Multivariate analysis of chromatographic retention data as a supplementary means for grouping structurally related compounds

• PCA and DA classify chemically related compounds based on LC retention data.
• The minimal number of LC–MS analyses necessary for a proper classification is determined.
• The results may help to filter false positive annotations in analyte identification from LC–MS data.

In the present study a series of 45 metabolite standards belonging to four chemically similar metabolite classes (sugars, amino acids, nucleosides and nucleobases, and amines) was subjected to LC analysis on three HILIC columns under 21 different gradient conditions with the aim to explore whether the retention properties of these analytes are determined from the chemical group they belong. Two multivariate techniques, principal component analysis (PCA) and discriminant analysis (DA), were used for statistical evaluation of the chromatographic data and extraction similarities between chemically related compounds. The total variance explained by the first two principal components of PCA was found to be about 98%, whereas both statistical analyses indicated that all analytes are successfully grouped in four clusters of chemical structure based on the retention obtained in four or at least three chromatographic runs, which, however should be performed on two different HILIC columns. Moreover, leave-one-out cross-validation of the above retention data set showed that the chemical group in which an analyte belongs can be 95.6% correctly predicted when the analyte is subjected to LC analysis under the same four or three experimental conditions as the all set of analytes was run beforehand. That, in turn, may assist with disambiguation of analyte identification in complex biological extracts.