A Novel Deep Siamese Convolution Network for Detecting Fentanyl Analogs From Mass Spectra

Area of Science:

• Analytical Chemistry
• Forensic Science
• Computational Chemistry

Background:

• Rising mortality rates linked to fentanyl and analog misuse necessitate improved detection methods.
• Existing classification models struggle with the accurate identification of diverse fentanyl analogs due to easy synthesis and rapid emergence.
• Small sample size detection of fentanyl analogs from electron impact (EI) or electrospray ionization (ESI) mass spectra presents a significant challenge.

Purpose of the Study:

• To propose a novel deep Siamese convolutional network (DSCN) for accurate classification and detection of fentanyl analogs.
• To address the limitations of current machine learning and deep learning models in identifying fentanyl analogs from mass spectral data.
• To enhance the detection capabilities for fentanyl analogs, particularly in scenarios with limited sample sizes.

Main Methods:

• Mass spectra augmentation to create input spectral pairs for the Siamese network.
• Utilizing a 1D convolutional neural network (CNN) within the Siamese architecture for effective spectral feature extraction.
• Employing a classification network with fully connected (FC) and Softmax layers for final analog detection.
• Combining contrastive loss and cross-entropy loss functions for DSCN parameter training.

Main Results:

• The proposed DSCN model demonstrated superior performance in detecting fentanyl analogs compared to other established machine learning and deep learning techniques.
• The DSCN effectively extracts relevant spectral features for distinguishing between fentanyl analogs.
• The model achieved higher accuracy in classification tasks involving small sample sizes of mass spectra.

Conclusions:

• The developed DSCN offers a promising and effective solution for the accurate detection of fentanyl analogs.
• This deep learning approach significantly advances the capabilities for identifying illicit substances based on mass spectral analysis.
• The DSCN model holds potential for application in forensic science and public health initiatives aimed at combating the opioid crisis.