Attribute selection and model evaluation for the maternal and paternal imprinted genes in bovine (Bos Taurus) using supervised machine learning algorithms.

Imprinted genes display biased expression of paternal and maternal alleles in mammals. They are marked through epigenetic process during gametogenesis. Characterization of imprinted genes has expanded our understanding of the regulation and function of genes. In the current study, 22 experimentally validated imprinted genes in bovine (Bos Taurus) were analysed. Several supervised machine learning algorithms and attribute weighting methods were used to find characteristics of different types of imprinted genes and suggest a classification method for finding maternally and paternally expressed genes in bovine. For assessing the best model and comparing attributes in other organisms, we have also conducted a comparative analysis for human and sheep imprinted genes. According to the results of the present study, GC contents 10 and 100 kb upstream, Gly and Gln amino acids, Ile/ATC codon usage, LINE and SINE in 100kbup and length of first intron were significantly different between the maternal and paternal genes in cattle. Considering all species together, we found that GC content 100 kb up, LINE 100 kb up and the frequency of amino acids like Gly, Gln and Met were the most important attributes for identifying the paternal and maternal imprinted genes. These findings could imply conservation pattern in the attributes among these species. [ABSTRACT FROM AUTHOR]