Prediction of O-linked glycosylation sites in amino acid sequences by LDA-Normal Bayes approach

O-linked glycosylation is one of the important types of the mammalian protein glycosylation and is known to be serine or threonine specific. In this work, the author propose a new method of LDA (using the concept of Normal Bayes classification) to predict O-glycosylation sites in amino acid sequences under two different window sizes, 11 and 21. The protein sequences are encoded using hydropathy encoding technique. The prediction can be viewed as a 2-class (positive and negative) classification problem. The classification algorithm is designed by clubbing the concepts of Linear Discriminant Analysis and Normal Bayes classification. By using these techniques, the data sets are transformed into discriminating functions. The protein sequence to be predicted is assigned to the class which has the maximum value for the discriminating function. The prediction accuracy for non-O-linked glycosylation sites (negative sites) is about 76-94%, and the accuracy for O-linked glycosylation sites (positive sites) is about 90-98%.