S2L-PSIBLAST

S2L-PSIBLAST implements a supervised two-layer search framework to improve protein remote homology detection by enhancing identification of distant homologous proteins.

Key Features:

• First-Level Search (SMI-BLAST + double-link strategy): Utilizes the SMI-BLAST framework combined with a double-link strategy to filter out non-homologous protein sequences.
• Second-Level Search (profile-link similarity): Detects additional homologous proteins by leveraging profile-link similarity to refine remote homology detection.
• Learning-to-Rank Strategy: Implements a learning-to-rank approach to produce more accurate ranking lists of detected protein sequences.
• PSI-BLAST foundation: Builds upon the PSI-BLAST method and its profile-based search framework.

Scientific Applications:

• Remote homology detection: Identification of distant homologous proteins for evolutionary and functional inference.
• Protein evolutionary relationship analysis: Analysis of evolutionary relationships among proteins using improved sensitivity to distant homologs.
• Method benchmarking and comparative evaluation: Comparative performance assessment against PSI-BLAST, DELTA-BLAST, and PSI-BLASTexB using the Structural Classification of Proteins-extended benchmark dataset.

Methodology:

The approach applies a first-level SMI-BLAST search with a double-link filtering strategy, a second-level profile-link similarity search to detect additional homologs, and a learning-to-rank algorithm to reorder results; experimental evaluations used the updated Structural Classification of Proteins-extended benchmark dataset.