lcUMAPtSNE: use of non-linear dimensionality reduction techniques with genotype likelihoods

Area of Science:

• Conservation Genetics
• Population Genomics
• Bioinformatics

Background:

• Population structure analysis is crucial for effective conservation genetics, informing strategies for genetic diversity and adaptability.
• Dimensionality reduction techniques like t-distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) are effective for human and model organisms but underexplored in wild species using low-coverage sequencing data.

Purpose of the Study:

• To develop and demonstrate a workflow for applying UMAP and t-SNE to genotype likelihoods for population structure analysis in wildlife.
• To provide guidance on hyperparameter tuning and practical implementation for these methods in wildlife genetics.

Main Methods:

• A Jupyter Notebook-based workflow was created to apply UMAP and t-SNE on principal components derived from genotype likelihoods.
• The workflow was tested using medium to low-coverage whole-genome sequencing data from scimitar-horned oryx.

Main Results:

• The workflow successfully applied UMAP and t-SNE to genotype likelihood data, demonstrating their utility in revealing population genetic structures.
• The study provides practical implementation details and hyperparameter tuning guidance for these dimensionality reduction techniques.

Conclusions:

• The developed workflow (lcUMAPtSNE) serves as a complementary exploratory tool for wildlife genetics, supporting biodiversity conservation efforts.
• The application of UMAP and t-SNE on genotype likelihoods offers a valuable approach for understanding population connectivity and genetic diversity in wild species.