Prediction for Weight Loss and Regain Based on Multi-Omics and Phenotypic Features: Results from a Caloric-Restricted Feeding Trial

Objective To identify baseline multi-omics and phenotypic predictors and develop prediction models for weight and body composition loss and regain in the low-carbohydrate diet and time-restricted eating (LEAN-TIME) trial. Research Design and Methods Post hoc analysis of the LEAN-TIME trial using data from 88 adults with overweight/obesity completing a 12-week caloric-restricted weight loss phase and 79 completing a 28-week weight regain phase. Baseline dietary, metabolic, fecal metabolites, and gut microbiome data were candidate predictors for changes in weight, body fat mass (BFM), and soft lean mass (SLM). Multivariable regression and least absolute shrinkage and selection operator model were used to identify predictors and develop weighted-sum prediction models. Results Multi-omics and phenotypic models significantly outperformed phenotypic-only models (P < 0.05), demonstrating strong predictive performance during both phases. During weight loss, the model yielded R² values of 0.49, 0.61, and 0.54 for changes in weight, BFM, and SLM, with corresponding RMSEs of 1.59, 1.41, and 0.98 kg. For binary classification of clinically meaningful weight loss (≥5%), the model achieved an area under the curve of 0.95 (sensitivity 94.12%, specificity 86.79%). During weight regain, R² reached 0.72, 0.73, and 0.66 for weight, BFM, and SLM (RMSEs: 1.40, 1.62, and 0.73 kg). Several key baseline predictors, primarily gut microbes and fecal metabolites, such as N-Acetyl-L-aspartic acid, Ruminococcus callidus, and Bifidobacterium adolescentis, were shared for weight and body composition changes during both phases. Conclusions Baseline multi-omics and phenotypic data effectively predict weight and body composition loss and regain, offering insights for personalized weight management.