SOHSite: incorporating evolutionary information and physicochemical properties to identify protein S-sulfenylation sites

Abstract

Background: Protein S-sulfenylation is a type of post-translational modification (PTM) involving the covalent binding of a hydroxyl group to the thiol of a cysteine amino acid. Recent evidence has shown the importance of S-sulfenylation in various biological processes, including transcriptional regulation, apoptosis and cytokine signaling. Determining the specific sites of S-sulfenylation is fundamental to understanding the structures and functions of S-sulfenylated proteins. However, the current lack of reliable tools often limits researchers to use expensive and time-consuming laboratory techniques for the identification of S-sulfenylation sites. Thus, we were motivated to develop a bioinformatics method for investigating S-sulfenylation sites based on amino acid compositions and physicochemical properties.

Results: In this work, physicochemical properties were utilized not only to identify S-sulfenylation sites from 1,096 experimentally verified S-sulfenylated proteins, but also to compare the effectiveness of prediction with other characteristics such as amino acid composition (AAC), amino acid pair composition (AAPC), solvent-accessible surface area (ASA), amino acid substitution matrix (BLOSUM62), position-specific scoring matrix (PSSM), and positional weighted matrix (PWM). Various prediction models were built using support vector machine (SVM) and evaluated by five-fold cross-validation. The model constructed from hybrid features, including PSSM and physicochemical properties, yielded the best performance with sensitivity, specificity, accuracy and MCC measurements of 0.746, 0.737, 0.738 and 0.337, respectively. The selected model also provided a promising accuracy (0.693) on an independent testing dataset. Additionally, we employed TwoSampleLogo to help discover the difference of amino acid composition among S-sulfenylation, S-glutathionylation and S-nitrosylation sites.

Conclusion: This work proposed a computational method to explore informative features and functions for protein S-sulfenylation. Evaluation by five-fold cross validation indicated that the selected features were effective in the identification of S-sulfenylation sites. Moreover, the independent testing results demonstrated that the proposed method could provide a feasible means for conducting preliminary analyses of protein S-sulfenylation. We also anticipate that the uncovered differences in amino acid composition may facilitate future studies of the extensive crosstalk among S-sulfenylation, S-glutathionylation and S-nitrosylation.

Fig. 1

Analytical flowchart of SOHSite including data collection and preprocessing, features extraction and encoding, model construction and evaluation, and independent testing

Fig. 2

Amino acid composition of protein S-sulfenylation sites. a Comparison of amino acid composition between S-sulfenylation sites (blue) and non- S-sulfenylation sites (red). b Position-specific amino acid composition of S-sulfenylation sites. c Position-specific amino acid composition of non-S-sulfenylation sites. d TwoSampleLogo between S-sulfenylation sites (positive data) and non-S-sulfenylation sites (negative data)

Fig. 3

Comparison of the solvent-accessible surface area between S-sulfenylation and non-S-sulfenylation sites

Fig. 4

The predictive performance of PSSM model combined with forward selection of the top 20 physicochemical properties

Fig. 5

Comparison of the independent testing results between PSSM model and the hybrid model combining PSSM with the top 12 physicochemical properties

Fig. 6

Discrimination of S-sulfenylation sites from S-nitrosylation and S-glutathionylation sites. a Number of duplicate proteins among S-sulfenylation, S-nitrosylation and S-glutathionylation; (b) Number of duplicate sites among S-sulfenylation, S-nitrosylation and S-glutathionylation; (c) Significant differences in position-specific compositions among three PTMs as identified by TwoSampleLogo