CAMP

CAMP predicts peptide-protein interactions (PepPIs) and residue-level peptide binding sites from sequence using a convolutional attention-based neural network to support peptide therapeutic design and analysis.

Key Features:

• Multifaceted prediction: Predicts binary peptide-protein interactions and identifies specific peptide binding residues involved in those interactions.
• Sequence-based model: Uses a convolutional neural network (CNN) architecture combined with an attention mechanism to extract and weight informative sequence signals without requiring high-resolution structural data.
• Comprehensive sequence features: Integrates secondary structures, physicochemical properties, intrinsic disorder characteristics, and position-specific scoring matrices (PSSMs) as input features.
• Benchmark dataset: Includes a curated benchmark dataset of high-quality peptide-protein interaction pairs with corresponding annotated binding residues for training and evaluation.
• Performance: Demonstrates performance that outperforms existing state-of-the-art methods for binary peptide-protein interaction prediction and accurately identifies non-covalent binding residues.

Scientific Applications:

• Peptide therapeutic design: Supports design and prioritization of peptide candidates by predicting interaction partners and binding residues relevant to efficacy.
• Binding-site mapping and engineering: Enables identification of peptide residues for mutagenesis or engineering to alter binding specificity or affinity.
• Functional studies of PepPIs: Facilitates investigation of cellular processes mediated by peptide-protein interactions through prediction of interaction networks and contact sites.

Methodology:

Implements a convolutional attention-based neural network that integrates sequence-derived features (secondary structure, physicochemical properties, intrinsic disorder, PSSMs) via CNN layers and an attention mechanism, and is trained/evaluated on a benchmark dataset of peptide-protein pairs with annotated binding residues to predict binary interactions and residue-level binding sites.