ModLink+

ModLink+ predicts protein structural folds and refines family assignments by integrating sequence similarity and protein-protein interaction data.

Key Features:

• Integration of Sequence Similarity and Protein Interactions: ModLink+ leverages sequence homology and known protein-protein interactions to inform fold prediction by considering interacting partners and potential homologs.
• Improved Accuracy and Coverage: The method achieves a positive predictive value (PPV) of over 90% at an e-value cutoff of 10^-3, outperforming PSI-BLAST, HHSearch, and PRC in PPV.
• Enhanced Protein Coverage: ModLink+ assigns folds to 30-45% of proteins in tested datasets versus less than 25% for its predecessor and assigns domains to 3738 yeast proteins compared with fewer assignments by PSI-BLAST, HHSearch, and PRC.
• Benchmarked Performance: The approach was benchmarked on 1,434 proteins with known folds from the Structural Classification of Proteins (SCOP) and interacting partners from the Database of Interacting Proteins (DIP).
• Increased Specificity and Sensitivity: Fold assignment specificity increases from 54% to 75% and family assignment specificity at an e-value threshold of 10^-8 increases from 70% to 87%, with a slight increase in sensitivity.

Scientific Applications:

• Predicting Protein Structures: Assigns folds to uncharacterized proteins to support inference of structure and interaction context.
• Functional Annotation: Infers functions of unknown proteins by linking them to characterized proteins through shared interactions.
• Large-Scale Automated Discovery: Enables automated detection of remote relationships between protein sequences using combined sequence and interaction data for genomic-scale analyses.

Methodology:

ModLink+ combines sequence-based predictions with protein-protein interaction data by analyzing proteins in the context of their interaction networks to identify homologous sequences enriched by shared partners; validation used SCOP for structural classification and DIP for interaction data.