FoldRec-C2C

FoldRec-C2C performs protein fold recognition by integrating hierarchical homology relationships through a global protein similarity network to improve structural predictions, including applications to proteins related to COVID-19.

Key Features:

• Information Retrieval Framework: Treats protein fold recognition as an information retrieval task analogous to natural language processing.
• Learning to Rank: Generates preliminary ranking results from protein pairwise similarities using the Learning to Rank algorithm.
• Re-ranking Models: Applies three distinct re-ranking algorithms—sequence-to-sequence (seq-to-seq), sequence-to-cluster (seq-to-cluster), and cluster-to-cluster (C2C)—to refine initial rankings.
• Cluster-to-Cluster (C2C) Model: Incorporates interactions among protein clusters to inform and refine fold predictions.
• Global Protein Similarity Network: Globally adjusts ranking results using a comprehensive protein similarity network to capture hierarchical homology relationships beyond pairwise similarities.
• Benchmark Validation: Validated on the LINDAHL dataset and shown to outperform 34 state-of-the-art protein fold recognition methods.

Scientific Applications:

• COVID-19-related protein structure prediction: Predicts protein folds for proteins associated with COVID-19.
• Structural biology and homology inference: Improves fold assignment and comparative structural analysis by capturing hierarchical homology relationships.
• Method benchmarking: Enables comparative evaluation of fold recognition methods using the LINDAHL dataset.

Methodology:

Treats fold recognition as an information retrieval task, applies Learning to Rank on protein pairwise similarities to produce initial rankings, and refines these rankings with three re-ranking algorithms (seq-to-seq, seq-to-cluster, and cluster-to-cluster) while globally adjusting results via a global protein similarity network to incorporate hierarchical homology and cluster interactions.