Multi-omic modelling of body mass index response to a dietary weight loss intervention

Obesity is a multifactorial condition, and there is wide heterogeneity in responses to weight loss interventions. Although it remains challenging, modeling responses to weight loss interventions can help tailor treatments, increase weight loss success, or improve our understanding of underlying pathophysiology. We leveraged multi-omic (genetics; gut microbiota: taxonomy, inferred gene pathways and metabolite dynamics; blood metabolomics) and clinical data (e.g., lipids, blood glucose) from a 12-month behavioral weight loss trial of adults (n = 150) with overweight/obesity, to forecast longitudinal body mass index (BMI) and BMI change (ΔBMI) using Mixed Effects Random Forests (MERF) and GLMM-Lasso. Across modeling approaches and outcomes, routinely available clinical variables and blood metabolomics consistently improved prediction over basic demographics, and metabolomics added value beyond clinical information. Across models, the combined omic risk score most improved models of longitudinal BMI trajectories, explaining 20.5-26.0% marginal variance (R²m), whereas metabolomic risk scores most improved BMI change prediction (R²m = 52.9-59.3%). Gut microbial taxonomy and inferred gene pathways offered modest but significant gains for some models and outcomes, while metabolite dynamics consistently failed to enhance performance. The most important features in the models included insulin, glycoprotein acetyls, lipoprotein sizes, and certain amino acids, aligning with known inflammatory and metabolic mechanisms. These findings support that select blood-based biomarkers correlate with individual responses to weight loss efforts.