Simplified amino acid alphabets based on deviation of conditional probability from random background.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Biology

Background:

• Amino acid substitution matrices like Miyazawa-Jernigan and BLOSUM are crucial for protein sequence analysis.
• These matrices are derived from residue pair frequency counts, representing conditional probabilities.
• Reducing the complexity of the amino acid alphabet is desirable for efficient analysis.

Purpose of the Study:

• To develop a scheme for reducing the amino acid alphabet based on conditional probability deviations from random background.
• To compare reduced alphabets derived from Miyazawa-Jernigan and BLOSUM data.
• To evaluate the effectiveness of reduced alphabets in protein homology detection.

Main Methods:

• Calculating conditional probability distributions for each amino acid from pair frequency counts.
• Proposing an alphabet reduction scheme based on deviations from random background probabilities.
• Generating coarse-grained substitution matrices from reduced alphabets.
• Testing homology detection using the SCOP40 homologous sequence database.

Main Results:

• A novel scheme for amino acid alphabet reduction was successfully developed.
• Discrepancies were observed between reduced alphabets derived from Miyazawa-Jernigan and BLOSUM data.
• Homology detection using reduced alphabets and coarse-grained matrices proved effective on the SCOP40 dataset.
• The reduced alphabets were shown to retain significant information from the original 20-letter alphabet.

Conclusions:

• The proposed alphabet reduction scheme effectively simplifies amino acid representation while preserving crucial information.
• Reduced alphabets show promise for efficient and accurate protein homology detection.
• Further investigation into the discrepancies between different data sources for alphabet reduction is warranted.