PSI-Search: iterative HOE-reduced profile SSEARCH searching

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Sequence Analysis

Background:

• Iterative similarity searches using position-specific score matrices (PSSMs) like PSI-BLAST enhance homolog detection.
• However, PSSMs can be contaminated by homologous over-extension (HOE), where alignments extend into unrelated protein domains, reducing accuracy.

Purpose of the Study:

• To develop a novel method, PSI-Search, that improves the accuracy and selectivity of iterative protein homology searches.
• To mitigate the impact of homologous over-extension (HOE) on PSSM-based searches.

Main Methods:

• PSI-Search integrates the SSEARCH optimal Smith-Waterman local alignment algorithm with the PSI-BLAST profile construction strategy.
• An optional sequence boundary-masking procedure is implemented to prevent alignment extension and reduce HOE errors in PSSMs.

Main Results:

• PSI-Search demonstrates approximately 4-fold better selectivity compared to PSI-BLAST, with similar sensitivity at 50% and 60% family coverage.
• The method also yields 2- to 4-fold fewer false-positives than JackHMMER, albeit with a slight decrease in sensitivity (~5%).

Conclusions:

• PSI-Search offers a significant improvement in selectivity for protein homology detection by effectively reducing HOE errors.
• This method provides a more accurate and reliable tool for identifying protein homologs compared to existing iterative search strategies.