IDSL_MINT

IDSL_MINT predicts molecular fingerprint descriptors from tandem mass spectrometry (MS/MS) spectra using transformer-based deep learning to improve annotation in untargeted metabolomics and exposomics.

Key Features:

• Transformer Model Utilization: Employs transformer architectures adapted from large language models to translate complex MS/MS spectra into molecular fingerprint descriptors.
• Customizable Training: Trains models on user-provided reference MS/MS libraries and accepts customizable molecular fingerprint descriptors for tailored model development.
• Benchmarking and Performance: Benchmark tests using the LipidMaps database reported improved annotation rates for MS/MS spectra that were unannotated by conventional mass spectral libraries.

Scientific Applications:

• Untargeted Metabolomics: Enhances structural annotation of MS/MS spectra to increase identification coverage in untargeted metabolomics datasets.
• Exposomics: Improves annotation of exposome-related MS/MS spectra to support detection and characterization of environmental chemicals.
• Biomarker and Pathway Studies: Converts raw spectra into molecular fingerprints to facilitate biomarker discovery and insights into metabolic pathways.

Methodology:

Trains transformer-based deep learning models on reference MS/MS libraries provided by users to predict molecular fingerprint descriptors from raw MS/MS spectra, with support for customizable fingerprint descriptor sets.