PSCP

PSCP predicts protein structural classes and ligand binding properties from protein tertiary structures using image-derived features for structural similarity and binding-site inference.

Key Features:

• Image-Based Feature Extraction: Extracts features from the distance matrix of protein tertiary structures including Local Binary Pattern (LBP) histogram, Gabor filtered LBP histogram, Separate Row Multiplication Matrix with uniform LBP histogram, Neighbor Block Subtraction Matrix with uniform LBP histogram, and atom bond features.
• Supervised Machine Learning: Applies supervised algorithms including Random Forest for prediction of structural classes and ligand binding properties, with performance evaluated on benchmark datasets.
• Hybrid LBP: Utilizes Hybrid Local Binary Pattern (Hybrid LBP) as an image-based descriptor that yields high accuracy for protein structural class prediction.
• Protein–Ligand Binding Prediction: Predicts protein–ligand binding sites using the same set of image-based features combined with a Similarity-Based Clustering approach.

Scientific Applications:

• Structural Classification: Classifies proteins into structural classes based on tertiary structure-derived image features to support interpretation of protein function and interactions.
• Ligand Binding Prediction: Predicts potential ligand binding sites to inform virtual screening and drug design efforts.

Methodology:

Image-based features are extracted from the distance matrix (Local Binary Pattern (LBP) histogram, Gabor filtered LBP histogram, Separate Row Multiplication Matrix with uniform LBP histogram, Neighbor Block Subtraction Matrix with uniform LBP histogram, atom bond features, and Hybrid LBP), supervised machine learning algorithms including Random Forest are applied for prediction, and Similarity-Based Clustering is used for ligand binding site prediction.