Aarhus University Seal / Aarhus Universitets segl

Kirstine Lykke Nielsen

Assessment of XCMS Optimization Methods with Machine-Learning Performance

Publikation: Bidrag til tidsskrift/Konferencebidrag i tidsskrift /Bidrag til avisTidsskriftartikelForskningpeer review

The metabolomics field is under rapid development. In particular, biomarker identification and pathway analysis are growing, as untargeted metabolomics is usable for discovery research. Frequently, new processing and statistical strategies are proposed to accommodate the increasing demand for robust and standardized data. One such algorithm is XCMS, which processes raw data into integrated peaks. Multiple studies have tried to assess the effect of optimizing XCMS parameters, but it is challenging to quantify the quality of the XCMS output. In this study, we investigate the effect of two automated optimization tools (Autotuner and isotopologue parameter optimization (IPO)) using the prediction power of machine learning as a proxy for the quality of the data set. We show that optimized parameters outperform default XCMS settings and that manually chosen parameters by liquid chromatography-mass spectrometry (LC-MS) experts remain the best. Finally, the machine-learning approach of quality assessment is proposed for future evaluations of newly developed optimization methods because its performance directly measures the retained signal upon preprocessing.

OriginalsprogEngelsk
TidsskriftAnalytical Chemistry
Vol/bind93
Nummer40
Sider (fra-til)13459-13466
Antal sider8
ISSN0003-2700
DOI
StatusUdgivet - okt. 2021

Bibliografisk note

Funding Information:
We are thankful to innovation fund Denmark for funding (TraceAge). P.V. and J.L. is supported by a PhD grant from AUFF NOVA (Aarhus University Research Foundation).

Publisher Copyright:
© 2021 American Chemical Society

Se relationer på Aarhus Universitet Citationsformater

ID: 225936394