proprint

proprint predicts protein-protein interactions from amino acid sequences using Support Vector Machine (SVM) models that integrate multiple sequence composition descriptors and Position Specific Scoring Matrix (PSSM) information to support functional interpretation of genomic data.

Key Features:

• SVM-based classification: Uses Support Vector Machine models to distinguish interacting and non-interacting protein pairs from sequence information alone.
• Sequence composition descriptors: Employs amino acid, dipeptide, biochemical property, split amino acid, and pseudo amino acid compositions as input features.
• Evolutionary information: Incorporates Position Specific Scoring Matrix (PSSM) compositions to capture evolutionary profiles.
• Model scope: Develops both species-specific and general models to account for organismal variation in PPIs.
• Performance metrics: Reports Matthews correlation coefficient (MCC) values including 1.00 for Escherichia coli, 0.52 for Saccharomyces cerevisiae, and 0.74 for Helicobacter pylori using a dipeptide-based SVM at the default threshold, and notes a decrease to 0.67 for Escherichia coli on a new dataset.
• Amino acid propensity analysis: Identifies specific amino acids that are more prevalent in interacting pairs than in non-interacting pairs for sequence-based descriptor interpretation.

Scientific Applications:

• Interactome prediction: Predicts protein-protein interactions to help reconstruct organismal interactomes from sequence data.
• Functional interpretation of genomes: Assists in inferring functional relationships among proteins for post-genomic analyses.
• Functional genomics and systems biology: Supports studies of cellular networks and systems-level functional organization by providing PPI predictions and sequence-based insights.

Methodology:

Support Vector Machine (SVM)-based classification using amino acid, dipeptide, biochemical property, split amino acid, and pseudo amino acid compositions plus Position Specific Scoring Matrix (PSSM) compositions, with species-specific and general models trained and evaluated using Matthews correlation coefficient (MCC), including reported dipeptide-based SVM performance at the default threshold.