MS2Quant

MS2Quant predicts concentrations of unidentified chemicals detected by nontarget liquid chromatography-high-resolution mass spectrometry (LC-HRMS) by estimating ionization efficiencies from fragmentation (MS²) spectra using machine learning.

Key Features:

• Machine Learning Model: xgbTree regression trained on ionization efficiency data derived from 1191 unique chemicals spanning eight orders of magnitude.
• Ionization Efficiency Prediction: Predicts ionization efficiencies from structural fingerprints computed from SMILES for identified chemicals or from MS² spectra for unidentified compounds using SIRIUS+CSI:FingerID.
• Concentration Estimation: Converts predicted ionization efficiencies together with MS response into concentration estimates for unidentified compounds detected in nontarget LC-HRMS.
• Performance Metrics: Reported root mean square errors of 0.55 (3.5×) log-units on the training set and 0.80 (6.3×) log-units on the test set.
• Validation and Comparison: Validated on a set of 39 environmental pollutants with a mean prediction error of 7.4× (geometric mean 4.5×, median 4.0×) and compared to PaDEL descriptor-based models (mean 9.5×; geometric mean 5.6×; median 5.2×).

Scientific Applications:

• Environmental monitoring: Enables quantification of unidentified compounds in nontarget LC-HRMS datasets to support monitoring of environmental pollutants.
• Chemical risk assessment: Provides concentration estimates for unidentified chemicals detected via LC-HRMS to inform risk assessment workflows.

Methodology:

xgbTree regression was trained on ionization efficiency data from 1191 chemicals spanning eight orders of magnitude using structural fingerprints computed from SMILES or predicted from MS² spectra by SIRIUS+CSI:FingerID to predict ionization efficiencies and estimate concentrations from fragmentation spectra.