Improved Accuracy and Reliability in Untargeted Analysis with LC-ESI-QTOF/MS¹ by Ensemble Averaging

Untargeted liquid chromatography coupled with high-resolution mass spectrometry (LC-HRMS) is a powerful tool for comprehensive chemical analysis. Such techniques allow the detection and quantification of thousands of compounds in a sample. However, the complexity and variability in the data can introduce significant errors, impacting the reliability of the results. This study investigates ensemble averaging to mitigate these errors and improve signal-to-noise (S/N) ratios, feature detection, and data quality. In this work, 256 LC-qTOF/MS¹ data sets from the analysis of Morning Glory seeds were averaged to generate merged data sets. The numbers of the pooled data sets in the merged files were varied, and the number of features, the S/N ratio, the accuracy and precision of the accurate masses, relative intensities, and migration time were examined. It was proved that ensemble averaging allows an increase in the S/N up to a factor of 10, and the relative standard deviation of the accurate masses and retention time decreased by a factor of 10. Moreover, the average number of features mined per data set increased from 1192 ± 129 with the original data set to 4408 when all data sets were averaged into one. Using known target compounds, ensemble averaging benefits on quantitative analysis were investigated. The measured and theoretical relative intensities between the [M+1]+H⁺, [M+2]+H⁺, and [M+3]+H⁺ and [M]+H⁺ isotopes of known alkaloids were used. The standard deviation decreased by up to a factor of 10, and the absolute error between theoretical and experimental relative intensities was below 3%, making the theoretical isotopic pattern a valid criterion for confirming a putative molecular formula. Using a targeted approach to recover quantitative data from the original data sets from information in the merged data sets provides an accurate quantitative means. Peak lists from the merged data sets and quantitative information from the original data sets were fused to obtain a robust clustering approach that allows recognizing features (adducts, isotopes, and fragments) generated by a common chemical in the ionization chamber. Two hundred and four clusters were obtained, characterized by two or more features with migration times that differ by less than 0.05 min and with similar response patterns.