SBL

SBL implements sparse Bayesian learning to perform multiple-locus mapping for genome-wide association studies (GWAS) and quantitative trait locus (QTL) mapping, enabling simultaneous detection of genetic markers that contribute to trait variance.

Key Features:

• Multiple-locus model: Simultaneously detects all markers within a single framework and addresses challenges such as large matrix inversions and overly conservative significance thresholds like Bonferroni correction.
• Coordinate descent algorithm: Estimates marker-effect parameters iteratively by updating one parameter at a time while conditioning on current values of other parameters, supporting scalability to datasets with sample sizes exceeding 100,000.
• L2-type penalty: Incorporates an L2 penalty to promote sparsity in high-dimensional marker-effect estimation and help control false positives.
• High statistical power and sensitivity: Simulation studies report higher power and consistent detection of true loci with extremely small P-values, indicating robustness to stringent significance thresholds.

Scientific Applications:

• Genome-wide association studies (GWAS): Detection of multiple associated loci and subtle genetic effects across the genome.
• Quantitative trait locus (QTL) mapping: Simultaneous mapping of QTLs in high-dimensional genotype data.
• Large-scale genomic datasets: Analysis of datasets with thousands to millions of markers and very large sample sizes (including >100,000 samples).
• Estimation of genetic contribution: Assessment of the aggregate genetic contribution to trait variance in complex traits.

Methodology:

SBL applies a sparse Bayesian multiple-locus model fitted via a coordinate descent algorithm that iteratively updates marker-effect parameters under an L2-type penalty.