SH3-Hunter

SH3-Hunter predicts potential interactions between SH3 domains and proline-rich short peptide sequences to identify peptide-domain binding sites and infer binding specificity.

Key Features:

• Recognition of Interaction Sites: Identifies peptides containing poly-proline binding motifs and pairs them with SH3 domains to predict potential interaction sites on query proteins.
• Propensity Scoring and Evaluation Metrics: Assigns a propensity score to each predicted peptide-domain interaction and reports sensitivity and precision estimates for prediction reliability.
• Neural Network-Based Prediction Model: Employs neural network models that integrate experimental pep-spot data with structural information from known SH3-peptide complexes.
• Extensive Interaction Database (SH3-specific matrix): Uses a multidimensional matrix of position-specific contact frequencies derived from structural studies and panning experiments on peptide libraries to evaluate binding likelihoods.
• Machine Learning Encoding: Implements a novel encoding based on domain-peptide contact residues to represent interaction information with a reduced number of variables.
• High Accuracy and Generalization Capability: Reports accuracy exceeding 90% for detecting new binders to known SH3 domains and can infer specificity for unknown SH3 domains with structural similarity to those in the database.

Scientific Applications:

• Binding partner discovery in proteomes: Identifies candidate SH3-binding peptides across large protein databases such as SWISSPROT.
• Prediction of binding specificity: Infers peptide binding specificity from SH3 domain sequences and structural similarity.
• Analysis of protein-protein and protein-ligand interactions: Helps elucidate rules governing SH3-mediated interactions and proline-rich motif recognition.
• Identification of interaction sites on query proteins: Predicts location of potential SH3-binding peptides within submitted protein sequences.

Methodology:

Combines position-specific contact frequencies from known SH3/peptide complex structures and pep-spot panning experiments into an SH3-specific matrix, encodes domain-peptide contact residues with a reduced-variable scheme, and applies neural network–based machine-learning models to compute propensity scores and evaluation metrics.