The power of word-frequency-based alignment-free functions: a comprehensive large-scale experimental analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomic Data Analysis

Background:

• Alignment-free (AF) distance/similarity functions are crucial for sequence analysis.
• Existing studies often focus on false positive rates and limited sequence lengths, neglecting the power of AF functions.
• Current assessment of AF function power is insufficient for modern applications with variable sequence lengths.

Purpose of the Study:

• To conduct a comprehensive evaluation of the power of word-frequency-based AF functions.
• To assess both the power (true similarity detection) and Type I error (false positives) of selected AF functions.
• To provide guidance for selecting appropriate AF functions for diverse sequence analysis tasks.

Main Methods:

• Evaluated a representative set of word-frequency-based AF functions.
• Utilized models of CIS Regulatory Modules and Horizontal Gene Transfer for genomic feature representation.
• Tested functions across a wide range of sequence lengths (thousands to millions) and k values.
• Performed a coherent and uniform evaluation of AF function power and Type I error.

Main Results:

• Identified significant variations in the power of different AF functions.
• Four out of 15 evaluated AF functions demonstrated superior performance across various sequence lengths.
• Characterized the strengths and weaknesses of each AF function, offering practical insights.
• Observed minimal performance differences between short and long sequences for the top-performing functions.

Conclusions:

• The study provides the first uniform evaluation of AF function power across diverse genomic scenarios and sequence lengths.
• The findings offer a practical guide for researchers to select optimal AF functions for their specific analysis needs.
• A public Big Data platform and open-source software are available to support the validation of future AF functions.