A Machine Learning Analysis of Big Metabolomics Data for Classifying Depression: Model Development and Validation

Many metabolomics studies of depression have been performed, but these have been limited by their scale. A comprehensive in silico analysis of global metabolite levels in large populations could provide robust insights into the pathological mechanisms underlying depression and candidate clinical biomarkers.

Depression-associated metabolomics was studied in 2 datasets from the UK Biobank database: participants with lifetime depression (N = 123,459) and participants with current depression (N = 94,921). The Whitehall II cohort (N = 4744) was used for external validation. CatBoost machine learning was used for modeling, and Shapley additive explanations were used to interpret the model. Fivefold cross-validation was used to validate model performance, training the model on 3 of the 5 sets with the remaining 2 sets for validation and testing, respectively. Diagnostic performance was assessed using the area under the receiver operating characteristic curve.

In the lifetime depression and current depression datasets and sex-specific analyses, 24 significantly associated metabolic biomarkers were identified, 12 of which overlapped in the 2 datasets. The addition of metabolic features slightly improved the performance of a diagnostic model using traditional (nonmetabolomics) risk factors alone (lifetime depression: area under the curve 0.655 vs. 0.658 with metabolomics; current depression: area under the curve 0.711 vs. 0.716 with metabolomics).

The machine learning model identified 24 metabolic biomarkers associated with depression. If validated, metabolic biomarkers may have future clinical applications as supplementary information to guide early and population-based depression detection.