RedHom

RedHom predicts interatomic Cα distances within protein structures using two complementary data-driven methods to derive structural constraints from sequence information.

Key Features:

• Dual Prediction Methods: Two independent approaches are provided: statistically derived probability distributions for estimating pairwise residue distances and a neural network using contextual windows that consider surrounding residues.
• Threshold-Based Sequence Similarity: Implements a defined threshold that determines when sequence similarity implies structural similarity for use in distance prediction.
• Performance Optimization: Evaluates multiple distance thresholds to identify information-rich constraints and reports that neural networks outperform probability density functions for distance prediction.
• Enhanced Accuracy with Sequence Profiles: Incorporates sequence profiles to improve the accuracy of distance predictions.
• Threading Methodology: Provides a threading approach that uses predicted distances to assist protein structure modeling.

Scientific Applications:

• Protein Structure Prediction: Predicts interatomic Cα distances to support modeling of protein structures from sequence data.
• Structural Bioinformatics Research: Derives structural constraints from sequence information for studies of protein folding and function.
• Comparative Structural Analysis: Facilitates comparison of predicted structures with known templates to inform evolutionary and functional analyses.

Methodology:

Analyzes datasets derived using a sequence-similarity-to-structure-similarity threshold; predicts interatomic Cα distances via statistically derived probability distributions and a neural network with contextual windows; evaluates and optimizes distance thresholds, incorporates sequence profiles, and applies a threading method based on predicted distances.