Mass spectral reconstruction of LC/MS data with entropy minimization

Highlights

• A novel deconvolution approach for LC-MS data.

• Resolve asymmetric, co-eluted and trace peaks in LC-MS data.

• A multivariate curve resolution approach that does not require any parameter settings.

Abstract

Liquid chromatography-mass spectrometry (LC/MS) is a major technique for the analysis of complex samples. Data processing for LC/MS untargeted analysis is very challenging due to the presence of asymmetric, co-eluted at trace-level peaks. We report a multivariate curve resolution approach based on entropy minimization to resolve LC/MS data. Our liquid chromatography band-target entropy minimization (lc-BTEM) approach enables automated resolution of complex LC/MS peaks without the requirement of any parameter inputs. We demonstrated the performance of our approach on a co-eluted and trace-level peaks in an American ginseng extract. Consequently, we were able to putatively annotate the presence of more than 50 ginsenosides. The extracted pure mass spectra and in-source fragments information further enabled the annotation of [M+H]⁺, [M+H-Neu]⁺ and [M-H]^- ions. This approach is expected to benefit LC/MS untargeted analysis and data independent analysis of complex samples.

Introduction

Liquid chromatography-mass spectrometry (LC/MS) is a widely used technique in biochemical and biomedical research [[1], [2], [3], [4], [5]]. The technique offers a combination of sensitivity and selectivity as well as enables the analysis of a wide range of non-volatile compounds. However, LC/MS produces complex data and demands systematic, stringent and innovative data processing steps to extract meaningful information. The major challenges include resolving asymmetric, co-eluted and trace-level peaks. Current solutions to this challenge rely mainly on the optimization of experimental protocols. However, this can be very time-consuming and may not always resolve the problem, especially for complex biological samples.

Spectral deconvolution is a viable solution to resolve co-eluted and trace-level peaks in LC/MS data. Numerous empirical peak shape modeling techniques have been introduced over the years, such as the bi-Gaussian model and exponentially modified Gaussian model for this purpose [[6], [7], [8], [9]]. Other approaches apply feature selection algorithms (e.g., CODA, CAMERA, centWave) and its variants [[10], [11], [12], [13]]. However, a one-size-fit-all modeling solution may not be available since the peak shape and width in LC/MS data can vary from one equipment to another or even across repeated data from the same equipment. The analysis often requires fine adjustments of various parameter settings to ensure accurate deconvolution and is thus not easily accessible to non-experts.

In recent years, information entropy minimization approach has been implemented as a deconvolution technique [14,15]. As a blind source separation algorithm, the family of band-target entropy minimization (BTEM) algorithms can extract pure signals from mixture signals by finding the linear combination of pure components and without the need for any a priori information [[16], [17], [18], [19]]. In comparison to empirical peak shape modeling techniques, BTEM approach does not require any parameter settings and is robust in handling different forms of peak shape and width. Recently, rBTEM was introduced for the deconvolution of GC/MS data and the result showed that it was able to increase the sensitivity and accuracy for analysis of co-eluting and trace-level components in comparison to empirical-based approach [20].

In this work, we extend the information entropy minimization approach to deconvolute co-eluted and trace-level peaks in LC/MS scan data. The rBTEM variant herein, known as lc-BTEM, considers the higher resolution of data generated by a triple quadrupole system to enable the extraction of pure mass spectra without the need for any parameter settings. We demonstrate its application on LC/MS scan data of American ginseng (Panax quinquefolius) extract. The scan data contains valuable information related to molecular ions and their fragment ions resulting from in-source collisional activation dissociation at the ESI interface, which enables the application of the lc-BTEM algorithm which is based on finding the linear combination of pure components in a mixture system. The availability of pure mass spectra and inherent in-source fragments information enabled a seamless recognition of [M+H]⁺, [M+H-Neu]⁺ and [M-H]^- ions and subsequently led to the putative annotation of ginsenoside subtypes for more than 50 ginsenosides.