Gut Microbes May Predict Who Keeps Weight Off After Fasting, Study Finds

A five-day fast can trim pounds and shift the trillions of microbes living in the human gut, but the real puzzle has always been what happens afterward. Some people hold on to the benefits for months; others drift back to their starting point almost as quickly as the fast ends. A new randomized, controlled trial published in Genome Medicine suggests that the answer may already be written in the body before the first skipped meal. Using machine learning trained on baseline gut microbiome profiles and routine clinical measurements, researchers led by a team from the Max Delbrück Center for Molecular Medicine, Charité – Universitätsmedizin Berlin, and their collaborators were able to predict how much body mass index an individual would lose three months after a prolonged fasting intervention. The findings point toward a future in which fasting prescriptions could be tailored to a person’s biology rather than applied as a one-size-fits-all metabolic reset.

The study, known as LEANER, enrolled 38 healthy adults who completed a five-day fasting protocol and were followed for twelve weeks afterward in a randomized, waitlist-controlled design. Fasting interventions have surged in popularity over the past decade, yet high-quality controlled data in healthy adults, with longer follow-up and multi-omics profiling, have remained scarce. This trial was designed to close that gap. Researchers tracked body mass index and body composition, sequenced gut microbiome composition, and measured both plasma and fecal metabolites before the fast, at its conclusion, and throughout the three-month follow-up period. Changes over time and between fasting and waitlist groups were evaluated with regression-based models and paired non-parametric tests, while permutation-based multivariate testing was applied to microbiome and metabolome data to detect coordinated shifts across hundreds of microbial species and metabolites simultaneously.

The acute results confirmed what smaller studies had hinted at: fasting rapidly lowered body mass index, and the reduction was driven predominantly by the loss of fat mass rather than lean tissue. Crucially, some of that improvement persisted at twelve weeks, even after participants returned to their ordinary eating habits. Alongside the changes on the scale, the fast produced marked remodeling of the gut microbiome and of metabolite profiles in both blood and stool. The researchers observed that post-fasting and longer-term changes in microbial diversity were associated with each participant’s baseline microbiome diversity, an observation with potentially significant implications. If the state of the gut ecosystem before fasting shapes how it responds and rebounds afterward, then the microbiome is not merely a passive bystander in the fasting response but a determining factor in its durability.

The heart of the study, however, lay in its predictive modeling. Rather than simply cataloguing what changed, the team asked whether pre-intervention data alone could forecast who would respond best. They built a data-driven machine learning model combining baseline microbiome features with clinical variables to predict individual body mass index response at twelve weeks, training and testing it with rigorous cross-validation inside the trial. The approach is technically demanding: with hundreds of candidate microbial species and dozens of clinical covariates measured in a modestly sized cohort, the risk of overfitting is substantial. Cross-validation, in which the model repeatedly trains on most participants and is tested on held-out individuals, provides a safeguard, ensuring that reported predictive performance reflects genuine signal rather than memorized noise.

Among the most prominent predictors identified were two specific gut bacteria: an unclassified species of Faecalibacterium, a genus widely associated with anti-inflammatory activity and the production of short-chain fatty acids, and Oscillibacter sp. 50_27. Clinical variables also carried weight, with baseline low-density lipoprotein cholesterol and systolic blood pressure emerging as key predictors alongside the microbial features. The inclusion of both domains is what distinguishes the model from earlier attempts at personalized nutrition forecasting. Microbial taxa capture the ecology of the gut ecosystem, while lipoprotein levels and blood pressure reflect systemic cardiometabolic status, and the model’s reliance on both suggests that sustained weight-loss response after fasting is a whole-body phenomenon rather than a purely intestinal one.

The most convincing evidence for the model came from its performance beyond the original trial. The team validated the model externally in three independent cohorts of people undergoing prolonged fasting protocols: individuals with metabolic syndrome, patients with multiple sclerosis who had been exposed to repeated fasting, and healthy volunteers fasting for six to twelve days. These cohorts differed in health status, fasting duration, and study context, yet the model retained its ability to stratify expected responses. External validation of this kind is rare in microbiome-prediction research, where many published models collapse when applied outside their discovery dataset. The authors note that their study was prospectively registered at ClinicalTrials.gov under identifier NCT04452916, and that the analysis pipeline and feature lists were made available, supporting reproducibility.

The mechanistic implications are already intriguing researchers. Faecalibacterium and Oscillibacter are both fermenters that produce metabolites capable of influencing host energy balance, inflammation, and even blood pressure regulation. The team supplemented their modeling with correlation analyses linking these species to fecal metabolites, hinting at concrete biochemical routes through which the gut ecosystem might modulate how the body stores and sheds fat. One plausible reading is that a microbiome already enriched in certain butyrate-producing organisms sets the metabolic stage for a more durable response to caloric deprivation, while a different starting community biases the body toward rebound weight gain. The new study does not prove causation, and the authors are appropriately cautious on this point, but it establishes a clear testable framework for intervention studies that manipulate the microbiome before fasting.

The clinical stakes are considerable. Obesity and metabolic syndrome affect hundreds of millions of people worldwide, and prolonged fasting has moved from fringe practice to mainstream interest, with medically supervised programs in Germany and elsewhere offering fasts lasting from five days to several weeks. Yet response is notoriously variable, and no simple biomarker currently tells a clinician whether a given patient will benefit. A validated predictive panel combining a microbiome profile with standard clinical labs such as LDL cholesterol and blood pressure could, in principle, allow physicians to identify likely responders before recommending an intensive intervention, and to direct those predicted to respond poorly toward alternative strategies. It would also help researchers design better trials, stratifying participants by predicted response and reducing the noise that has historically obscured modest but real effects in nutrition studies.

Limitations remain. The discovery cohort comprised 38 healthy adults, a sample small by machine learning standards even if the external validation cohorts bolster confidence in generalizability. The fasting protocol was short and the follow-up limited to twelve weeks, so predictions of response over one year or longer remain untested. Microbiome sequencing and metabolomics also vary across platforms and laboratories, meaning the specific model would need retraining or calibration before deployment in new settings. Still, the study represents one of the most rigorous demonstrations to date that routine pre-intervention biology can forecast the long-term outcome of a dietary intervention. As the authors conclude, baseline microbiome and clinical characteristics can help stratify expected longer-term responses, supporting the development of individualized, fasting-based interventions. In a field crowded with hype and one-size-fits-all promises, that measured claim, backed by cross-validation and replication across three independent populations, may prove to be the study’s most durable contribution.

Subject of Research: Machine learning prediction of sustained weight loss after prolonged fasting using baseline gut microbiome and clinical data

Article Title: Machine learning identifies microbiome and clinical predictors of sustained weight loss following prolonged fasting

Article References: Machine learning identifies microbiome and clinical predictors of sustained weight loss following prolonged fasting. (n.d.). https://doi.org/10.1186/s13073-026-01765-0

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01765-0

Keywords: prolonged fasting, gut microbiome, machine learning, weight loss, metabolic health, Faecalibacterium, Oscillibacter, multi-omics, personalized nutrition, precision medicine, body mass index, Genome Medicine