Bigram-PGK

Bigram-PGK predicts phosphoglycerylation sites in proteins by using PSSM-derived profile bigrams and a support vector machine to identify phosphoglycerylated lysine residues.

Key Features:

• Evolutionary information utilization: Leverages evolutionary information of amino acids to incorporate conservation patterns across species.
• Profile bigram occurrences: Calculates profile bigram occurrences from position-specific scoring matrices (PSSMs) derived from protein sequences.
• Support vector machine classifier: Employs a support vector machine classifier for binary classification of lysine residues as phosphoglycerylated or not.
• Benchmark dataset utilization: Trains and validates the model using an experimentally labeled benchmark dataset.
• Segment-based feature extraction: Extracts features from sequence segments of ±32 amino acids around candidate lysine sites.
• Cross-validation: Evaluates performance using 10-fold cross-validation with varying train/test sample combinations.
• Performance metrics: Reported metrics include Sensitivity 0.9642, Specificity 0.8973, Precision 0.8253, Accuracy 0.9193, Matthews Correlation Coefficient 0.8330, and AUC 0.9306.

Scientific Applications:

• Phosphoglycerylation site mapping: Predicts phosphoglycerylated lysine residues to support mapping of this post-translational modification.
• Proteomics and cell biology studies: Facilitates investigation of protein regulation and cellular processes linked to phosphoglycerylation.
• Disease and therapeutic research: Assists identification of candidate modification sites relevant to disease mechanisms and potential therapeutic targets.

Methodology:

Feature extraction using ±32 amino acid segments, generation of PSSMs, calculation of profile bigram occurrences, training of a support vector machine on an experimentally labeled benchmark dataset, and evaluation by 10-fold cross-validation with different train/test sample combinations.