TeraPCA

TeraPCA computes principal components from tera-scale genotype matrices (millions of individuals genotyped at millions of markers) to support population-structure analysis and large-scale studies of genetic variation.

Key Features:

• Scalability and Efficiency: Handles datasets on the order of millions of individuals and millions of markers and computes principal components with low memory requirements (on the order of a few gigabytes of RAM when applicable).
• In-Core and Out-of-Core Capabilities: Operates in-core when sufficient memory is available and out-of-core using disk storage when datasets exceed memory capacity.
• Minimal Dependencies: Requires BLAS (Basic Linear Algebra Subprograms) and LAPACK (Linear Algebra Package) as external libraries.
• Performance: In multi-threaded mode can compute the 10 leading principal components for a dataset with one million individuals genotyped at one million markers in under five hours.
• Accuracy and Reliability: Experimental analyses reported show fast and accurate recovery of principal components for population-structure and genetic-variation studies.

Scientific Applications:

• Population-structure analysis: Principal component analysis of large genotype matrices to infer population structure in human genetics.
• Genetic diversity and evolutionary biology: Examination of genetic variation and patterns relevant to evolutionary studies using large-scale genotype data.
• Disease association studies: Large-cohort PCA for controlling population stratification in association analyses.
• Tera-scale genotype dataset analysis: Enabling PCA on tera-scale datasets that comprise millions of samples and markers.

Methodology:

The core method is Randomized Subspace Iteration using random projections and iterative refinement to compute principal components efficiently in-core or out-of-core without loading entire datasets into memory.