An enhanced framework for local genetic correlation analysis

We compared HDL-L and LAVA for estimating heritability, genetic covariance, and genetic correlation across 50 simulation replicates. Six pairs of true heritability values were considered for the two phenotypes: \({h}_{1}^{2}\,=\,0.2,\,{h}_{2}^{2}\,=\,0.4\); \({h}_{1}^{2}\,=\,0.2,\,{h}_{2}^{2}\,=\,0.8;\) \({h}_{1}^{2}\,=\,0.6,\,{h}_{2}^{2}\,=\,0.8\); \({h}_{1}^{2}\,=\,0.3,\,{h}_{2}^{2}\,=\,0.3;\) \({h}_{1}^{2}\,=\,0.5,\,{h}_{2}^{2}\,=\,0.5;\) \({h}_{1}^{2}\,=\,0.7,\,{h}_{2}^{2}\,=\,0.7\). a-d, Chromosome 22 results: The boxplots depict estimation bias (Estimated minus true values), and the scatterplots with LOESS-fitted curves in the right panel show HDL-L’s improved accuracy. For each box, the horizontal line represents the median, the central box indicates the interquartile range (IQR), and whiskers extend up to 1.5 times the IQR.

HDL-L and LAVA were evaluated for estimating heritability, genetic covariance, and genetic correlation over 50 simulation replicates. Line plots show median mean squared error across 2,468 regions. The heatmap indicates the heritability setting for each phenotype, where \({h}_{1}^{2}\,\) and \({h}_{2}^{2}\) denote the true genome-wide heritability for trait 1 and trait 2, respectively. 10% of SNPs were randomly selected as causal SNPs.

We considered six different heritability combinations of the two phenotypes: \({h}_{1}^{2}=0.2,\,{h}_{2}^{2}=0.4\); \({h}_{1}^{2}=0.2,\,{h}_{2}^{2}=0.8;\) \({h}_{1}^{2}=0.6,\,{h}_{2}^{2}=0.8\); \({h}_{1}^{2}=0.3,\,{h}_{2}^{2}=0.3;\) \({h}_{1}^{2}=0.5,\,{h}_{2}^{2}=0.5;\) \({h}_{1}^{2}=0.7,\,{h}_{2}^{2}=0.7\). The true genetic correlation (\({r}_{G}\)) was set at 0 (null hypothesis). 10% of SNPs were randomly selected as causal SNPs.

We considered six different heritability combinations of the two phenotypes: \({h}_{1}^{2}=0.2,\,{h}_{2}^{2}=0.4\); \({h}_{1}^{2}=0.2,\,{h}_{2}^{2}=0.8;\) \({h}_{1}^{2}=0.6,\,{h}_{2}^{2}=0.8\); \({h}_{1}^{2}=0.3,\,{h}_{2}^{2}=0.3;\) \({h}_{1}^{2}=0.5,\,{h}_{2}^{2}=0.5;\) \({h}_{1}^{2}=0.7,\,{h}_{2}^{2}=0.7\), with nine settings of true genetic correlation (\({r}_{G}\) from −1 to 1). 10% of SNPs were randomly selected as causal SNPs.

We randomly selected 16 loci. After performing eigen-decomposition to the LD matrix, leading eigenvalues explaining different amount of variances of the LD matrix and their corresponding eigenvectors were taken to approximate the LD matrix. Each boxplot shows the distribution of genetic covariance bias across simulations, with the box spanning the first quartile to the third quartile (Q1–Q3) and the horizontal line indicating the median.

The figure presents the comparative analysis of HDL-L and LAVA in estimating heritability and genetic covariance from 50 simulation repeats. The predetermined genome-wide true genetic correlation was varying from −0.5, 0, 0.5, with the heritability of trait 1 (\({h}_{1}^{2}\)) on the whole genome set at 0.2, the heritability of trait 2 (\({h}_{2}^{2}\)) on the whole genome set at 0.4, and 10% of SNPs were randomly selected as causal variants. The analysis was conducted on 2,468 genomic loci and 50 simulation replicates. For the boxplots, the central line in each box marks the median value, the lower and upper edges represent the first and third quartiles (Q1 and Q3), and the whiskers extend 1.5 \(\times\) IQR (interquartile range) beyond Q1 or Q3. a, Boxplots of false positive rate. b, P-value distribution under null hypothesis. The P-value was derived by comparing the two-sided likelihood ratio test (LRT) statistic to a chi-squared distribution with 1 degree of freedom. c, Boxplots of true positive rate. d, Boxplots of mean squared error for heritabilities and genetic covariance.