Ancestral informative marker selection and population structure visualization using sparse Laplacian eigenfunctions.

Area of Science:

• Population Genetics
• Genomics
• Bioinformatics

Background:

• Identifying population structure informative markers is crucial for genetic studies, but traditional methods struggle with admixed populations.
• Principal Component Analysis (PCA) has been used to select SNPs correlated with population structure, applicable to admixed groups.

Purpose of the Study:

• To propose a novel geometric approach for selecting population structure informative markers.
• To demonstrate the effectiveness of this new method compared to PCA, especially for admixed populations.

Main Methods:

• Developed a novel algorithm based on graph Laplacian eigenfunctions to summarize population structure.
• Validated the algorithm using simulations and the Human Genome Diversity Project (HGDP) dataset.
• Compared marker informativeness and redundancy against PCA using Support Vector Machine (SVM) for continental membership prediction.

Main Results:

• The proposed geometric method efficiently selects the most informative markers.
• Selected markers by the novel method are more informative and less redundant than those selected by PCA.
• The algorithm successfully recovers underlying population structure with a small fraction of markers.

Conclusions:

• The novel graph Laplacian eigenfunction-based algorithm is a promising tool for selecting informative markers in population genetics.
• This method facilitates efficient population substructure detection and ancestral inference, with applications in genome-wide association studies.