Comparative accuracy of methods for protein sequence similarity search

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein sequence database searching is crucial for understanding protein structure and function.
• Two novel methods, Probabilistic Smith-Waterman (PSW) and WU-BLAST2, utilize Hidden Markov Models and Sum statistics for gapped alignments, respectively.

Purpose of the Study:

• To compare the effectiveness of PSW and WU-BLAST2 against established methods: Smith-Waterman (SSEARCH), FASTA, and BLASTP.
• To evaluate search accuracy using statistical criteria including equivalence number, minimum errors, and receiver operating characteristic.

Main Methods:

• Comparative analysis of five protein sequence search algorithms: PSW, WU-BLAST2, SSEARCH, FASTA, and BLASTP.
• Utilized a curated dataset (PIR 39) for evaluating performance with both complete-length and partial-length protein query sequences.

Main Results:

• Probabilistic Smith-Waterman (PSW) demonstrated superior accuracy for complete-length protein queries from large families.
• WU-BLAST2 and other methods showed varying performance depending on query length; BLAST and SSEARCH were effective for partial-length sequences.
• Sensitivity, indicated by false negatives, was identified as a greater challenge than selectivity across all tested methods when using optimized thresholds.

Conclusions:

• New methods like PSW offer enhanced accuracy, particularly for specific use cases (e.g., complete-length sequences).
• The choice of search method should consider the nature of the query sequence (complete vs. partial length).
• Addressing sensitivity limitations remains a key area for improvement in protein sequence homology searching.