MSNovelist

MSNovelist predicts de novo molecular structures from tandem mass spectrometry (MS²) spectra to enable structure elucidation of small molecules without relying on reference spectral libraries.

Key Features:

• Fingerprint prediction and encoder–decoder neural network: Combines predicted molecular fingerprints with an encoder–decoder neural network architecture to generate molecular structures directly from MS² spectra.
• De novo structure elucidation: Predicts structures de novo, enabling identification of novel or poorly represented analyte classes absent from spectral databases.
• Performance metrics: Evaluation on 3,863 MS² spectra from the Global Natural Product Social Molecular Networking site (GNPS) yielded a first-rank correct prediction rate of 25%, retrieved 45% of structures overall, and reproduced 61% of correct database annotations; in the CASMI 2016 challenge MSNovelist alone correctly predicted 26% of structures, retrieved 57%, and recovered 64% of correct database annotations.
• Application to bryophyte MS² data: In a bryophyte MS² dataset, de novo predictions by MSNovelist outperformed the best database candidates for seven spectra.

Scientific Applications:

• Natural Product Discovery: Facilitates identification and structural characterization of novel natural products from MS² data.
• Metabolomics and Cheminformatics: Enhances analysis of complex biological samples by proposing structures for novel or underrepresented metabolites from MS² spectra.
• Drug Discovery and Development: Assists structural elucidation of potential drug candidates derived from mass spectrometry analyses.

Methodology:

Combines fingerprint prediction with an encoder–decoder neural network to generate molecular structures directly from MS² spectra.