Computational Prediction of Lysine Pupylation Sites in Prokaryotic Proteins Using Position Specific Scoring Matrix into Bigram for Feature Extraction

Abstract

Post-transcriptional modification (PTM) in a form of covalently attached proteins like ubiquitin (Ub) are considered an exclusive feature of eukaryotic organisms. Pupylation, a crucial type of PTM of prokaryotic proteins, is modification of lysine residues with a prokaryotic ubiquitin-like protein (Pup) tagging functionally to ubiquitination used by certain bacteria in order to target proteins for proteasomal degradation. Pupylation plays an important role in regulating many biological processes and accurate identification of pupylation sites contributes in understanding the molecular mechanism of pupylation. The experimental technique used in identification of pupylated lysine residues is still a costly and time-consuming process. Thus, several computational predictors have been developed based on protein sequence information to tackle this crucial issue. However, the performance of these predictors are still unsatisfactory. In this work, we propose a new predictor, PSSM-PUP that uses evolutionary information of amino acids to predict pupylated lysine residues. Each lysine residue is defined through its profile bigrams extracted from position specific scoring matrices (PSSM). PSSM-PUP has demonstrated improvement in performance compared to other existing predictors using the benchmark dataset from Pupdb Database. The proposed method achieves highest performance in 10-fold PSSM-PUP with accuracy value of 0.8975, sensitivity value of 0.8731, specificity value of 0.9222, precision value of 0.9222 and Matthews correlation coefficient value of 0.801.

PupStruct: Prediction of Pupylated Lysine Residues Using Structural Properties of Amino Acids

“…The support vector machine (SVM) [76,77] is engaged in both regression and classification applications and is also used in many state-of-the-art predictors for pupylation sites [35][36][37][38][39]41,43]. The literature shows that the SVM produces a lower prediction error compared to other classifiers when large numbers of features are considered, as in this study there are 104 features.…”

“…The progress and challenges faced in protein pupylation sites prediction were discussed in [20]. CIPPN [42] was developed using a neural network and, most recently, PSSM-PUP [43] employed PSSM, which was converted into bigram probabilities for feature extraction with an LibSVM classifier was developed.…”