Knowledge of the importance of input features towards decisions made by machine-learning models is essential to increase our understanding of both the models and the underlying data. Here, we present a new approach to estimating feature importance with neural networks based on the idea of distributing the features of interest among experts in an attentive mixture of experts (AME). AMEs use attentive gating networks trained with a Granger-causal objective to learn to jointly produce accurate predictions as well as estimates of feature importance in a single model. Our experiments show (i) that the feature importance estimates provided by AMEs compare favourably to those provided by state-of-theart methods, (ii) that AMEs are significantly faster at estimating feature importance than existing methods, and (iii) that the associations discovered by AMEs are consistent with those reported by domain experts.

Highlights d Sleep states regulate more than 50% of all metabolite features detected in human breath d Major pathways are up-or downregulated during wake, SWS, and REM sleep d Sleep-regulated pathways show stepwise logic across different sleep stages

Introduction: Sleep insufficiency or decreased quality have been associated with Alzheimer’s disease (AD) already in its preclinical stages. Whether such traits are also present in rodent models of the disease has been poorly addressed, somewhat disabling the preclinical exploration of sleep-based therapeutic interventions for AD. Methods: We investigated age-dependent sleep-wake phenotype of a widely used mouse model of AD, the Tg2576 line. We implanted electroencephalography/ electromyography headpieces into 6 months old (plaque-free, n=10) and 11 months old (moderate plaque-burdened, n=10) Tg2576 and age-matched wild-type (WT, 6 month old n =10, 11 month old n =10) mice and recorded vigilance states for 24 hours. Results: Tg2576 mice exhibited significantly increased wakefulness and decreased non-rapid eye movement sleep over a 24-hour period compared to WT mice at 6, but not at 11 months of age. Concomitantly, power in the delta frequency was decreased in 6-month old Tg2576 mice in comparison to age-matched WT controls, rendering a reduced slow-wave energy phenotype in the young mutants. Lack of genotype-related differences over 24 hours in overall sleep-wake phenotype at 11 months of age appears to be the result of changes in sleep-wake characteristics accompanying the healthy aging of WT mice. Discussion/Conclusion: Therefore, our results indicate that at plaque-free disease stage, diminished sleep quality is present in Tg2576 mice which resembles aged healthy controls, suggesting an early-onset of sleep-wake deterioration in murine AD. Whether such disturbances in the natural patterns of sleep could in turn worsen disease progression warrants further exploration.

Sleep insufficiency or decreased quality have been associated with Alzheimer disease (AD) already in its preclinical stages. Whether such traits are also present in rodent models of the disease has been poorly addressed, somewhat disabling the preclinical exploration of sleep-based therapeutic interventions for AD. We investigated age-dependent sleep-wake phenotype of a widely used mouse model of AD, the Tg2576 line. We implanted electroencephalography/electromyography headpieces into 6 months old (plaque-free, n=10) and 11 months old (moderate plaque-burdened, n=10) Tg2576 and age-matched wild-type (WT) mice and recorded vigilance states for 24 hours. Tg2576 mice exhibited significantly increased wakefulness and decreased non-rapid eye movement sleep over a 24-hour period compared to WT mice at 6, but not at 11 months of age. Concomitantly, delta power appeared decreased in 6-month old Tg2576 mice in comparison to age-matched WT controls, yielding a reduced slow-wave energy phenotype in the young mutants. Lack of genotype-related differences over 24 hours in overall sleep-wake phenotype at 11 months of age appears to be the result of the natural aging of WT mice. Therefore, our results indicate that at plaque-free stages of the disease, diminished sleep quality is present in Tg2576 mice which resembles aged healthy controls, suggesting an early-onset of sleep-wake deterioration in murine AD. Whether such disturbances in the natural patterns of sleep could in turn worsen disease progression warrants further exploration.