Journal: npj Antimicrobials and Resistance

Article Title: Accelerating natural product discovery with linked MS-genomics and language/transformer-based models

doi: 10.1038/s44259-026-00206-7

Figure Lengend Snippet: Putative producers of compounds of interest are prioritized through cross-validation of the top ranked strains via Protein Language Model (PLM) embeddings and via our in-house developed Workflow for Intelligent Structural Elucidation (WISE).

Article Snippet: The data processing Workflow for Intelligent Structural Elucidation (WISE) is partially based on Mestrenova v15.0.1-35756 which is commercially available from Mestrelab Research S. L. ( www.mestrelab.com ).

Techniques: Biomarker Discovery

Journal: npj Antimicrobials and Resistance

Article Title: Accelerating natural product discovery with linked MS-genomics and language/transformer-based models

doi: 10.1038/s44259-026-00206-7

Figure Lengend Snippet: A Overview of the cheminformatics analysis workflow (WISE) to annotate individual molecules from raw liquid chromatography-tandem mass spectral (LC-MS/MS) data of fermentation extracts. B Distribution of WISE scores for correct structures (Real) versus isobaric decoy structures (within 0.005 m/z) with different molecular formulae (Decoy) from the Critical Assessment for Small Molecule Identification (CASMI2022) dataset. Low and high WISE score thresholds of 0.35 and 0.65 respectively are annotated for reference. C Receiver Operating Characteristic (ROC) curve demonstrating the discrimination of correct structures from decoys with differing molecular formulae within a 0.005 m/z window. AUC = Area Under Curve. CI = Confidence Interval. Detailed data in Table .

Article Snippet: The data processing Workflow for Intelligent Structural Elucidation (WISE) is partially based on Mestrenova v15.0.1-35756 which is commercially available from Mestrelab Research S. L. ( www.mestrelab.com ).

Techniques: Liquid Chromatography, Liquid Chromatography with Mass Spectroscopy

Journal: npj Antimicrobials and Resistance

Article Title: Accelerating natural product discovery with linked MS-genomics and language/transformer-based models

doi: 10.1038/s44259-026-00206-7

Figure Lengend Snippet: A Elucidated biosynthetic gene cluster BGC0000709, responsible for neomycin production. B Similarity (%) of each embedded protein and sequence similarity (%) of the top-scoring protein matches. C Predicted protein structures of the top matches for Fe-S oxidoreductase ( CAF33317 ) from neomycin BGC. D Heatmap of WISE scores for strain-media combinations derived from the liquid chromatography-tandem mass spectrometry (LC-MS/MS) data of fermentation extracts. Only strains with at least 1 media combination with non-zero WISE score are shown. Activated strains (if any) are referred to by their parent strain instead of individually (details of the activated strains are in Table ). E Structure of neomycin B. F Workflow of the co-occurrence strategy and matrix ranking table with 0.35 WISE score and 0.75 PLM score cutoff. Strains that have been cross validated are highlighted in green or yellow, indicating whether these compounds were present (green) or absent (yellow). Lines indicating higher confidence thresholds of 0.65 WISE score (in red) and 2.5 PLM score (in blue) are also included.

Article Snippet: The data processing Workflow for Intelligent Structural Elucidation (WISE) is partially based on Mestrenova v15.0.1-35756 which is commercially available from Mestrelab Research S. L. ( www.mestrelab.com ).

Techniques: Sequencing, Derivative Assay, Liquid Chromatography, Mass Spectrometry, Liquid Chromatography with Mass Spectroscopy