Large-scale classification of metagenomic samples: a comparative analysis of classical machine learning techniques vs...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Classical machine learning struggles with high-dimensional biological data, impacting accuracy.
• High dimensionality in biological datasets presents significant analytical challenges.

Purpose of the Study:

• To evaluate hyperdimensional computing (HDC) as a supervised machine learning approach for bioinformatics.
• To compare HDC performance against established methods using metagenomic classification tasks.

Main Methods:

• Comparative analysis of HDC and classical machine learning techniques.
• Supervised classification of heterogeneous metagenomic samples using quantitative microbial profiles.
• Utilized publicly available microbiome datasets for validation.

Main Results:

• HDC demonstrated comparable or superior classification accuracy to classical methods.
• HDC showed potential for improved computational efficiency with large-scale datasets.
• Metagenomic sample classification using HDC yielded promising results.

Conclusions:

• HDC is a promising tool for bioinformatics, especially for high-dimensional data.
• HDC can complement or surpass established machine learning techniques in performance.
• A Galaxy-powered toolset is provided for accessible adoption of HDC methods.