A novel and fast approach for population structure inference using kernel-PCA and optimization.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Population structure confounds genome-wide association studies (GWAS), leading to false positives.
• Existing methods like ADMIXTURE and STRUCTURE are computationally intensive for large datasets.
• Non-model-based approaches (sNMF, EIGENSTRAT) offer better scalability.

Purpose of the Study:

• Introduce a novel, scalable method for population structure inference.
• Enable accurate admixture coefficient and principal component estimation.
• Determine the number of founder populations efficiently.

Main Methods:

• Developed PSIKO (population structure inference using kernel-PCA and optimization).
• Utilizes a combination of linear kernel-PCA and least-squares optimization.
• Non-model-based approach designed for large genomic datasets.

Main Results:

• PSIKO achieves comparable accuracy to existing leading methods.
• Demonstrates significant speed improvements, up to 30x faster than sNMF for long sequences.
• Scales exceptionally well with increasing dataset size.

Conclusions:

• PSIKO provides a computationally efficient and accurate solution for population structure inference.
• Offers a valuable tool for large-scale genetic studies, including GWAS.
• The method is freely available with accompanying documentation.