PCA-based population structure inference with generic clustering algorithms.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Population structure inference from large genotype datasets is computationally intensive.
• Existing methods like STRUCTURE, while powerful, are time-consuming due to Markov Chain Monte Carlo (MCMC) parameter estimation.
• Efficiently analyzing genetic variation across numerous loci is crucial for understanding population dynamics.

Purpose of the Study:

• To introduce a computationally efficient principal component analysis (PCA) based method for inferring population structure.
• To evaluate the performance of PCA combined with clustering algorithms against established methods.
• To identify optimal methods for predicting the number of subpopulations.

Main Methods:

• Applying PCA to high-density genotype data.
• Selecting significant principal components using the Tracy-Widom distribution.
• Utilizing clustering algorithms (K-means, soft K-means, spectral clustering) for subpopulation assignment.
• Comparing performance against the STRUCTURE algorithm.
• Investigating Bayesian Information Criterion (BIC) and likelihood for predicting subpopulation numbers.

Main Results:

• The proposed PCA-based approach demonstrates comparable performance to STRUCTURE in population structure inference.
• Soft K-means with BIC accurately predicted the number of subpopulations on simulated datasets, matching STRUCTURE's predictions.
• BIC proved to be a more effective index than likelihood for predicting subpopulation numbers in real datasets.

Conclusions:

• The PCA-based method offers a fast and scalable alternative for population structure inference.
• The choice of algorithm should be guided by specific application requirements and computational constraints.
• This approach significantly reduces the time required for analyzing large-scale genetic data.