A first-stage approximation to identify new imprinted genes through sequence analysis of its coding regions

Abstract

In the present study, a positive training set of 30 known human imprinted gene coding regions are compared with a set of 72 randomly sampled human nonimprinted gene coding regions (negative training set) to identify genomic features common to human imprinted genes. The most important feature of the present work is its ability to use multivariate analysis to look at variation, at coding region DNA level, among imprinted and non-imprinted genes. There is a force affecting genomic parameters that appears through the use of the appropriate multivariate methods (principle components analysis (PCA) and quadratic discriminant analysis (QDA)) to analyse quantitative genomic data. We show that variables, such as CG content, [bp]% CpG islands, [bp]% Large Tandem Repeats, and [bp]% Simple Repeats, are able to distinguish coding regions of human imprinted genes.

Figure 1

The separation of the training set into four groups: I1, I2, NO_I1 and NO_I2. Notice that both PCs are responsible for the separation.

Figure 2

Plot of the loading values of the selected variables used in the training set.

Figure 3

Scores for the predicted imprinted genes.