Decoding nature’s chemistry: Mass spectral databases power the annotation and discovery of natural products

Nanjing, China — Natural products (NPs) have long held a significant position in medicine, food, agriculture, and materials science. More than half of the approved drugs worldwide are derived from NPs and their derivatives, which play key roles in treating cancer, infectious diseases, and cardiovascular disorders. Yet the chemical space covered by NPs remains vast and largely underexplored: traditional extraction, isolation, and characterization processes are challenging, as most NPs are naturally low-content compounds that are easily overlooked in complex mixtures or degraded during separation. A comprehensive review by Fengyao Yang, Zeyuan Liang, Haoran Zhao, Jiayi Zheng, Lifang Liu, Huipeng Song, and Guizhong Xin, published in *Chinese Journal of Natural Medicines* (2025, 23(4): 410–420), examines the current mainstream mass spectral (MS) databases and annotation methodologies, offering valuable guidance for MS-based NP annotation and the discovery of novel natural entities.

The review first provides a detailed inventory of the most commonly used MS databases, which supply the three-dimensional criteria for molecular identification — exact mass, fragmentation pattern, and retention time. LC-MS databases are categorized into three types, each with distinct advantages and limitations. Public databases such as GNPS (the largest MS platform for NPs, uniquely supporting molecular networking with iterative dereplication), MassBank (the first public repository of shared MS data), HMDB, and LIPID MAPS offer free access and broad coverage, but suffer from variable data quality. Commercial databases — including mzCloud with its distinctive mass spectral trees, METLIN, the NIST Tandem Mass Spectral Library, and Wiley MSforID — deliver strictly audited, high-quality data with mature analytical tools, at the price of paid subscription. Meanwhile, self-built databases tailored to specific research objects, such as those for Gegen-Qinlian Decoction components, plant-derived saponins, and steroid structures generated through combinatorial principles, achieve higher accuracy and specificity but require significant investment to construct and maintain. GC-MS databases, exemplified by the NIST/EPA/NIH Mass Spectral Library and Wiley Registry 2023, benefit from the high reproducibility of electron ionization.

Recognizing that less than 1% of known compounds have associated spectral information, the review then systematically presents three families of computational annotation methodologies that extend beyond the limitations of databases. In-silico spectra prediction bridges structural and mass spectral databases: quantum chemical approaches such as QCEIMS enable prediction without training sets, while machine learning frameworks like CFM-ID 4.0, LipidBlast, MetFrag, and MS-FINDER offer faster and increasingly accurate spectral simulations. Molecular fingerprint-based methodologies start directly from MS/MS spectra — the SIRIUS 4.0 platform, integrating CSI: FingerID and CANOPUS, alongside the COSMIC confidence-scoring workflow and the de novo structure generator MSNovelist (which inferred seven novel compound structures from a bryophyte dataset), now enables high-confidence annotation even of compounds absent from spectral libraries. Molecular networking (MN), exemplified by GNPS with its cosine-score similarity metric, propagates annotations across structurally related compounds, with advanced variants including feature-based molecular networking (FBMN), ion identity molecular networking (IIMN), and knowledge-guided multi-layer networks such as KGMN/MetDNA2, which raised known metabolite identification accuracy to over 95%. The review also highlights integrated and biology-driven strategies, from the NP3 MS Workflow that ranks bioactive compounds directly from complex mixtures to reverse metabolomics approaches that mine public data repositories such as MassIVE and ReDU.

In its concluding perspectives, the authors emphasize that no single perfect method exists: each approach possesses distinct characteristics, and researchers should select the most appropriate technique for their objectives. Future directions include standardized storage of diversified data, enhanced complementarity and open sharing across databases, and a holistic focus on the entire analytical workflow — from sample preparation and data acquisition (including MS probes, EIEIO fragmentation, and dpDDA) to data filtering strategies such as diagnostic ion filtration, neutral loss filtration, and mass defect filtration. Importantly, the authors stress that annotation should not be the ultimate goal but rather the groundwork for functional annotation and the exploration of NP activity and efficacy. By providing this systematic overview, the review serves as both a practical reference and a source of inspiration for researchers seeking to discover and annotate the countless valuable natural products that remain hidden in abundant natural resources.