growing, there is little interaction and limited sharing of knowledge across platforms. Machine learning methods are being applied in other fields to extract knowledge locked in text (e.g., journal articles, patient records, social media posts), with the goal of recognizing relations among entities (e.g., "drug X causes adverse event Y"). Cognitive neuroscientists also determine relations, specifically between brain regions and cognitive, perceptual, and motor processes (e.g., "mental function X activates brain network Y"), but are hampered in using high-throughput automated methods on the ever-growing published text. It is not a simple process to identify which cognitive processes were studied in a given project, or what brain networks were identified as related to which mental process. The investigators propose an integrative metadata framework that describes the experimental design characteristics and results, as well as the knowledge that the research provides. Efficient knowledge sharing may best be achieved via an interactive data ecosystem that uses standards for transparency and openness when describing knowledge derived from cognitive neuroimaging experiments. Developing this integrative metadata framework for neuroimaging will increase the community's ability to share data and evaluate reliability in the resulting relationships between mind and brain. This project aims to provide improvements in large-scale integration of the scientific literature, with more rapid understanding of the complexity of brain research and neurocognitive models, within an educational setting for training STEM students and accelerated research productivity.Neuroscientific research frequently requires efficiency, transdisciplinary collaborations, and cross-domain flexibility. Efficient knowledge sharing may best be achieved via an interactive data ecosystem that relies on an integrative metadata framework. Such a framework would address scientific reproducibility by providing standards for transparency and openness when describing knowledge derived from cognitive neuroimaging experiments. Moreover, development of an integrative metadata framework for cognitive neuroimaging will enhance interaction between existing neuroinformatics resources, increasing the community's ability to share data and evaluate reliability in experimental findings. This project will develop knowledge modeling tools for cognitive neuroimaging studies, as well as large-scale meta-analytic evaluations of cognitive models. The investigators will build on previous work extracting experimental design features from the text to create an ensemble of classifiers for full text papers. The investigators will work with an External Advisory Board for evaluation and feedback, and will use the framework to automatically extract knowledge regarding mind/brain models within the exemplar domains of executive function, affective processing, and reward feedback. The investigators will integrate classifiers and methods with other international standards for data and results sharing (e.g., NI-DM, CEDAR and ISA-TAB, BioCaddie) and other repositories (e.g. Neurosynth, BrainSpell) for broader use in the community. The intellectual merit of this project is the enhanced access to cognitive neuroscience knowledge that is currently locked in text. This project's success will allow the research community to collectively address hurdles such as annotating their own data and sharing their data/results via integrated annotations in a public repository, journal, or knowledge discovery platforms, and ultimately lead to long-term strategies for cross-domain neurocognitive model development. This project has been designed to have high integrative value and will interact, harmonize, and share data and algorithms with existing neuroinformatics resources that will benefit from enhanced knowledge modeling techniques. Moreover, in an effort to promote transparency, reliability, and reproducibility, this project will be publicly available on the Open Science Framework and Github (e.g., Labels, Classifiers, Code). Such an integrative metadata framework may be viewed as the connective tissue that will facilitate a new generation of cognitive model development, providing a potentially transformative strategy for modeling the literature, and ultimately leading to more informed, evidence-based, and reproducible neurocognitive models of brain function.
