Summary: A new machine-learning approach can estimate a person’s internal body clock by analyzing metabolites in blood. This technique enables personalized guidance for sleep timing, meal scheduling, and clinical sampling by predicting circadian phase with minimal blood sampling. It offers a less invasive, practical route to align daily behaviour and treatments with an individual’s biology, potentially lowering the health risks associated with poor sleep and mistimed eating.
Source: University of Surrey
A machine-learning method has been developed to predict the timing of the human internal body clock, providing a practical way to improve sleep, eating patterns, and health-related decisions.
Researchers at the University of Surrey and the University of Groningen trained a machine-learning model to read metabolite patterns in plasma and estimate the phase of the circadian timing system. The method aims to provide a convenient estimate of dim light melatonin onset (DLMO), the standard marker for circadian phase, using a small number of blood samples rather than continuous monitoring or more invasive tests.
Currently, the most reliable technique for assessing circadian timing is measuring the timing of the natural melatonin rhythm, specifically the dim light melatonin onset (DLMO), which marks when melatonin production begins in dim conditions. However, DLMO measurement typically requires frequent sampling under controlled conditions, which limits real-world usability. This new metabolomics-based machine-learning approach seeks to retain accuracy while greatly simplifying the sampling process.
The study collected a time series of plasma samples from 24 healthy adult volunteers—12 men and 12 women—who followed regular sleep schedules and did not smoke in the week before testing. Researchers measured rhythms in more than 130 metabolites using targeted metabolomics and then applied a partial least squares regression (PLSR) machine-learning model to predict each participant’s DLMO from one or two blood samples.
Professor Debra Skene of the University of Surrey, a co-author of the study, said the new method offers a less intrusive way to estimate DLMO: “After taking two blood samples from our participants, our method was able to predict the DLMO of individuals with an accuracy comparable to or better than previous, more intrusive estimation methods.”
Professor Skene added that the approach is promising but requires broader validation: “We are excited but cautious about our method. It requires fewer samples and is more convenient than some existing tools. With validation in diverse populations and real-world conditions, it could help optimise treatment for circadian rhythm sleep disorders and support recovery after injury.”
The research emphasizes real-world applicability: the protocol was designed to reflect normal daily conditions rather than highly controlled laboratory settings. In those testing conditions, metabolomics-based estimates optimized separately for women and men performed as well as or better than more resource-intensive methods that rely on RNA sequencing.
Professor Roelof Hut of the University of Groningen, also a study co-author, highlighted potential applications: “These results could lead to affordable, practical ways for people to estimate their personal circadian rhythm. That information would help to optimise the timing of behaviours, diagnostic sampling, and medical treatment.”
About this machine learning and circadian rhythm research news
Author: Dalitso Njolinjo
Source: University of Surrey
Contact: Dalitso Njolinjo – University of Surrey
Image: The image is in the public domain
Original Research: Closed access. “Machine learning estimation of human body time using metabolomic profiling” by Debra Skene et al., published in PNAS.
Abstract
Machine learning estimation of human body time using metabolomic profiling
Circadian rhythms shape physiology, metabolism, and many molecular processes. Estimating an individual’s body time — their circadian phase — is important for tailoring daily behaviour (sleep timing, meal schedules, exercise), for planning diagnostic sampling, and for optimising medical treatments and interventions for circadian rhythm disorders.
This study presents a partial least squares regression (PLSR) machine-learning approach that uses plasma metabolomics from one or more blood samples to estimate DLMO as a proxy for circadian phase. The protocol was intentionally close to real-life conditions to increase practical applicability.
Results indicate that a targeted blood metabolomics approach, when optimized for sex-specific patterns under entrained conditions, can deliver circadian phase estimates that match or exceed the performance of existing, more labor-intensive RNA sequencing techniques. While further validation is needed—particularly among shift workers, clinical populations, and other real-world scenarios—this method shows promise as a robust and feasible tool to support personalised timing of behaviour and clinical care following appropriate validation in broader patient groups.