Deep MDRB: An integrated platform enabling high-coverage in vivo metabolite annotation in natural medicines

Comprehensive identification of in vivo metabolites from natural medicines (NMs) remains challenging due to their chemical complexity and the limited reference spectra. Here, we presented the deep mass dataset relevance bridging (Deep MDRB) platform, an integrated platform powered by a diagnostic weighted cosine similarity (DWCS) algorithm that efficiently facilitates high-coverage annotation of in vivo metabolites. By integrating mass spectral databases with biosynthetic and metabolic reaction rules, Deep MDRB enables fully automated non-targeted analysis, including peak extraction, background correction, multimodal data alignment, multilevel similarity matching, and parent–metabolite network bridging. As a proof of concept, Deep MDRB was applied to in vitro liver microsome models of verapamil and berberine, as well as an in vivo metabolic study involving mixed reference standards (MRS) and Tongxinluo (TXL) capsule in rats. The platform identified 1565 parent compounds with a 94.4% identification rate for the TXL capsule and constructed a biotransformation network of 29 in vitro , 83 (MRS) and 2069 (TXL capsule) in vivo metabolite nodes, respectively, with a correlation rate exceeding 90%. Compared to existing software, Deep MDRB substantially improves annotation coverage and network integrity. Overall, the “knowledge-data” dual-driven framework of Deep MDRB provides a novel strategy for metabolite profiling in complex systems.