Multidimensional scaling method for prediction of lysine glycation sites.

Similar to the regular enzymatic glycosylation, lysine glycation also attaches a sugar molecule to a peptide, but it does not need the help of an enzyme. It has been found that lysine glycation is involved in various biological processes and is closely associated with many metabolic diseases. Thus, an accurate identification of lysine glycation sites is important to understand its underlying molecular mechanisms. The glycated residues do not show significant patterns, which make both in silico sequence-level predictions and experimental validations a major challenge. In this study, a novel predictor named MDS_GlySitePred is proposed to predict lysine glycation sites by using multidimensional scaling method (MDS) and support vector machine algorithm. As illustrated by the average results of tenfold cross-validation repeated 50 times, MDS_GlySitePred achieves a satisfactory performance with a sensitivity of 95.08%, a specificity of 97.65%, an accuracy of 96.58%, and Matthew's correlation coefficient of 0.93 on the extensively used benchmark datasets. Experimental results indicate that MDS_GlySitePred significantly outperforms four existing glycation site predictors including NetGlycate, PreGly, Gly-PseAAC, and BPB_GlySite. Therefore, MDS_GlySitePred can be a useful bioinformatics tool for the identification of glycation sites. [ABSTRACT FROM AUTHOR]