Computational Drug Discovery and Design

The preprocessed data should be stored in an object as soon as

the background correction, normalization, probe-specific correc-

tion, and summarization methods have been performed. Then, the

log intensity expression values at probe set level are extracted

(Fig.7).

3.3 DE Genes
Determination

The gene expression ratios between healthy and T2DM partici-

pants were determined bylimmapackage by fitting a linear model

on each gene with the Bayes’ approach. The “group-means”

parametrization is applied, and the each coefficient for the mean

gene expression within a specific group is estimated (Fig.8).

Then, the parameters of the linear model are determined. The

contrasts of interest are extracted from the coefficient vectors. The

estimated contrast effects and standard errors are estimated, and a

hierarchical Bayes’ model is established. The log fold change

(FC) and adjustedP value for the probe set in each array are

obtained by setting the parameterlfc¼ 0 ,p.value¼ 1.

As a result, the log FC and adjustedPvalue are provided for

each probe set in csv format. We annotate each probe set by the

hgu133a.dbpackage according to the array platform used in the

original literature (Fig.9). Before annotation, we open the gen-

eExp file, and name the probe set column ID. The probe sets that

cannot be annotated or duplicatively annotated should be omitted.

The genes with adjustedPvalues less than 0.05 and log FCs

larger than 1 or less than
1 were considered DE genes. There are

701 DE genes identified according to these criteria. We copy the

Fig. 7Normalization of the intensities indicating gene expression levels. (a) R code for extracting the
normalized probe-set level expression. (b) A normalized log intensity expression value matrix at probe set level

188 Sze Chung Yuen et al.