I-sites

I-sites predicts initiation sites and local three-dimensional structural elements in protein sequences using a curated library of short sequence patterns that correlate with specific structural motifs.

Key Features:

• Pattern-Based Library: The core library contains 82 distinct sequence patterns derived from documented and newly identified sequence-structure relationships, including a diverging type-II beta-turn, a frayed helix, and a proline-terminated helix.
• Automated Correlation Identification: The library was generated by an automated method that identifies correlations between protein sequences and their local three-dimensional structures.
• Segment Matching and Prediction: I-sites scans query sequences for segments of 7–19 residues that match library patterns and assigns each matching segment the corresponding three-dimensional structure.
• Backbone Torsion Angle Prediction: The tool predicts backbone torsion angles for entire protein sequences by integrating mutually compatible segment predictions.

Scientific Applications:

• Local Structure Prediction: Predicts local structures within proteins, providing information on regions that traditional secondary structure methods may overlook.
• High-Confidence Predictions: In a test set of 55 proteins, approximately 50% of all residues and 76% of residues covered by high-confidence predictions fall within eight-residue segments that are less than 1.4 Å from their true structures.
• Complementary to Traditional Methods: Offers more specific predictions in turn regions, complementing conventional secondary structure prediction approaches.
• Ab Initio Tertiary Structure Prediction and Fold Recognition: Supplies detailed local structural information useful for ab initio tertiary structure prediction and fold recognition tasks.

Methodology:

The library was compiled using an automated method to identify sequence–structure correlations; query sequences are scanned for 7–19 residue segments that match library patterns, each matched segment is assigned the corresponding three-dimensional structure, and mutually compatible segment predictions are integrated to predict backbone torsion angles for entire sequences.