ProtFold-DFG

ProtFold-DFG implements a Directed Fusion Graph (DFG) framework that integrates ranking lists from DeepSVM-fold(CCM), DeepSVM-fold(PSFM), MotifCNN-fold(CCM), and MotifCNN-fold(PSFM) and uses transitive closure, Kullback-Leibler (KL) divergence, and the PageRank algorithm to perform protein fold recognition.

Key Features:

• Integration of Multiple Methods: Integrates ranking lists generated by DeepSVM-fold(CCM), DeepSVM-fold(PSFM), MotifCNN-fold(CCM), and MotifCNN-fold(PSFM) into a fused representation.
• Directed Fusion Graph (DFG): Constructs a Directed Fusion Graph and applies transitive closure to capture extended relationships among proteins.
• Relationship Quantification: Employs Kullback-Leibler (KL) divergence to quantify pairwise relationships between proteins.
• Global Interaction Consideration: Applies the PageRank algorithm on the DFG to incorporate global network interactions into final rankings.

Scientific Applications:

• Structural Genomics: Supports protein fold prediction for structural genomics and large-scale structural annotation projects.
• Drug Discovery: Aids inference of target protein folds relevant to structure-based drug discovery.
• Functional Annotation: Assists functional annotation of proteins through fold-based inference.
• Benchmarking and Evaluation: Validated on the LINDAHL dataset and reported to outperform 35 competing methods.

Methodology:

Combine outputs from predictive models into ranking lists, construct a Directed Fusion Graph using transitive closure, compute pairwise relationships with Kullback-Leibler (KL) divergence, and apply the PageRank algorithm on the DFG to produce final fold rankings.