0321 Luna: Merging Open-source Tools, Data and Models for Sleep Signal Analyses

Abstract Introduction Data-sharing is increasingly recognized as a critical driver of discovery and reproducibility in modern biomedical science. To realize the promise of data-sharing, important corollaries are the need for open tools, standards and models to work with and make sense of those data. Here we outline a new software framework (Luna), which is tightly coupled to data in the National Sleep Research Resource (NSRR) and supports predictive models based on these and other datasets. Methods Luna is well-documented, multi-platform open-source software package that supports a variety of analyses of sleep macro- and micro-architecture and is being integrated into the NHLBI BioData Catalyst ecosystem. Web-based Moonlight & Moonbeam components support the direct access, interactive viewing and analysis of NSRR data. We are also developing modular and extensible model-based prediction capabilities in Luna, to predict outcomes such as biological age directly from raw sleep (EEG) signals. Results We will present specific results from three exemplar applications of Luna: 1) a sleep stager and biological/brain age prediction model tailored to pediatric contexts, trained on over 5000 individuals, 2) risk models for schizophrenia from high-density sleep EEG data, and 3) sleep-based biomarkers of cognitive aging in older adults. As well demonstrating the performance of these models, we will discuss challenges to validity and interpretation more generally. Conclusion The power of open sleep data is greatly augmented by robust tools, standards and models to facilitate working with those data. Greater community-wide sharing and transparency of tools and methods is an important step to greater reproducibility and rigor in sleep science. Support (if any) NIH/NIA 1R01AG070867; NIH/NHLBI 5R01HL146339; NIH/NHLBI 75N92019C00011.