CBMAFF-Net: An Intelligent NMR-Based Nontargeted Screening Method for New Psychoactive Substances

Area of Science:

• Forensic Chemistry
• Analytical Chemistry
• Computational Chemistry

Background:

• New psychoactive substances (NPSs) rapidly evolve, challenging traditional identification methods.
• Complex NPS compositions in seizures hinder effective regulation and early warning.
• Database-based searches struggle with novel and mixed NPS samples.

Purpose of the Study:

• To develop an intelligent screening method for rapid classification of unknown confiscated substances.
• To address the limitations of traditional methods in identifying complex NPS seizures.
• To utilize nuclear magnetic resonance (NMR) data for enhanced NPS detection.

Main Methods:

• Proposed CBMAFF-Net (CNN BiLSTM Multistep Attentional Feature Fusion Network).
• Employed convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) for feature extraction.
• Utilized an attention mechanism for weighted feature fusion of 13C NMR and 1H NMR data.

Main Results:

• CBMAFF-Net achieved 99.8% accuracy and 99.8% F1 score on simulated data.
• The model demonstrated 97.6% accuracy on 42 actual NPS seizure cases.
• Outperformed conventional database-based similarity search algorithms significantly.

Conclusions:

• CBMAFF-Net shows substantial promise for rapid screening and classification of NPSs.
• The method enhances the ability to identify complex NPS compositions.
• Offers a powerful tool for forensic analysis and regulatory control of NPSs.