Machine Learning for Enhanced Identification Probability in RPLC/HRMS Nontargeted Workflows

Area of Science:

• Analytical Chemistry
• Mass Spectrometry
• Machine Learning

Background:

• High-resolution mass spectrometry (HRMS)-based nontargeted analysis (NTA) relies heavily on spectral matching for chemical identification, especially without retention data.
• Accurate chemical identification is critical for interpreting complex sample compositions in various scientific fields.

Purpose of the Study:

• To develop and validate a novel machine learning (ML)-driven approach to enhance chemical identification probability (IP) in HRMS-based NTA.
• To leverage predicted retention time indices (RTIs) derived from molecular fingerprints and cumulative neutral losses to improve spectral matching accuracy.

Main Methods:

• Developed three ML models: molecular fingerprint (MF)-to-RTI, cumulative neutral loss (CNL)-to-RTI, and a binary classification model for true positive/negative spectral matches.
• Trained models on extensive datasets, including calibrants and experimental spectra, and validated using independent test sets.
• Integrated predicted RTIs with spectral library searches and evaluated performance using metrics like F1 score and Matthews correlation coefficient for pesticide identification.

Main Results:

• High correlation (R²=0.96 training, 0.88 testing) between MF- and CNL-derived RTI values indicated reduced errors in true positive spectral matches.
• The k-nearest neighbors algorithm achieved a weighted F1 score of 0.65 and MCC of 0.30 for pesticide identification (1-1000 ppb) in blank samples.
• Average pesticide IPs increased by 46.7-54.5% compared to library matching alone across different sample dilutions.

Conclusions:

• Machine learning models effectively predict RTIs, reducing identification uncertainties in HRMS-NTA.
• The proposed ML approach significantly boosts the confidence of chemical identifications, particularly for pesticides in complex matrices.
• This methodology offers a powerful tool for advancing chemical discovery and characterization in analytical sciences.