Latest / Sleep Science Friday Series / Leveraging the Metabolic Fingerprint of Sleep Deprivation and Sleep Restriction for Forensic Applications: A Machine Learning Study in Oral Fluid Metabolomics

Transcript

Leveraging the Metabolic Fingerprint of Sleep Deprivation and Sleep Restriction for Forensic Applications: A Machine Learning Study in Oral Fluid Metabolomics · Sleep Science Friday Series

• 0:00
  SPEAKER_01: Hello, listeners.
• 0:01
  SPEAKER_01: I am here with Dr.
• 0:03
  SPEAKER_01: Michael Scholes, first author of the publication Leveraging the Metabolic Fingerprint of Sleep Deprivation and Sleep Restriction for Forensic Applications, a machine learning study in oral fluid metabolomics that was published in the Journal of Proteome Research in May 2026.
• 0:24
  SPEAKER_01: Welcome, Michael.
• 0:26
  SPEAKER_00: Hello, and thank you very much for the invitation.
• 0:29
  SPEAKER_01: So, first of all, what motivated you to investigate whether sleep deprivation leaves a detectable metabolic signature?
• 0:38
  SPEAKER_01: And what gap in current approaches were you hoping to address?
• 0:43
  SPEAKER_00: So, um that is a very important thing to clarify first.
• 0:47
  SPEAKER_00: My background is forensics.
• 0:49
  SPEAKER_00: And well there is a common misbelief that forensics only deals with the deceased.
• 0:55
  SPEAKER_00: But actually, in my field of forensic toxicology, most of our clients are well alive.
• 1:01
  SPEAKER_00: And in fact, the majority of the casework that we do has something to do with drugged or drunk drivers.
• 1:09
  SPEAKER_00: But also drowsy driving is a common thing.
• 1:12
  SPEAKER_00: And I believe most people can relate here and remember at least one episode where they possibly thought they may have been too tired to drive.
• 1:22
  SPEAKER_00: Now, from the legal perspective, drowsy driving is penalized under the same provision as any other form of impaired driving.
• 1:31
  SPEAKER_00: So it is also our task to prove this.
• 1:34
  SPEAKER_00: Yet this turns out to be quite difficult since sleep deprivation does not imply any consumption of any exogenous substance that we could possibly detect easily.
• 1:46
  SPEAKER_00: And that was when my supervisor once had the idea of analyzing the metabolome to see whether we find characteristic changes there that could be used for forensic purposes as a proof of sleep deprivation and possibly impairment.
• 2:03
  SPEAKER_01: Your study identifies a metabolic fingerprint of acute sleep deprivation.
• 2:09
  SPEAKER_01: What exactly do you mean by a metabolic fingerprint?
• 2:13
  SPEAKER_01: And what was the most surprising finding when you first saw these results?
• 2:18
  SPEAKER_00: The term fingerprint is also a remainder of our forensic background.
• 2:23
  SPEAKER_00: As you know, every fingerprint is unique, and so is a metabolic fingerprint.
• 2:29
  SPEAKER_00: In this case, it means that the pattern of metabolites has changed in a distinct way under acute sleep deprivation.
• 2:37
  SPEAKER_00: And like a fingerprint at a crime scene can lead to the perpetrator, the metabolic fingerprint can be used as a forensic proof.
• 2:45
  SPEAKER_00: Now to be honest, at first it was pretty surprising to see for me how clear and distinct this fingerprint was under acute sleep deprivation.
• 2:55
  SPEAKER_00: But then again, when you think somewhat more about it, our sleep drive is such a strong force, one that we ultimately cannot evade.
• 3:05
  SPEAKER_00: And also, I like this comparison, the effects of sleep lead to deep unconsciousness for several hours and initiates extensive bodily reactions.
• 3:16
  SPEAKER_00: But there are not so many drugs out there, if any, that could do the same thing.
• 4:18
  SPEAKER_00: At the bottom line, acute sleep loss was much more detrimental to performance in all tests and at the same time led to much stronger changes in the metabolome.
• 4:31
  SPEAKER_00: My hypothesis is that during total sleep deprivation, some essential metabolic processes did not run at all, especially those taking place during the first hours of sleep.
• 4:45
  SPEAKER_00: Interestingly, however, amongst our study participants, it was always a very lively debate which scenario was subjectively the hardest for them.
• 4:55
  SPEAKER_01: Why did you choose oral fluid for this analysis?
• 4:59
  SPEAKER_01: What advantages does it offer over other biological samples?
• 5:04
  SPEAKER_00: Also, here we were very much focusing on the forensic scenario, especially the drowsy driving case.
• 5:11
  SPEAKER_00: For police and law enforcement, but also for prevention use, it would be best to be able to test at a roadside.
• 5:20
  SPEAKER_00: And that means you want some non-invasive techniques.
• 5:24
  SPEAKER_00: Nevertheless, in the original study, we also collected blood samples as well as exhaled breath and even fingers.
• 5:32
  SPEAKER_00: Now those specimens are not covered in the given paper, but logically, also there, the metabolic fingerprint of sleep loss was detectable to a certain extent.
• 5:44
  SPEAKER_01: Your machine learning approach was able to distinguish sleep-deprived from rested participants using a relatively small set of molecular features, without requiring an individual baseline sample.
• 5:57
  SPEAKER_01: How robust was this finding, and what factors could potentially affect the accuracy of such a model in real-world populations?
• 6:07
  SPEAKER_00: Right, so the goal was always finding a robust pattern without the need of a second reference sample from the same individual, say under rested conditions for baseline comparison.
• 6:20
  SPEAKER_00: Because again, here in forensic practice, you simply do not have the chance, alright, to collect a second sample, whereas in the clinic that would be usually not much of a problem.
• 6:33
  SPEAKER_00: And that was just one of the many, let's say, forensic particularities of our approach.
• 6:40
  SPEAKER_00: Also, quite logically, we knew from the beginning that this approach would require then some sort of feature panel.
• 6:48
  SPEAKER_00: You know, our bodies are quite complex and there's not this single substance that regulates sleep-wake behavior alone.
• 6:56
  SPEAKER_00: But then, as it turned out, the metabolic fingerprint we detected was built upon only 10 metabolites, and adding more was not improving the results anymore.
• 7:08
  SPEAKER_00: Technically, you do not even need all 10 metabolites here, you can work with less for the cost of accuracy.
• 7:16
  SPEAKER_00: The big question is now, since our findings are based on a clinical study, how many of these metabolites are robust enough to withstand real-world effects?
• 7:29
  SPEAKER_01: Another intriguing finding was that the metabolic signature vary depending on the time of day.
• 7:36
  SPEAKER_01: What does this tell us about the interaction between sleep deprivation, circadian rhythms, and metabolism?
• 7:44
  SPEAKER_00: Absolutely.
• 7:45
  SPEAKER_00: Time of day has a big influence on our metabolism, regardless of sleep status.
• 7:51
  SPEAKER_00: And indeed, we have seen this influence very clearly in our final models.
• 7:56
  SPEAKER_00: Quite interestingly was the fact that the models would always pick some metabolites that actually did not exhibit any significant differences between sleep-deprived individuals and baseline.
• 8:08
  SPEAKER_00: But as soon as we would restrict the models from picking these metabolites, the scores decreased a lot.
• 8:16
  SPEAKER_00: In the end, uh it turned out that our models use these metabolites and as time-of-day indicators, basically asking what time is it now?
• 8:28
  SPEAKER_00: And that has subsequently strong influence on interpretation of the other metabolite concentrations.
• 8:34
  SPEAKER_00: And since we also monitored dimlite melatonin onset of every participant by salivary melatonin analysis, we could show that some of these timekeeping metabolites are truly following our circadian rhythm.
• 8:50
  SPEAKER_00: That again is another good example of why we need more than just one or two metabolites to reliably detect sleep loss.
• 9:00
  SPEAKER_01: The study involved a relatively small, homogeneous group of young men under controlled experimental conditions.
• 9:08
  SPEAKER_01: How far can we currently generalize these findings, and what populations or real-world conditions need to be studied next?
• 9:18
  SPEAKER_00: Well, 20 participants may look like a small group on paper, but it was actually the largest group studied so far in such a setting.
• 9:27
  SPEAKER_00: And from the power calculations we did beforehand, we knew that we need at least 17 people.
• 9:33
  SPEAKER_00: So we accomplished that statistical mission, so to say.
• 11:29
  SPEAKER_00: From the regulatory point of view, we do not see high obstacles.
• 11:34
  SPEAKER_00: As I said, drowsy driving is already illegal in most countries.
• 11:40
  SPEAKER_00: To be honest, I see bigger obstacles in the acceptance of consequences of a future sleep loss test.
• 11:48
  SPEAKER_00: So sleep loss is a worldwide problem in our modern world, and I feel that we have already reached a point where we go beyond what common sense actually tells us.
• 12:01
  SPEAKER_00: So, my personal utopian vision is that we should create a world where your boss tells you not to work more, but to sleep more.
• 12:13
  SPEAKER_01: Michael, thank you very much for this interview.