Self Adjusting Algorithm for the Nontargeted Feature Detection of High Resolution Mass Spectrometry Coupled with Liquid Chromatography Profile Data

Abstract

Nontargeted feature detection in data from high resolution mass spectrometry is a challenging task, due to the complex and noisy nature of data sets. Numerous feature detection and preprocessing strategies have been developed in an attempt to tackle this challenge, but recent evidence has indicated limitations in the currently used methods. Recent studies have indicated the limitations of the currently used methods for feature detection of LC-HRMS data.To overcome these limitations, we propose a self-adjusting feature detection (SAFD) algorithm for the processing of profile data from LC-HRMS. SAFD fits a three-dimensional Gaussian into the profile data of a feature, without data preprocessing (i.e., centroiding and/or binning). We tested SAFD on 55 LC-HRMS chromatograms from which 44 were composite wastewater influent samples. Additionally, 51 of 55 samples were spiked with 19 labeled internal standards. We further validated SAFD by comparing its results with those produced via XCMS implemented through MZmine. In terms of ISs and the unknown features, SAFD produced lower rates of false detection (i.e., ≤ 5% and ≤10%, respectively) when compared to XCMS (≤11% and ≤28%, respectively). We also observed higher reproducibility in the feature area generated by SAFD algorithm versus XCMS.

From Centroided to Profile Mode: Machine Learning for Prediction of Peak Width in HRMS Data

“…Moreover, these parameters, when optimized, were in close agreement with the parameters optimized for the self-adjusting feature detection algorithm. 16 …”

“…There are different open-source/access algorithms for processing (e.g., feature detection) of both profile and centroided data during non-targeted workflows. 1 , 2 , 15 Some of these data processing tools are specifically designed to handle the profile data 16 , 17 while others can only process the centroided data. 18 , 19 Most of these algorithms employ a set of user defined (i.e., applied to all peaks) parameters such as mass tolerance.…”

“…These mass tolerances are used as a means to group the signals that belong to the same chemical constituents, for example, all the measured points of a chromatographic feature in both time and mass domains. 16 , 18 This implies that such parameters are dependent on the distribution of the measured m / z values in the mass domain and thus the mass peak width. However, previous studies have shown that such parameter setting strategies may cause a high level of uncertainty in the final outcome, particularly for complex samples with a wide variety of chemicals and concentration levels.…”

Inter-laboratory mass spectrometry dataset based on passive sampling of drinking water for non-target analysis

“…While these algorithms have shown increasing success, there still remain some challenges with NTA and the underlying assumptions. Accordingly, recent studies have highlighted that further improvements are needed to be able to generate reproducible results 8 , 15 , 24 , 25 .…”

Naive Bayes classification model for isotopologue detection in LC-HRMS data

“…Feature lists were generated with the self adjusting feature detection (SAFD) algorithm, using the following settings: 10 000 maximum number of iterations, a minimum intensity of 500, resolution of 20 000, 0.02 m/z minimum window size in the mass domain, 0.75 minimum regression coefficient, a maximum signal increment of 5, a signal to noise ratio of 2, and a minimum and maximum peak width in the time domain of 3 and 200 s, respectively. 12 These feature lists were used for the performance evaluation of the classification model on real samples.…”