Missing value imputation and data cleaning in untargeted food chemical safety assessment by lc-hrms

Grégoire Delaporte (1) , Mathieu Cladière (1) , Valérie Camel (1)

1 GENIAL - Ingénierie, Procédés, Aliments

Untargeted food safety assessment by the use of LC-HRMS instrumentation combined to chemometric tools is a rather new field. As a consequence there is a lack of methodological assessment of the different steps of the data treatment workflow. Thus, we propose a comparison of different methods applied to two major steps of data matrix pretreatment, namely missing value imputation and ion selection. To that end, a missing value classification method has been proposed for the first time for MS data. Several metrics have also been proposed to assess pretreatment step performance as well as to investigate global untargeted approach efficiency for all method combinations considered. Different contaminants were considered as "tracers" to address their detection rates. Pretreatment methods were applied here on two data sets, aiming at illustrating either a simple contamination case to detect or a more complicated application. The data sets used in this study were from the EML-EBI Metabolights data exchange platform (MTBLS752 and MTBLS754), offering other research groups the opportunity to develop and compare their own data treatment strategies with the combinations discussed in this work.

Mots clés: non-targeted; food contaminants; variable selection; filtration

Domaines: Sciences du Vivant [q-bio]

Grégoire Delaporte, Mathieu Cladière, Valérie Camel. Missing value imputation and data cleaning in untargeted food chemical safety assessment by lc-hrms. Chemometrics and Intelligent Laboratory Systems, 2019, 188, pp.54-62. ⟨10.1016/j.chemolab.2019.03.005⟩. ⟨hal-02154537⟩