Principal Investigator: Dr Shoi Shi
University of Tokyo, Tokyo, JapanTags: 48357, Machine Learning, sleep
Sleep deficiency is a common public health problem in many countries, such as the United States, U.K., or Japan. If we know how long people should sleep, we can diagnose sleep deficiency quantitatively. However, we still do not know this. We have already developed a machine learning based algorithm which enable us to detect sleep in high accuracy from the activity data. By applying this algorithm to the 100,00 activity data stored in the U.K. BioBank, we can obtain the largest data set of sleep in the world and the highly accurate distribution of sleep length. We plan to use the first 6 months for adjusting our algorithm to the data obtained by the AX3 sensor which is used in the U.K. BioBank. With the later 6 months, we will analyze all data (100,000 subjects) and prepare for publication.
Project extension – January 2020
Recent genetic studies have also identified several sleep genes whose mutant mice or flies have increased or decreased sleep amount per day. Consistently, in human studies, many genetic associations with sleep phenotype have been identified (Jones et al., 2019), which indicates that sleep amount per day is regulated by genes (Shi et al., 2019). With the development of a high-accurate sleep phenotyping algorithm based on acceleration data, we can read-out relatively accurate sleep phenotype including sleep length, the wake after sleep onset (WASO) and other sleep parameters from the big data stored in the UK Biobank. By comparing these sleep parameters with genetic information, we can find new sleep-related genetic associates. In this study, we would like to perform correlation analyses between the exome sequencing data and GWAS data and sleep parameters calculated by our developed algorithm and find new sleep-related genetic associates.
Last updated Jan 27, 2020