OCMA: Fast, Memory-Efficient Factorization of Prohibitively Large Relationship Matrices.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic Relationship Matrices (GRMs) are crucial for genetic analyses like heritability estimation and genomic prediction.
• Large biobank datasets exceed active memory capacity, hindering classical eigen-decomposition and SVD analyses.
• Existing methods struggle with the scale of modern genomic data.

Purpose of the Study:

• To develop a memory-efficient, disk-based tool for large-scale eigen and Singular Value Decomposition (SVD) of genomic matrices.
• To enable scalable genetic analyses on big data using readily available computing resources.
• To overcome computational bottlenecks in analyzing massive genotype and GRM data.

Main Methods:

• Developed Out-of-Core Matrices Analyzer (OCMA), a disk-based computational tool.
• Integrated memory mapping (mmap) and advanced matrix factorization libraries.
• Implemented multi-threading capabilities for enhanced performance on desktops and HPC clusters.

Main Results:

• OCMA efficiently solved eigen-decomposition for a 10,000-individual GRM in 55 seconds on a personal computer.
• OCMA computed top singular values for a 1,000,000-individual genotype matrix within hours.
• Demonstrated significant speed and memory efficiency for large-scale genomic matrix analyses.

Conclusions:

• OCMA effectively addresses memory limitations for analyzing large genomic matrices.
• The tool facilitates scalable application of classical and emerging genetic analysis methods.
• OCMA democratizes big genomics data analysis by enabling use of lightweight computing resources.