MetaboAnnotate: An AI-powered Multi-agent Framework for Integrating Annotation Tools for Untargeted Metabolomics
TL;DR
Evaluation on the CASMI 2016 and CASMI 2022 benchmarks shows that agreement among independent annotation tools substantially reduces false discovery rates and improves annotation accuracy, and application to a fecal metabolomics dataset further demonstrates the utility of multi-tool consensus for identifying high-confidence putative metabolites.
Abstract
Metabolite annotation remains a major bottleneck in untargeted metabolomics, limiting biological interpretation of large-scale mass spectrometry datasets. Although substantial advances have been made through spectral libraries, machine learning–based annotation models, and community benchmarking efforts, many recently developed tools remain difficult to incorporate into routine workflows because they are distributed as research-oriented software with complex dependencies and nonstandard interfaces. Here we present MetaboAnnotate, a web-based framework that uses a large language model (LLM) in a multi-agent system to orchestrate multiple metabolite annotation tools through a unified natural-language interface. The system enables users to submit MS/MS spectra and execute multi-tool annotation workflows without local installation or programming expertise. The current implementation integrates complementary methods, including SIRIUS, FLARE, JESTR, and DiffMS. Evaluation on the CASMI 2016 and CASMI 2022 benchmarks shows that agreement among independent annotation tools substantially reduces false discovery rates and improves annotation accuracy. Application to a fecal metabolomics dataset further demonstrates the utility of multi-tool consensus for identifying high-confidence putative metabolites. MetaboAnnotate is available at: https://hassounlab.cs.tufts.edu/MetaboAnnotate/