Evaluating individual genome similarity with a topic model.

Area of Science:

• Genomics
• Population Genetics
• Bioinformatics

Background:

• Evaluating genome similarity is crucial for understanding population structure and human history.
• Current methods like PCA struggle to fully utilize rare genetic variants and can be difficult to interpret.
• Advanced sequencing generates vast amounts of genomic data, necessitating improved analysis techniques.

Purpose of the Study:

• To introduce a probabilistic topic model, latent Dirichlet allocation (LDA), for enhanced individual genome similarity evaluation.
• To demonstrate the efficacy of LDA in analyzing large-scale genomic datasets.
• To compare LDA's performance with existing methods like PCA.

Main Methods:

• Applied latent Dirichlet allocation (LDA), a probabilistic topic model, to analyze genome similarity.
• Utilized genomic data from 2535 individuals in the 1000 Genomes Project (KGP).
• Investigated the impact of variant selection (rare and sparse variants) and model parameters on results.

Main Results:

• Identified that rare (0.001
• Achieved significantly clearer and more cohesive population structure visualization compared to PCA.
• Observed that global genome similarities align with known geographical, historical, and cultural factors.

Conclusions:

• LDA provides a powerful and interpretable approach for evaluating individual genome similarity.
• The method effectively leverages rare genetic variants for population structure analysis.
• Findings support the utility of LDA in uncovering population genetics insights from large-scale genomic data.