ResPRE

ResPRE predicts long-range residue–residue contacts in proteins using precision matrices derived from multiple sequence alignments (MSAs) and a deep residual convolutional neural network (ResNet).

Key Features:

• Precision Matrix-Based Features: Uses inverse covariance matrices computed from multiple sequence alignments to capture coevolutionary signals and reduce noise in contact prediction.
• Residual Neural Network Architecture: Implements a deep residual convolutional neural network (ResNet) with shortcut connections to model complex sequence–structure relationships.
• Long-Range Contact Prediction: Predicts residue–residue contacts between distant positions in protein sequences.
• MSA-Dependent Prediction: Utilizes high-quality multiple sequence alignments to improve contact map accuracy.

Scientific Applications:

• Protein Structure Prediction: Supports structural modeling of proteins by predicting long-range residue contacts.
• Protein Function Analysis: Assists in identifying structural constraints relevant to protein function.
• Template-Free Structure Modeling: Facilitates structure prediction for proteins lacking homologous structural templates.

Methodology:

ResPRE computes precision matrices from multiple sequence alignments and uses these features as input to a deep residual convolutional neural network (ResNet) to predict long-range residue–residue contacts.