A machine learning approach for predicting methionine oxidation sites

摘要: The oxidation of protein-bound methionine to form sulfoxide, has traditionally been regarded as an oxidative damage. However, recent evidences support the view this reversible reaction a regulatory post-translational modification. perception that sulfoxidation may provide mechanism redox regulation wide range cellular processes, stimulated some proteomic studies. these experimental approaches are expensive and time-consuming. Therefore, computational methods designed predict sites attractive alternative. As first approach matter, we have developed models based on random forests, vector machines neural networks, aimed at accurate prediction oxidation. Starting from published data regarding oxidized methionines, created hand-curated dataset formed by 113 unique polypeptides known structure, containing 975 methionyl residues, 122 which were oxidation-prone (positive dataset) 853 oxidation-resistant (negative dataset). We use machine learning generate predictive datasets. Among multiple features used in classification task, them contributed substantially performance models. Thus, (i) solvent accessible area residue, (ii) number residues between analyzed next found towards N-terminus (iii) spatial distance atom sulfur closest aromatic among most relevant features. Compared other classifiers also evaluated, forests provided best performance, with accuracy, sensitivity specificity 0.7468±0.0567, 0.6817±0.0982 0.7557±0.0721, respectively (mean ± standard deviation). present computationally detect become vivo response signals. These insights into structural context residue either or oxidation-prone. Furthermore, should be useful prioritizing methinonyl for further studies determine their potential modification sites.