TY  - JOUR
T1  - Machine Learning in Tandem Mass Spectrometry for Natural Product Discovery Through Spectral Annotation, Molecular Networking, Dereplication, and Structure Prioritization
A1  - Pieter Botha
A1  - Anel Van Wyk
A1  - Johan Marais
JF  - International Journal of Pharmaceutical And Phytopharmacological Research
JO  - Int J Pharm Phytopharmacol Res
SN  - 2250-1029
Y1  - 2025
VL  - 15
IS  - 6
DO  - 10.51847/1PW0moEsyp
SP  - 105
EP  - 114
N2  - Tandem mass spectrometry has shifted from a predominantly library-centered identification technique toward a computational discovery environment in which spectra can be embedded, networked, propagated, classified, ranked, and used to generate structural hypotheses. For natural products, this expansion is attractive because chemical diversity is high and authenticated reference spectra remain incomplete, yet the resulting outputs span substantially different levels of structural certainty. A systematic mapping review was conducted across analytical chemistry, cheminformatics, metabolomics, and natural-product informatics literature. DOI-unique records published from 2017 through 2025 were screened for peer-reviewed evidence in which MS/MS materially contributed to spectral annotation, molecular networking, dereplication, formula or substructure inference, candidate ranking, de novo generation, or contextual prioritization. Evidence was charted by discovery task, representation, validation design, inference level, uncertainty source, and transferability boundary. Forty-two unique records entered screening and 30 studies were included after title/abstract and full-text assessment. The evidence map shows movement from database search and fragmentation-based inference toward learned spectral representations, analogue retrieval, graph and transformer models, joint spectrum–molecule embeddings, and generative approaches. Better retrieval or ranking nevertheless remains conditional on library composition, candidate-space construction, acquisition conditions, network anchors, and independent structural confirmation. Machine learning broadens the structural hypotheses extractable from MS/MS, but its outputs should be interpreted as task-specific evidence rather than interchangeable forms of identification. The review proposes a task-and-confidence mapping that separates computational reach from structural confirmation and identifies cross-instrument, chemically disjoint, prospective validation as a major priority.
UR  - https://eijppr.com/article/machine-learning-in-tandem-mass-spectrometry-for-natural-product-discovery-through-spectral-annotati-6qb9j0yycowfjsc
ER  -