AllergenFP: allergenicity prediction by descriptor fingerprints

Area of Science:

• Computational biology
• Bioinformatics
• Protein analysis

Background:

• Allergenicity prediction is challenging for traditional alignment-based methods due to complex encoding.
• A novel alignment-free approach is needed to accurately identify allergens.

Purpose of the Study:

• To develop and present a novel alignment-free, descriptor-based fingerprint approach for allergen identification.
• To create a universal computational tool applicable to various classification problems in bioinformatics.

Main Methods:

• Protein sequences are characterized by amino acid principal properties (e.g., hydrophobicity, size).
• Auto- and cross-covariance (ACC) transforms sequence strings into fixed-length vectors.
• Vectors are converted into binary fingerprints for Tanimoto coefficient-based comparison.

Main Results:

• The method correctly identified 88% of 2427 known allergens and 2427 non-allergens.
• Achieved a Matthews correlation coefficient of 0.759.
• The descriptor fingerprint approach captures key physicochemical and structural properties of amino acids.

Conclusions:

• The developed descriptor fingerprint approach is universal and effective for allergen classification.
• Auto-covariance transformation effectively handles variable protein sequence lengths.
• The AllergenFP web server provides free access to this classification tool.