Fentanyl analogs classification via Siamese network and mass spectral library searching. (1st May 2023)

• Record Type: Journal Article
• Authors: Xu, Ping; Zhao, Yang; Xue, Lingyun; Liu, Yian; Yan, Ming; Zhu, Lei; Lin, Weng; Hu, Shundi; Wen, Luhong
• Is Part Of: Expert systems with applications. Volume 217(2023)
• Publication Date: 2023-05-01
• Subjects: Fentanyl analogs classification -- Mass spectrum -- Siamese network -- Deep learning; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2023.119534
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A novel classification model via Siamese network for fentanyl analogs was proposed; Siamese network and spectral library searching were combined; The proposed model can achieve relatively high classification accuracy; Problem of small sample classification of fentanyl analogs was resolved effectively. Abstract: Fentanyl and its analogs, as emerging psychotropic drugs, have led to a sharp increasing fatality due to their abuse in recent years. It is difficult to identify their differences due to the diversified molecular structures and small sample characteristics. This paper proposed a novel deep classification model based on Siamese network and mass spectral library searching to classify fentanyl analogs accurately. After embedding the query mass spectrum and reference spectrum into a low-dimensional space, the best matched spectrum is obtained by calculating their similarity, so as to determine the category of the query analogue. Three experiments were performed to verify the classification performance of the proposed model on two open datasets of fentanyl analogs. Compared with two spectral library searching methods of simple match factor (sMF) and hybrid match factor (hMF), four machine learning methods of linear discriminant analysis (LDA), support vector machine (SVM), random forest (RF) and Adaboost, and two deep learning methods of deep clustering and contrastive learning, the proposed model can achieve the highest classification accuracy of 96.13%, 95.83% and 94.17%, respectively.