BAGO: A Self-Optimizing Tool for LC-MS Gradient Design in Metabolomics

A self-driving metabolomics laboratory has long been envisioned but remains largely unrealized due to the complexity of analytical method design. As an initial step toward this goal, we developed BAGO, a self-optimizing framework for automated liquid chromatography (LC) gradient design in mass spectrometry-based untargeted metabolomics. BAGO aims to enhance global metabolite detection by improving the separation of all compounds, regardless of whether their identities are known or unknown. It operates through a data-driven Bayesian optimization process that iteratively learns from acquired MS data to propose improved gradients. To support this, we propose a global separation index that quantifies coelution among both annotated and unannotated features, enabling robust and structure-agnostic optimization across diverse sample types. Benchmarking across four metabolomics assays involving diverse sample matrices, column chemistries, and gradient durations, BAGO achieved substantial improvements within only 10 optimization iterations by balancing exploration and exploitation. The optimized gradients led to increased numbers of Gaussian-shaped peaks, higher MS/MS acquisition rates, and more annotated metabolites using both identity and analog search approaches. We further applied BAGO to a sex-differentiated metabolomics study of Drosophila abdominal carcasses, completing the workflow in parallel under both initial and optimized gradients. The optimized method resulted in a 41.9% increase in Gaussian-shaped peaks, a 36.8% increase in MS/MS-acquired peaks, and the identification of 18 additional biologically significant metabolites, including sex-associated compounds such as octopamine and pyroglutamic acid. BAGO (https://github.com/HuanLab/bago) is freely available as an open-source tool and represents a generalizable step toward fully automated, self-optimizing experimental workflows in untargeted metabolomics.