Interpretable machine learning model to detect chemically adulterated urine samples analyzed by high resolution mass...

Area of Science:

• Analytical Chemistry
• Forensic Science
• Biomarker Discovery

Background:

• Urine sample manipulation poses a significant challenge in drug testing, abstinence monitoring, and doping control.
• Simultaneous detection of sample manipulation and drugs in a single analysis is highly desirable for efficiency and reliability.
• Machine learning (ML) offers a powerful approach to identify complex patterns in analytical data.

Purpose of the Study:

• To develop and validate a machine learning model for the simultaneous detection of chemical urine manipulation and prohibited drugs.
• To integrate sample integrity testing into routine drug analysis workflows.
• To identify potential biomarkers indicative of urine adulteration.

Main Methods:

• Utilized liquid chromatography coupled to quadrupole time-of-flight mass spectrometry (LC-QTOF-MS) to analyze 702 authentic human urine samples, including chemically treated and control samples.
• Employed an artificial neural network (ANN) trained on retention time-aligned LC-MS data (33,448 features) from 500 samples.
• Applied local interpretable model-agnostic explanations (LIME) for feature importance analysis.

Main Results:

• The trained ANN achieved high performance metrics: 88.9% sensitivity, 92.0% specificity, 91.9% positive predictive value, and 89.2% negative predictive value after 10-fold cross-validation.
• A diverse test set of 202 samples was correctly classified with an overall accuracy of 95.4%.
• Identified 14 important features and four potential biomarkers associated with urine manipulation.

Conclusions:

• Established a reliable ML model using interpretable LC-MS data for rapid detection of chemical urine manipulation.
• The model enables simultaneous LC-MS analysis and sample integrity testing in a single analytical run.
• This approach has the potential to revolutionize drug testing by enhancing efficiency and accuracy in clinical and forensic settings.