Improving Isotope Ratio Accuracy in Metabolic Labeling Using Orbitrap Mass Spectrometry: A Machine Learning...

Stable isotope labeling is widely used to study metabolic fluxes. Mass spectrometry is the primary tool for measuring isotope ratios in biomolecules, but there are often trade-offs between mass resolution, sensitivity, analytic speed, and, most importantly for flux measurements, accuracy and precision of relative isotope abundances. Orbitrap mass spectrometers have been found to exhibit high isotope ratio measurement accuracy and precision in targeted measurements on a narrow m/z range but are biased in untargeted measurements of multiple biomolecules concurrently across a wide m/z range. This measurement bias is known to be caused by multiple factors, including the ion signal intensity. Here, we developed a scan-by-scan, machine-learning-based correction method to address the bias and predict bias-free mass isotopomer ratios. The fully trained random forest model reduces the mean absolute percentage error of isotopic measurements in detected metabolites for the M1 mass isotopomer from 21.3 to 3.5% and for M2 from 25.8 to 3.1%. The model also provides a ranking of factors by their impact on the measurement bias, and it was found that the ion signal intensity/TIC ratio of individual scans is the most dominant and is a previously neglected bias-causing factor. Use of this correction model improves the accuracy of metabolic flux measurements from heavy water (²H₂O) labeling studies. Our correction method overcomes the limitation that current Orbitrap mass spectrometers cannot achieve isotope measurement accuracy in methods covering a wide m/z range. The approach presented here enables stable isotope labeling experiments to be high-throughput and may advance stable isotope labeling toward untargeted "fluxomics".