Enhancing compound confidence in suspect and non-target screening through machine learning-based retention time...

Area of Science:

• Analytical Chemistry
• Environmental Science
• Computational Chemistry

Background:

• Retention time (RT) is critical for identifying contaminants of emerging concern (CECs) using liquid chromatography-high-resolution mass spectrometry (LC-HRMS) in non-targeted screening (NTS).
• Accurate RT prediction aids in database matching and enhances the reliability of NTS analysis for CECs.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting the RTs of CECs in NTS analysis.
• To improve the confidence and efficiency of CEC identification in complex environmental samples.

Main Methods:

• Evaluated multiple ML models including Random Forest, XGBoost, Support Vector Regression (SVR), and Artificial Neural Network (ANN).
• Utilized molecular fingerprints and chemical descriptors for model training with 1051 CEC standards.
• Validated the optimal SVR model through laboratory NTS compound characterization and application to wastewater treatment plant samples.

Main Results:

• The SVR model using chemical descriptors achieved high predictive accuracy (R²ext = 0.850, r² = 0.925).
• Successfully identified 40 S1 and 234 S2 level CECs in wastewater samples.
• Predicted RTs for S2 compounds, classifying 153 with high confidence (ΔRT < 2 min).

Conclusions:

• The developed ML model significantly enhances NTS analysis by providing robust RT predictions.
• This workflow improves the determination of compound confidence levels, aiding in the identification of CECs.
• RT prediction models are valuable tools for advancing analytical capabilities in environmental monitoring.