To carry out the study, the team recruited 20 healthy young adult men who typically sleep between seven and nine hours per night. Each participant completed three different sleep conditions in random order, spaced a week apart: a full night of sleep deprivation, four nights of restricted sleep (two hours less than usual), and a well-rested condition involving around eight hours of sleep.

Saliva samples were collected before and after each condition and analysed for metabolic changes. The researchers identified 10 key molecular differences between the sleep-deprived and well-rested states. However, they found no significant metabolic differences between the well-rested and sleep-restricted conditions.

Using these findings, the team developed a machine-learning model capable of predicting sleep deprivation based on saliva composition. The model correctly identified sleep-deprived samples 94 per cent of the time.

The researchers believe the errors stemmed from individual differences in metabolism. In some cases, participants who had been awake for 24 hours did not return to a fully "rested" metabolic state even after eight hours of sleep, suggesting recovery rates may vary between individuals.

The findings point to what researchers describe as a "sleepiness fingerprint" - a pattern of salivary metabolites that could eventually be used to detect dangerous fatigue in real-world settings such as roadside checks or clinical assessments.

The team is now planning a larger international study involving more than 1,000 samples, including shift workers, women, and frequent drivers, to further validate the model.

While still in its early stages, the research suggests saliva-based screening could one day provide a practical tool for identifying sleep-deprived individuals before accidents occur.