Performance evaluation of amino acid substitution matrices

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Amino acid substitution matrices are essential for sequence alignment and searching.
• Existing matrices vary in their derivation and performance.
• The selection of an appropriate matrix significantly impacts the accuracy of bioinformatics analyses.

Purpose of the Study:

• To evaluate the performance of different amino acid substitution matrices in protein searching and alignment.
• To compare matrices derived from direct alignments versus those based on evolutionary models.
• To assess the utility of a multiple matrix strategy.

Main Methods:

• Utilized the BLAST and FASTA search programs for evaluating matrix performance.
• Employed the Prosite catalog to assess matrix effectiveness across diverse protein families.
• Compared matrices derived from sequence-based and structure-based alignments against extrapolated matrices (Dayhoff model).
• Tested a multiple matrix strategy, including combinations of sequence- and structure-based matrices.

Main Results:

• Matrices derived from direct sequence or structure alignments performed significantly better than extrapolated matrices.
• BLOSUM 62 emerged as a top-performing single matrix.
• Improved performance was observed generally, not limited to specific protein families.
• A hybrid set of sequence- and structure-based matrices yielded good results, though not superior to the best single matrix.

Conclusions:

• Matrix selection is a critical factor in the accuracy of protein comparison tools.
• Directly derived matrices offer superior performance for sequence searching and alignment.
• A comprehensive evaluation approach is valuable for assessing bioinformatics tools.
• Hybrid matrix strategies may offer benefits in specific protein analysis scenarios.