Metabolomics Data Analysis (1/24/2015)

Sat, Jan 24, 2015 at 9:16 AM

Customer: We have analyzed the following using CE-TOF-MS: (1) 50 renal cancers and matched normals of mostly 2 histological subtypes: Clear cell and papillary cancer. We wish to identify metabolites with significantly different levels between normal and tumor and also between these two subtypes. (2) Blood plasma from 10 controls and 6 affected (carrying a mutation) individuals all from the same family – ideally looking to identify significantly different metabolites between the two groups and therefore potential biomarkers.

Sat, Jan 24, 2015 at 10:19 AM

AccuraScience LB: To identify individual metabolites that are present at significantly different levels across groups is straight-forward. Common univariate analysis methods such as t-test, ANOVA, followed by multiple testing control (p-value adjustment or false discovery rate (FDR) control) would suffice. I suspect, however, some multivariate analysis methods could be useful for your data. For example, principal component analysis (PCA) and/or independent component analysis (ICA) for purposes of visualizing how samples are organized next to each other in a 2-dimensional space, clustering analysis to examine how samples are organized within and between groups (in this case, "normal" and "tumor" can be considered as two groups, and the two subtypes can be considered as two separate groups too), supervised methods such as partial least square discriminant analysis (PLS-DA) models for predicting the group label ("normal", "tumor", and either of the subtypes) of a future sample. A class of feature selection techniques could be applied to extract a subset of metabolites most suitable for classifying the samples into the groups (which is what you referred to as biomarker identification - but these methods do do not necessarily focus on individual metabolites one at a time, rather, they examine the combined discrimination power of multiple metabolites at once, thus would be more powerful).