Machine learning- and multilayer molecular network-assisted screening hunts fentanyl compounds.

Area of Science:

• Forensic Chemistry
• Analytical Chemistry
• Computational Chemistry

Background:

• Fentanyl and its analogs pose significant global health risks.
• Accurate identification of fentanyl compounds is crucial for public health and safety.
• Existing methods for fentanyl detection can be limited in scope and accuracy.

Purpose of the Study:

• To introduce Fentanyl-Hunter, a novel screening platform for fentanyl identification.
• To leverage machine learning and molecular networks for enhanced fentanyl detection.
• To assess the prevalence of fentanyl compounds in various sample types.

Main Methods:

• Development of a machine learning classifier utilizing 772 fentanyl spectra.
• Implementation of spectral binning feature engineering for classification.
• Construction of a multilayer molecular network based on spectral similarity and mass distances.
• Application of Fentanyl-Hunter to biological, environmental, and public mass spectrometry datasets.

Main Results:

• The classification model achieved a high F1 score of 0.868 ± 0.02.
• The multilayer network successfully covered over 87% of known fentanyl compounds.
• Fentanyl-Hunter identified fentanyl members in biological and environmental samples, including 35 metabolites from four derivatives.
• Norfentanyl was identified as a major fentanyl compound in wastewater.
• Retrospective analysis of over 605,000 MS files revealed fentanyl, sufentanil, norfentanyl, or remifentanil acid in over 250 samples across eight countries.

Conclusions:

• Fentanyl-Hunter is an effective platform for the accurate identification and annotation of fentanyl compounds.
• The platform demonstrates the potential for widespread presence of fentanyl and its metabolites globally.
• This tool can aid in public health surveillance and forensic investigations related to fentanyl abuse.