CBMAFF-Net:\nAn Intelligent NMR-Based Nontargeted Screening\nMethod for New Psychoactive Substances

With the proliferation and rapid evolution of new psychoactive\nsubstances (NPSs), traditional database-based search methods face\nincreasing challenges in identifying NPS seizures with complex compositions,\nthereby complicating their regulation and early warning. To address\nthis issue, CBMAFF-Net (CNN BiLSTM Multistep Attentional Feature Fusion\nNetwork) is proposed as an intelligent screening method to rapidly\nclassify unknown confiscated substances using <sup>13</sup>C nuclear\nmagnetic resonance (NMR) and <sup>1</sup>H NMR data. Initially, we\nutilize the synergy of a convolutional neural network (CNN) and bidirectional\nlong short-term memory network (BiLSTM) to extract the global and\nlocal features of the NMR data. These features are sequentially fused\nthrough a weighted approach guided by an attention mechanism, thoroughly\ncapturing the essential NPS information. We evaluated the model on\na generated simulated data set, where it performed with 99.8% accuracy\nand a 99.8% F1 score. Additionally, testing on 42 actual seizure cases\nyielded a recognition accuracy of 97.6%, significantly surpassing\nthe performance of conventional database-based similarity search algorithms.\nThese findings suggest that the proposed method holds substantial\npromise for the rapid screening and classification of NPSs.