unknown metabolites in multi-experiments. While for targeted
metabolomics, it is a hypothesis-driven research and focuses on
absolute quantification of targeted metabolites involved in some
interested pathways. Appropriate sample preparation is required for
accurate quantification of target metabolites. The strength of this
strategy is high sensitivity and specificity; thus numerous metabo-
lites of low abundance in biological samples might be quantified.
However, an obvious weakness of this strategy is its narrow cover-
age of metabolite detection. According to characteristics of untar-
geted and targeted strategy, it is a powerful strategy to combine the
initially untargeted metabolomics for exploration with the subse-
quently targeted metabolomics for validation.
The experimental subject is an important factor that could
influence results of metabolomics. For example, Sprague-Dawley
(SD) and Wistar rats are experimental animals commonly inter-
changeable in practice. However, a metabolomics study revealed
that SD rats have higher individual metabolic variations than Wistar
rats [7], suggesting that the difference of subjects should be con-
sidered thoroughly when designing metabolomics experiment.
Biological bias, existed in all organisms substantially, greatly
affects the results of metabolomics study. Irrelevant biological bias
induced by some impact factors, such as gender, dose, and time,
should be eliminated by biological replicates (repetitive samples
obtained from different individuals). Thus one important aspect
of experimental design is to determine how many biological repli-
cates are required in statistics. In metabolomics study, biological
replicates are preferred over analytical replicates (repetitive analyses
of the same sample obtained from the same individual), since
biological bias almost exceeds analytical one. In practice, three
biological replicates are required and five replicates are preferred
[8]. Moreover, dozens of biological replicates are required for
model animals, while for human population investigation, a larger
scale of samples (usually hundreds to thousands samples) are nec-
essary for early diagnosis in disease [9].
For some metabolomics studies, small treatments and effects
can easily be overlaid by large variation between human subjects,
and thus conventional data analytical strategy can’t work effectively.
A solution for these studies is the case-control analysis based on
cross-over experiment design, where each subject in the study
population acts as case specimen and the other subjects act as
control in turn [10]. In addition, multivariate analysis methods
for paired data remarkably improve the statistical power and give
more reasonable interpretability to results of these metabolomics
studies [11–13]. Therefore, the data analytical strategy should be
chosen thoroughly according to the research goals.268 Jing Cheng et al.