Machine learning-assisted identification of new psychoactive substances in biological sample using miniaturized ambient mass spectrometer

Hang Su, Danxia Yu, Yixi Deng, Hongping Zeng, Anqi Chen, Shundi Hu, Miaoxiu Ge, Xiangyu Wang, Wei Xiong, Jiabin Jin, Luhong Wen,
Machine learning-assisted identification of new psychoactive substances in biological sample using miniaturized ambient mass spectrometer,
Microchemical Journal,
Volume 219,
2025,
115985,
ISSN 0026-265X,
https://doi.org/10.1016/j.microc.2025.115985.
(https://www.sciencedirect.com/science/article/pii/S0026265X25033338)
Abstract: New psychoactive substances (NPSs) have emerged as a global public health and forensic challenge, impacting social stability. In this study, we presented a groove-type blade spray ionization (GT-BSI) coupled with miniaturized mass spectrometer (mini-MS) for the analysis of NPS in hair, enabling the sensitive detection of etomidate, metomidate, and tiletamine. This proposed method achieved low limits of detection (LODs) of 0.018–0.034 ng/mg and low limits of quantification (LOQs) of 0.057–0.13 ng/mg, with relative standard deviations (RSDs) of less than 6.49 % for the three NPSs in hair samples, demonstrating significant advantages in analysis speed and stability. Especially, it exhibited a quantification deviation of less than 10 % compared to the laboratory confirmation method, underscoring its potential for analyzing authentic hair samples. To improve detection accuracy for low-content samples, the linear regression (LR) machine learning model was employed to classify NPS-contained and blank hair, achieving high-precision classification and identification with precision of 1 and accuracy of 100 % in training set. Additionally, when applied to authentic positive hair samples, the method achieved successful classification and identification with a validation accuracy of 97.88 %, demonstrating its effectiveness in accurate NPS identification. These findings highlight the potential of integrating ambient mass spectrometry with machine learning algorithms for the rapid identification and quantification of NPS in biological sample, contributing to on-site high-throughput screening.
Keywords: New psychoactive substances; Miniaturized mass spectrometer; Machine learning; Sensitive identification; Classification