Nonlinear projection methods for visualizing Barcode data and application on two data sets.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• DNA Barcoding is crucial for species identification, but visualizing complex sequence data presents challenges.
• Existing statistical visualization tools like Multidimensional Scaling (MDS) and Self-Organizing Maps (SOM) are often unsuitable for non-Euclidean Barcode data.
• Unsupervised learning methods are key for uncovering structure and patterns in biological datasets.

Purpose of the Study:

• To develop and validate a novel computational workflow for visualizing DNA Barcode data.
• To adapt and combine clustering and projection methods for effective dimensionality reduction and data structuring.
• To apply the workflow to real-world datasets for species identification and taxonomic analysis.

Main Methods:

• A four-step workflow was developed: data collapse, dissimilarity matrix computation, modified SOM for dissimilarity matrices, and MDS projection.
• The methodology integrates unsupervised statistical learning techniques for data visualization and structural analysis.
• The workflow was tested on DNA Barcode data from Astraptes fulgerator and Hylomyscus.

Main Results:

• The developed workflow successfully structured and visualized DNA Barcode data, demonstrating robustness with unbalanced species datasets.
• Distinct groupings were observed for Astraptes species, with minor overlaps for specific taxa.
• For Hylomyscus, the results corroborated known taxonomy, identified unnamed taxa, and suggested potential new species.

Conclusions:

• The novel workflow provides an effective approach for visualizing and analyzing DNA Barcode data, enhancing species identification.
• The method's ability to handle complex datasets and reveal taxonomic structures is validated by its application to diverse biological examples.
• This approach contributes to advancing computational tools for biodiversity research and taxonomic studies.