CANOPUS

CANOPUS predicts compound classes directly from tandem mass spectrometry (MS/MS) fragmentation spectra to enable systematic annotation and classification of metabolites, including unknowns lacking spectral or structural references.

Key Features:

• Input: processes tandem mass spectrometry (MS/MS) fragmentation spectra as input.
• Model: uses a deep neural network to assign spectra to compound classes.
• Class coverage: assigns spectra into one of 2,497 biologically relevant compound classes.
• Unknown compound classification: predicts compound classes without requiring spectral or structural reference data from existing libraries.
• Evaluation: evaluated on reference datasets with cross-validation achieving an average accuracy of 99.7%.
• Comparative performance: outperformed four baseline methods in comparative studies.

Scientific Applications:

• Exploratory metabolomics: provides compound-class-level annotation to reveal chemical diversity in biological systems.
• Natural product discovery: aids identification and characterization of novel natural products, including marine compounds.
• Microbiome studies: applied to analyze chemical diversity changes during microbial colonization of the mouse digestive system.
• Plant chemodiversity: applied to study chemodiversity in Euphorbia plants.

Methodology:

Uses a deep neural network to process MS/MS fragmentation spectra and assign them to one of 2,497 compound classes; performance was assessed on reference datasets using cross-validation (average accuracy 99.7%) and compared against four baseline methods.