(P26-014-25) Beyond the Known Metabolome: Predictive Strategies for Dark Matter Discovery

Objectives: A major obstacle in deciphering the interactions between metabolites and their associated protein counterparts, particularly within the mysterious category of "dark matter" metabolites. This study introduces a comprehensive workflow for predicting dark matter metabolites, emphasizing the usage and amalgamation of advanced computational techniques and machine learning algorithms.

Methods: The workflow begins with data collection from untargeted metabolomics analysis, followed by prioritization of key metabolites through filtering of probable (level 1 and 2) and improbable annotations to isolate dark matter metabolites. Clustering and predictive modeling techniques were employed, incorporating molecular similarity searches (PubChem) and target prediction tools such as GLASS (GPCR-ligand database), BindingDB, and SwissTargetPrediction. Additional computational tools included AutoDock for binding energy estimation and SWISS ADME for evaluating properties such as blood-brain barrier penetration, skin permeability, gastrointestinal absorption, and bioavailability. Metabolite pathway mapping was performed using the KEGG database to assess pathway influence. For metabolites that could not be identified due to insufficient spectral data or absence in existing databases, machine learning algorithms were utilized to predict molecular structures.

Results: The workflow facilitated a comprehensive evaluation of metabolite significance. Pilot data from blood serum analysis revealed 1021 identified metabolites out of 1836 dark matter candidates, representing a 56% improvement in metabolite identification. Applying SMILES-based clustering with Tanimoto similarity thresholds at different cutoffs yielded 234, 330, and 412 unique compounds at thresholds of 0.7, 0.6, and 0.5, respectively.

Conclusions: This workflow enhances the identification and characterization of dark matter metabolites by combining data-driven strategies with informed interpretation by addressing gaps in existing spectral libraries and enabling structural predictions for unidentified compounds.

Funding Sources: USDA-ARS, US Department of Agriculture – Agricultural Research Service