PS²MS: A Deep Learning-Based Prediction System for Identifying New Psychoactive Substances Using Mass Spectrometry.

Area of Science:

• Forensic Chemistry
• Computational Chemistry
• Drug Enforcement Science

Background:

• Emerging novel psychoactive substances (NPS) challenge traditional mass spectrometry due to a lack of reference spectra.
• Accurate identification of illicit drugs is crucial for public health and law enforcement.

Purpose of the Study:

• To introduce PS²MS, an AI-powered predictive system for identifying novel illicit drugs.
• To overcome limitations in identifying emerging NPS lacking spectral data.

Main Methods:

• PS²MS generates a synthetic NPS database by enumerating feasible derivatives of known substances.
• Deep learning is employed to predict mass spectra and chemical fingerprints for these derivatives.
• An integrated approach compares analyte mass spectra and chemical fingerprints against the synthetic database to deduce identities.

Main Results:

• The PS²MS system demonstrated effectiveness in identifying cathinone derivatives in real forensic evidence.
• The AI-powered system successfully identified unknown substances where conventional methods failed.

Conclusions:

• PS²MS shows significant potential for the practical identification of emerging drugs of abuse.
• This AI system offers a valuable tool for researchers and forensic experts in combating NPS proliferation.