A Comparative Analysis Between k-Mers and Community Detection-Based Features for the Task of Protein Classification

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Machine learning algorithms are essential for annotating biological sequences.
• Effective feature vector generation is critical for algorithm performance.
• Previous community detection methods using Hamming distance were effective for nucleotide sequences but not protein sequences.

Purpose of the Study:

• To adapt a community detection approach for protein sequence feature generation.
• To improve the informativeness of feature vectors for protein sequence classification.

Main Methods:

• Replaced Hamming distance with substitution scores for comparing protein k-mers.
• Constructed a network based on substitution scores between k-mers.
• Applied community detection to identify informative feature sets.

Main Results:

• The novel approach successfully generated informative feature vectors for protein sequences.
• Features derived using substitution scores outperformed traditional k-mers in classification tasks.
• The method demonstrated effectiveness across various learning scenarios.

Conclusions:

• The adapted community detection method effectively addresses limitations of previous approaches for protein sequences.
• Substitution scores provide a more suitable measure for protein k-mer comparison in feature generation.
• This work advances machine learning applications in protein sequence analysis.