bigstatsr

bigstatsr implements scalable statistical methods for large-scale genomic data analysis, enabling memory-mapped handling and analysis of millions of genetic variants for association studies, principal component analysis, linkage disequilibrium pruning, and polygenic risk score computation.

Key Features:

• Memory-mapping for efficient data handling: Accesses large genotype and phenotype matrices directly from disk via memory-mapping to reduce RAM usage and enable analyses of very large datasets.
• Integration with existing tools: Executes transparent system calls to existing genomic analysis software or provides updated implementations of common methods within R.
• Principal Component Analysis (PCA): Implements fast and accurate PCA computations for dimensionality reduction and population structure correction.
• Association studies: Supports genome-wide association analyses to identify genetic variants associated with traits and diseases.
• Linkage disequilibrium (LD) pruning: Provides functions to remove correlated single nucleotide polymorphisms (SNPs) to refine variant sets for downstream analyses.
• Polygenic risk scores (PRS): Implements algorithms to compute polygenic risk scores across millions of SNPs for genetic predisposition assessment.
• Large-scale dataset support: Designed to handle datasets comprising hundreds of thousands of individuals and millions of markers, as demonstrated in simulated genome-wide analyses.

Scientific Applications:

• Genome-wide association studies (GWAS): Performs association analyses to identify SNPs linked to traits or diseases.
• Polygenic risk score estimation: Calculates PRS across genome-wide marker sets to quantify genetic risk.
• Case-control analysis (celiac disease example): Applied to a celiac disease case-control dataset to perform association testing and PRS computation.
• Scalability evaluation on simulated datasets: Used to analyze a simulated genome-wide dataset of 500,000 individuals and one million markers to demonstrate large-scale performance.

Methodology:

Uses memory-mapping for optimized data access, transparent system calls or reimplementations of existing genomic tools, and implements fast PCA, association testing, LD pruning, and polygenic risk score algorithms.