Granular decision fusion systems for effective protein methylation pPrediction

Progress and challenges in predicting protein methylation sites

“…54 Therefore physicochemical and biochemical property has been applied to predict protein methylation. 29,30,38 Chemical property matrix, 28 …”

“…Ding et al manually examined publications on PubMed and collected PRMT1, PRMT4 and PRMT5 substrates to predict the methylated arginines. 28,29 In the field of methylation site prediction, the experimentally verified methylation sites are defined as the positive data set. Since only neighboring residues can influence the status of centered arginines or lysines, the sliding window strategy was utilized to extract positive data from protein sequences as training data, which were represented by peptide sequences with methylation sites symmetrically surrounded by flanking residues.…”

“…But the most appropriate window size is still unclear, most researchers have preliminarily tested various different fragment sizes, and then chosen the window size that gives the best predictive performance. 19,[21][22][23][28][29][30][31][32][33][34][35] Fig. 1 The system flow chart of the current methods for predicting protein methylation sites.…”

“…Except Plewczynski et al 37 defined the negative instances that were chosen randomly from those that do not include experimentally verified PTM of any type, almost all researchers of methylation prediction used a strategy in which arginine or lysine that is not annotated as methylated sites in the experimentally validated methylated proteins is defined as a non-methylation site. 19,[21][22][23][24]28,[29][30][31][32][33][34][35]38,39 The rational of this assumption is that as proteins with known methylation sites have been well studied with respect to methylation, sites in these proteins that are not known to be methylated are more likely to be true negative sites than those obtained by random. This strategy has been widely used in predicting other types of PTM, such as phosphorylation, 40 ubiquitination 41 and acetylation, 42 etc.…”

“…But the most appropriate window size is still unclear, most researchers have preliminarily tested various different fragment size, and then chosen the window size that gives the best predictive performance. 19,[21][22][23][28][29][30][31][32][33][34][35] Like phosphorylation site prediction, a rigorous investigation of window size would be invaluable for developers of future methylation site prediction tools, especially if consideration was given to the particular prediction algorithm used and the particular methylated residues and methyltransferase under consideration. 36 Negative data.…”