OCMA

OCMA performs out-of-core eigen-decomposition and Singular Value Decomposition (SVD) of large genomic matrices to enable analysis of Genomic Relationship Matrices (GRMs) and genotype matrices that exceed active memory capacity.

Key Features:

• Disk-Based Computation: Handles matrices that exceed active memory by performing disk-based (out-of-core) computation, enabling processing of large Genomic Relationship Matrices (GRMs) and genotype matrices.
• Eigen-Decomposition and SVD: Performs eigen-decomposition and SVD for analyses such as SNP-heritability estimation, Principal Component Analysis (PCA), and genomic prediction, with SVD available as a computationally faster alternative.
• Memory Efficiency: Integrates memory mapping (mmap) with matrix factorization libraries to minimize active memory usage.
• Performance Metrics: On a typical personal computer, completes full eigen-decomposition of an ordinary GRM with 10,000 individuals in ~55 seconds; computes top 200 singular values in ~0.5 hours, top 2,000 in ~0.95 hours, and all 5,000 singular values from a genotype matrix of 1,000,000 individuals by 5,000 markers in ~1.77 hours.
• Multi-Threading Support: Supports multi-threading to scale performance across desktop environments and high-performance computing (HPC) clusters.

Scientific Applications:

• Estimation of SNP-heritability: Provides eigen- and singular-value outputs used in SNP-heritability estimation workflows.
• Principal Component Analysis (PCA): Computes principal components from GRMs or genotype matrices for population structure and ancestry analyses.
• Genomic prediction: Supplies matrix factorization results applicable to genomic prediction models.

Methodology:

Performs disk-based (out-of-core) computation using memory mapping (mmap) and matrix factorization libraries to execute eigen-decomposition and SVD with multi-threading support.