Scalable classification of organisms into a taxonomy using hierarchical supervised learners

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Accurate organism identification from partial genetic material is crucial for understanding phylogenetic diversity.
• DNA barcodes offer an efficient method for species identification.
• Current DNA barcode classification methods face challenges with pre-alignment assumptions, scalability of high-performance classifiers like SVM, and reduced accuracy due to sequence mutations and noise.

Purpose of the Study:

• To propose a multi-level hierarchical classifier framework for automated DNA sequence taxonomy assignment.
• To address the limitations of existing DNA barcode classification approaches, including scalability and robustness to data imperfections.

Main Methods:

• Utilized an alignment-free spectrum kernel method for feature extraction from DNA sequences.
• Developed and evaluated a two-level hierarchical classifier on real DNA sequence data from the Barcode of Life Data Systems.
• Compared the hierarchical framework against regular classifiers.

Main Results:

• The proposed hierarchical classifier achieved a higher f1-score compared to regular classifiers.
• Demonstrated improved scalability for large datasets, allowing the use of memory-intensive, high-performance classifiers.
• Showcased enhanced robustness against mutations and noise in DNA sequence data.

Conclusions:

• The multi-level hierarchical classifier framework offers a more accurate, scalable, and robust solution for DNA sequence identification.
• This approach facilitates efficient exploration of phylogenetic diversity, even with imperfect sequence data and large-scale datasets.