% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@ARTICLE{Raimondo:1052071,
author = {Raimondo, Federico and Bi, Hanwen and Komeyer, Vera and
Kasper, Jan and Primus, Sabrina and Hoffstaedter, Felix and
Mandal, Synchon and Waite, Laura and Winkelmann, Juliane and
Oexle, Konrad and Eickhoff, Simon and Tahmasian, Masoud and
Patil, Kaustubh},
title = {{C}an we predict sleep health based on brain features? {A}
large-scale machine learning study using {UK} {B}iobank},
journal = {Brain communications},
volume = {8},
number = {1},
issn = {2632-1297},
address = {[Oxford]},
publisher = {Oxford University Press},
reportid = {FZJ-2026-00741},
pages = {fcag016},
year = {2026},
abstract = {Numerous correlational and group comparison studies have
demonstrated robust associations between sleep health and
large-scale brain organization. However, individual
differences play a critical role in this relationship,
highlighting the need for person-specific analyses. In this
study, we aimed to explore whether brain imaging features
could predict various sleep health-related traits at the
individual level using machine learning (ML) techniques.We
utilized data from 28,088 participants in the UK Biobank,
extracting 4,677 structural and functional neuroimaging
markers. These features were then used to predict a range of
self-reported sleep characteristics, including insomnia
symptoms, sleep duration, ease of waking in the morning,
chronotype, napping behavior, daytime sleepiness, and
snoring. For each of these seven traits, we trained both
linear and nonlinear ML models to evaluate how well brain
imaging data could account for individual differences.Our
analyses involved extensive computational resources,
equivalent to over 200.000 core-hours (equivalent to 25
years of compute time). Despite this, the predictive
performance of brain features was consistently low across
all models, with balanced accuracy scores ranging from 0.50
to 0.59. The highest accuracy achieved (0.59) came from a
linear model predicting the ease of getting up in the
morning. Notably, models using only demographic variables
such as age and sex achieved comparable performance,
suggesting that these basic characteristics may largely
explain the observed variability.These findings indicate
that, even when using a large, well-powered sample and
advanced ML techniques, multimodal brain imaging features
provide limited predictive value for sleep health at the
individual level. This low predictability underscores the
complexity of the relationship between sleep and brain
structure/function. It also suggests that other biological,
environmental, or psychological factors—possibly not
captured by current imaging modalities—may play a more
substantial role in shaping sleep-related behaviors.},
cin = {INM-7},
ddc = {610},
cid = {I:(DE-Juel1)INM-7-20090406},
pnm = {5252 - Brain Dysfunction and Plasticity (POF4-525)},
pid = {G:(DE-HGF)POF4-5252},
typ = {PUB:(DE-HGF)16},
doi = {10.1093/braincomms/fcag016},
url = {https://juser.fz-juelich.de/record/1052071},
}