Linkage Disequilibrium Score Regression (LDSC)

Quantifies the proportion of phenotypic variance explained by all common single-nucleotide polymorphisms (SNPs) on a genotyping array. LDSC estimates narrow-sense heritability (h²) by regressing per-SNP χ² association statistics against LD Scores—the sum of squared correlations (r²) between a variant and its neighbors.

• Mechanism: A variant tagging more of the genome (high LD Score) will have an inflated test statistic under a polygenic model.
• Output: Provides an unbiased estimate of observed-scale heritability, convertible to the liability scale for binary disease traits.
• Advantage: Unlike GREML methods (e.g., GCTA), LDSC requires only summary statistics, not individual-level genotype data.

Measures the genome-wide genetic correlation (r_g) between two complex traits or diseases using their respective GWAS summary statistics. This reveals the extent to which the same causal variants influence both phenotypes.

• Cross-Trait LDSC: Extends the univariate model by regressing the product of Z-scores from two studies against LD Scores.
• Interpretation: An r_g of +1 indicates perfect positive genetic overlap; 0 indicates no shared genetic architecture.
• Application: Used to identify pleiotropic relationships between psychiatric disorders or to validate the genetic overlap between a biomarker and a clinical endpoint.

Distinguishes genuine polygenic signals from spurious inflation caused by population stratification and cryptic relatedness. The LDSC intercept provides a direct estimate of the inflation factor attributable to confounding rather than true polygenicity.

• Unconstrained Intercept: Under a polygenic model with no confounding, the intercept should equal 1.0. Values significantly greater than 1.0 indicate residual population structure.
• LD Score Intercept: Corrects the genomic inflation factor (λ_GC) by partitioning inflation into polygenic signal versus bias.
• Diagnostic Utility: A critical quality control step before interpreting GWAS results or conducting downstream causal inference like Mendelian Randomization.

Decomposes the total SNP-heritability into contributions from specific functional genomic annotations. Stratified LDSC evaluates whether certain categories of SNPs are disproportionately enriched for heritability.

• Annotations: Tests enrichment in coding regions, conserved elements, cell-type-specific regulatory marks (e.g., from ENCODE or Roadmap Epigenomics), and minor allele frequency bins.
• Enrichment Score: Defined as the proportion of heritability explained by an annotation divided by the proportion of SNPs in that annotation.
• Biological Insight: Reveals that variants in fetal brain-specific enhancers are enriched for psychiatric disorder heritability, guiding functional follow-up experiments.

The foundational pre-processing step requiring a reference panel of individual-level genotypes to compute the LD Score for each variant. This score represents the total amount of genetic variation tagged by a given SNP.

• Reference Panels: Typically uses the 1000 Genomes Project or a population-matched cohort to ensure accurate LD structure representation.
• Sliding Windows: LD is calculated within a fixed window (e.g., 1 centimorgan) around the index SNP, excluding the MHC region due to its complex LD architecture.
• Quality Control: HapMap3 SNPs with high imputation quality (INFO > 0.9) are recommended to ensure robust regression estimates and avoid bias from poorly imputed variants.