A new method for species identification via protein-coding and non-coding DNA barcodes by combining machine learning...

Area of Science:

• Bioinformatics and Computational Biology
• Genetics and Genomics
• Molecular Ecology

Background:

• DNA barcoding is crucial for bioinventory, with cytochrome c oxidase subunit I (COI) as the standard for animals.
• Internal transcribed spacer (ITS) genes are proposed for animals and plants but face alignment challenges.
• Robust species identification methods are needed for both coding and non-coding DNA sequences.

Purpose of the Study:

• To introduce two novel machine learning-based methods, DV-RBF and FJ-RBF, for accurate species assignment using DNA barcodes.
• To evaluate the performance of these new methods against existing approaches for both coding and non-coding genetic markers.
• To demonstrate the effectiveness of DV-RBF and FJ-RBF across diverse taxa including mammals, fish, fungi, and algae.

Main Methods:

• Development of two new algorithms, DV-RBF and FJ-RBF, leveraging machine learning and bioinformatics for DNA sequence alignment and species identification.
• Application of the new methods to empirical datasets of neotropical bats (COI), marine fish (COI), rust fungi (ITS), and brown algae (ITS).
• Comparative analysis against Neighbor-joining (NJ) and Maximum Likelihood (ML) methods using random sub-sampling techniques.

Main Results:

• DV-RBF and FJ-RBF significantly outperformed NJ and ML methods for both COI and ITS barcodes when reference datasets had complete species coverage.
• For non-coding ITS sequences, the new methods maintained superior performance even with incomplete species coverage in reference data.
• Achieved 100% species identification success for bats and fish (COI) and high success rates for fungi and algae (ITS).

Conclusions:

• The DV-RBF and FJ-RBF methods offer a significant advancement in DNA barcoding for species identification, particularly for challenging non-coding regions.
• These machine learning approaches provide robust and accurate species assignment across diverse taxonomic groups and marker types.
• The proposed methods enhance bioinventory efforts by improving the reliability and efficiency of DNA-based species identification.