Improving taxonomic classification with feature space balancing.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Microbial Ecology

Background:

• High-throughput sequencing generates millions of DNA sequences requiring taxonomic assignment.
• Current methods like MMseqs2 (alignment-based) and BERTax (deep learning) face challenges with large datasets, runtime, hardware, and energy consumption.

Purpose of the Study:

• To develop a more efficient and accurate taxonomic classification method for metagenomic data.
• To enhance the predictive power of k-mer profiles through feature space balancing.

Main Methods:

• Utilized k-mer profiles of DNA sequences as features for taxonomic classification.
• Applied a novel feature space balancing approach to training data to improve classifier generalization.
• Implemented pipelines combining feature extraction and balancing with classifiers like bagged decision trees and feature subspace KNNs.

Main Results:

• The proposed pipelines significantly improved the predictive power of k-mer profiles.
• Achieved superior performance compared to state-of-the-art algorithms (BERTax, MMseqs2) on two datasets.
• Demonstrated high precision in classifying sequences from organisms not present in the training data.

Conclusions:

• The k-mer profile approach with feature balancing offers a highly effective and efficient solution for taxonomic classification.
• This method shows promise for accurate classification of both known and novel microbial species in metagenomic datasets.