Random forest models accurately classify synthetic opioids using high-dimensionality mass spectrometry datasets.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Forensic Science

Background:

• Untargeted detection of unknown chemical structures, like novel synthetic opioids, is a significant challenge.
• Machine learning (ML) offers potential for developing broad-spectrum analytical methods for threat agent identification.
• Existing methods struggle with the vast and evolving landscape of potential chemical threats.

Purpose of the Study:

• To develop and validate machine learning models for the untargeted classification of synthetic opioids.
• To assess the efficacy of logistic regression and random forest algorithms using mass spectrometry data.
• To establish a foundation for field-deployable analytical tools for emergent threat detection.

Main Methods:

• Utilized nominal and high-resolution mass spectrometry data from hundreds of synthetic opioids and non-opioid compounds.
• Trained and validated logistic regression and random forest machine learning models.
• Evaluated model performance based on accuracy, false positive, and false negative rates.

Main Results:

• Random forest models achieved over 95% validation accuracy in classifying opioids versus non-opioids.
• Both nominal and high-resolution mass spectrometry data were effective with random forest.
• The developed random forest models accurately predicted the classification of previously unseen compounds.

Conclusions:

• Random forest models demonstrate high accuracy and low error rates for opioid detection using mass spectrometry.
• ML-driven analysis is crucial for developing practical, field-deployable instruments for identifying emergent chemical threats.
• This approach supports broad-spectrum screening of complex mixtures containing unknown threat agents.