PepBindPred

PepBindPred predicts peptide binding regions in protein sequences to identify short linear motifs (SLiMs) and potential interactions with peptide-binding domains.

Key Features:

• Training Dataset: Trained on a curated, non-redundant dataset derived from the ELM database containing protein sequences with annotated peptide binding regions (SLiMs).
• Focus on Disordered Regions: Targets intrinsically disordered regions of proteins that often mediate peptide–domain interactions.
• Vina Docking Scores: Utilizes Vina docking scores computed for overlapping tripeptides within a 50-residue region surrounding each SLiM to evaluate potential binding affinity.
• Predicted Secondary Structure: Incorporates predicted secondary structure of the protein sequence as an input feature.
• Predicted Disorder Score: Includes predicted disorder scores as an input feature.
• Machine Learning Approach: Integrates multiple input features using a bidirectional recurrent neural network (BRNN) to predict peptide binding regions.
• Performance Metrics: Demonstrates an Area Under the Curve (AUC) of 0.72, outperforming Vina docking scores alone.

Scientific Applications:

• Binding site prediction: Predicts potential peptide binding sites within protein sequences that may interact with peptide-binding domains.
• Protein–protein interaction analysis: Supports analysis of protein–protein interactions mediated by short linear motifs.
• Intrinsically disordered protein studies: Aids study of intrinsically disordered proteins by identifying disordered regions likely to bind structured domains.
• Signaling pathway research: Informs investigations of signaling pathways involving SLiM-mediated interactions.

Methodology:

Model trained on an ELM-derived non-redundant dataset using Vina docking scores for overlapping tripeptides within 50-residue SLiM regions, predicted secondary structure and disorder scores as input features, and a bidirectional recurrent neural network (BRNN) for prediction.