SMI-BLAST

SMI-BLAST improves protein remote homology detection by refining Position-Specific Scoring Matrices (PSSMs) to mitigate Incorrectly Selected Homology (ISH) errors.

Key Features:

• Supervised iterative search framework: Applies a supervised search approach to iteratively update BLAST-derived PSSMs to reduce ISH errors.
• ISH error identification: Detects and classifies three distinct types of Incorrectly Selected Homology (ISH) errors in PSSMs.
• PSSM refinement: Corrects identified ISH errors to improve the accuracy of PSSMs used by PSI-BLAST.
• Benchmark performance on SCOPe: Demonstrated detection of 1.6–2.87 times more remote homologs and a 35.66% higher ROC1 score compared to PSI-BLAST on the SCOPe dataset.
• Integration with other search tools: Can be integrated into JackHMMER, DELTA-BLAST, and PSI-BLASTexB to enhance their remote-homology detection performance.

Scientific Applications:

• Remote homology detection: Enhances sensitivity for identifying remote homologous protein sequences.
• Functional and evolutionary analysis: Improves sequence-based inference of protein function and evolutionary relationships using refined PSSMs.
• Improving profile-based search methods: Boosts performance of profile/profile and profile/sequence search tools when integrated into JackHMMER, DELTA-BLAST, or PSI-BLASTexB.

Methodology:

Implements a supervised, iterative BLAST-based search that identifies three types of ISH errors and refines PSSMs, with performance evaluated on the SCOPe dataset and integration demonstrated with JackHMMER, DELTA-BLAST, and PSI-BLASTexB.