Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 Mar 8;10(10):eadj6834.
doi: 10.1126/sciadv.adj6834. Epub 2024 Mar 8.

Accurate detection of acute sleep deprivation using a metabolomic biomarker-A machine learning approach

Katherine Jeppe  1   2 , Suzanne Ftouni  1   2 , Brunda Nijagal  3 , Leilah K Grant  1   2   4   5 , Steven W Lockley  1   2   4   5 , Shantha M W Rajaratnam  1   2   4   5 , Andrew J K Phillips  1 , Malcolm J McConville  3 , Dedreia Tull  3 , Clare Anderson  1   2   6
Affiliations

Accurate detection of acute sleep deprivation using a metabolomic biomarker-A machine learning approach

Katherine Jeppe et al. Sci Adv. .

Abstract

Sleep deprivation enhances risk for serious injury and fatality on the roads and in workplaces. To facilitate future management of these risks through advanced detection, we developed and validated a metabolomic biomarker of sleep deprivation in healthy, young participants, across three experiments. Bi-hourly plasma samples from 2 × 40-hour extended wake protocols (for train/test models) and 1 × 40-hour protocol with an 8-hour overnight sleep interval were analyzed by untargeted liquid chromatography-mass spectrometry. Using a knowledge-based machine learning approach, five consistently important variables were used to build predictive models. Sleep deprivation (24 to 38 hours awake) was predicted accurately in classification models [versus well-rested (0 to 16 hours)] (accuracy = 94.7%/AUC 99.2%, 79.3%/AUC 89.1%) and to a lesser extent in regression (R2 = 86.1 and 47.8%) models for within- and between-participant models, respectively. Metabolites were identified for replicability/future deployment. This approach for detecting acute sleep deprivation offers potential to reduce accidents through "fitness for duty" or "post-accident analysis" assessments.

PubMed Disclaimer

Figures

Fig. 1.
Fig. 1.. Features isolated by HILIC LC-MS and showing linear and/or cyclical trends across each sleep deprivation experiment.
Heatmaps display significant (FDR-adjusted P value of <0.05) group-level trends during sleep deprivation for linear (A) and cycling (B) trends for experiment 1 (n = 12) and linear (D) and cycling (E) trends for experiment 2 (n = 11). For all heatmaps, purple corresponds to the highest and green corresponds to the lowest values for z-scored median-normalized peak area. Venn diagrams display the number and overlap of cycling, increasing or decreasing features for experiment 1 (C) and experiment 2 (F). The Venn ring sizes correspond to the number of features, and percentages are relative to the total number of features analyzed (929).
Fig. 2.
Fig. 2.. Significant linear changes at the individual level for each of the five final biomarker candidates across both sleep deprivation experiments.
(A) Horizontal bar charts summarize the percent of participants displaying an increasing or decreasing trend in z-scored median-normalized peak area for each metabolite for sleep deprivation training set (experiment 1) and testing set (experiment 2). (B) Heatmaps display significant linear trends observed for each participant with increasing TSW for both experiment 1 (participants A to L) and experiment 2 (participants M to W). Linear trends are displayed as percent change per hour TSW, with decreasing trends in green and increasing trends in purple. Non-significant metabolite/participants are displayed in dark gray. Metabolites that were also significantly rhythmic are indicated with a white “R.”
Fig. 3.
Fig. 3.. Classification random forest model results for within- and between-participant analyses.
Top: Relative importance of the five final biomarker candidates for training models using within-participant (A) and between-participant (C) data. Testing model accuracies (lower to upper 95% CI) are displayed in each panel. Bottom: Receiver operating characteristic curves for within-participant (B) and for between-participant data (D). Training models are displayed in light green, and test models are in dark green. Acc, accuracy; AUC, area under the curve; MDA%, mean decreased accuracy on variable removal; V4S, vanillin 4-sulfate; I3PA, indole-3-propionate; Sac, monosaccharide 1; PI, PI(16:0/18:1); LPC, LPC(18:3).
Fig. 4.
Fig. 4.. Random forest model regression results for within- and between-participant analyses.
Top: Relative importance of the five final biomarker candidates for training models using within-participant (A) and between-participant data (C). Testing R2 is shown in each panel. Bottom: Model predicted TSW versus actual TSW in test data for within-participant (B) and between-participant models (D). Line types indicate percentiles with corresponding prediction error shown in hours (e.g., 50% of individual points are predicted within 2.1 or 5.1 hours of the actual time awake for within- and between-participant, respectively). Two outlier participants (P = □; S = ▽) are highlighted (adjusted R2 for between-participant is 63.8% with their removal). MSE%, increase in mean square error on variable removal; TSW, time since wake.
Fig. 5.
Fig. 5.. Group-level trends of the five final biomarker candidates in sleep deprivation and matched control experiments.
(A and B) Mean candidate metabolite levels (z-score) from a 4-hour block pre- and post-habitual sleep interval were compared between sleep deprivation and matched control protocols. Pre- and post-habitual sleep were compared using both clock time–matched (A) (sleep deprivation—pre: 2 to 6 hours TSW/post: 26 to 30 hours TSW; matched control pre: 2 to 6 TSW day 1/post: 2 to 6 TSW day 2) and (B) evening/morning (sleep deprivation—pre: 12 to 16 hours TSW/post: 26 to 30 hours TSW; matched control pre: 12 to 16 hours TSW day 1/post: 2 to 6 hours TSW day 2). (C) A comparison of a WR day 3 from sleep deprivation (CR) and matched control (CP) protocols. Dashed lines indicate main meals (dark gray) and snacks (light gray) in the matched control protocol. Snacks were provided hourly in the sleep deprivation protocols. Series colors indicate the protocol [sleep deprivation (experiment 1): light green, sleep deprivation (experiment 2): dark green, and matched control: charcoal].
Fig. 6.
Fig. 6.. Raster plots depicting the in-laboratory protocols used in sleep deprivation experiments 1 and 2 and the matched control study.
(A) The 6-day protocol used in sleep deprivation experiments. Days 1 and 2 were baseline days with 8-hour:16-hour sleep/wake cycle based on average habitual sleep before admission (AD). On day 3, the 40-hour CR commenced ending at the end of day 4. Days 5 and 6 were recovery days with up to 12-hour sleep opportunities before discharge (DC) on day 6. (B) The 3-day in-laboratory protocol used in the matched control study. Day 1 was the baseline day, and each day followed a self-selected 8-hour:16-hour sleep/wake cycle. Black diamonds represent blood samples taken for metabolomics analysis. White bars represent wake intervals in 100 lux, black bars represent sleep intervals in 0 lux, and gray bars represent wake intervals in <3 lux ambient light. The protocol is shown in relative clock time with a relative bedtime of midnight. Study events were scheduled relative to each individual’s pre-study self-selected wake time.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Uehli K., Mehta A. J., Miedinger D., Hug K., Schindler C., Holsboer-Trachsler E., Leuppi J. D., Kunzli N., Sleep problems and work injuries: A systematic review and meta-analysis. Sleep Med. Rev. 18, 61–73 (2014). - PubMed
    1. Bioulac S., Micoulaud-Franchi J. A., Arnaud M., Sagaspe P., Moore N., Salvo F., Philip P., Risk of motor vehicle accidents related to sleepiness at the wheel: A systematic review and meta-analysis. Sleep 40, zsx134 (2017). - PubMed
    1. J. L. Lim, D. F. Dinges, in Molecular and Biophysical Mechanisms of Arousal, Alertness, and Attention, D. W. Pfaff, B. L. Kieffer, Eds. (Blackwell Publishing, 2008), vol. 1129, pp. 305–322. - PubMed
    1. Anderson C., Horne J. A., Sleepiness enhances distraction during a monotonous task. Sleep 29, 573–576 (2006). - PubMed
    1. Collet J., Ftouni S., Clough M., Cain S. W., Fielding J., Anderson C., Differential impact of sleep deprivation and circadian timing on reflexive versus inhibitory control of attention. Sci. Rep. 10, 7270 (2020). - PMC - PubMed