NPS-MS

NPS-MS predicts MS/MS spectra from chemical structures to enable identification of known and novel new psychoactive substances (NPS) without requiring reference standards for forensic and toxicological analysis.

Key Features:

• Deep learning MS/MS prediction: Predicts tandem mass spectrometry (MS/MS) spectra from chemical structures using a deep learning model.
• Transfer learning training: The model is trained via transfer learning from a generic MS/MS prediction model on an extensive dataset of MS/MS spectra.
• Reference-standard-free identification: Enables identification of NPS in the absence of costly or unavailable reference standards.
• Large predicted database: Provides search capability against approximately 8.7 million predicted NPS compounds from DarkNPS and 24.5 million predicted ESI-QToF-MS/MS spectra.
• Novel derivative detection: Identifies novel derivatives, exemplified by detection of a new phencyclidine (PCP) derivative in a seized powder.
• Improved identification accuracy: Enhances accuracy of identifying NPS from experimentally acquired MS/MS spectra relative to existing methodologies.

Scientific Applications:

• Forensic toxicology: Identification of known and newly emerging NPS in samples analyzed by forensic and toxicological laboratories.
• MS-based spectral matching: Matching experimentally acquired MS/MS spectra to predicted spectra, including ESI-QToF-MS/MS, for compound identification.
• Illicit drug surveillance: Rapid screening and identification of novel NPS derivatives in seized samples to support forensic investigations.

Methodology:

Uses deep learning to predict MS/MS spectra from chemical structures and is trained via transfer learning from a generic MS/MS prediction model on an extensive dataset of MS/MS spectra, producing predicted ESI-QToF-MS/MS spectra for database searching.