Fig. 2: Polygenic score contribution of LNL-ISO (PGSLNL-ISO) and schizophrenia (PGSSCZ) to schizophrenia risk and heritability estimates.
From: Polygenic contribution to the relationship of loneliness and social isolation with schizophrenia
A PGS predictions of LNL-ISO (PGSLNL-ISO) and its constituent phenotypes (see legend) on an independent schizophrenia case-control sample (NSCZ = 1927; NHC = 1561). Explained variance attributable to PGS was calculated as the increase in Nagelkerke’s pseudo-R2 between a linear model with and without the PGS variable. P-values were obtained from the binomial logistic regression of SCZ phenotype on PGS, accounting for Linkage Disequilibrium (LD) and including sex, age, and ten multidimensional scalings (MDS) ancestry components as covariates. Significant PGS predictions after FDR correction (pFDR < 0.05) are marked with an asterisk. See Supplementary Fig. 1 for R2 values for PGS predictions on the liability scale estimated using UK Biobank prevalence for LNL-ISO constituent phenotypes. For a full detailed description and results see Supplementary Methods 3 and Supplementary Data 1. B PGS predictions of schizophrenia (PGSSCZ) on an independent schizophrenia case-control sample (NSCZ = 1927; NHC = 1561). We used schizophrenia GWAS summary statistics overlapping with LNL-ISO summary statistics (SCZ(ALL)) and three subsets of them based on their effects on LNL-ISO: variants not associated with LNL-ISO (SCZ[noLNL]) and those associated with LNL-ISO with either concordant (SCZ[CONC]) or discordant (SCZ[DISC]) allele effects in each trait. Explained variance attributable to PGS was calculated as the increase in Nagelkerke’s pseudo-R2 between a linear model with and without the PGS variable. Pseudo-R2 was converted to liability scale following the procedure proposed by Lee et al.85 assuming a prevalence of schizophrenia in the general population of 1%86. P-values were obtained from the binomial logistic regression of SCZ phenotype on PGS, accounting for LD and including sex, age, and ten MDS ancestry components as covariates. Significant PGS predictions after FDR correction (pFDR < 0.05) are marked with an asterisk. For a full detailed description and results see Supplementary Methods 4 and Supplementary Data 2 A. C Quantile plot of PGSSCZ predictions from the partitions described in B. The target sample is separated into deciles of increasing PGSSCZ. The case-control status of each decile is compared to the median (5th decile), one by one, using a logistic regression model with covariates (sex, age, and ten MDS ancestry components). OR values for each comparison were estimated from regression coefficients of these decile-status predictors. Significant comparisons (pFDR < 0.05) are marked with an asterisk. For a full detailed description and results see Supplementary Methods 4 and Supplementary Data 2B. D Proportion of SNP-based heritability (h2SNP) and heritability enrichment (h2SNP/NSNP) of the annotations in schizophrenia were estimated by LD-score regression (LDSR). 95% confidence intervals based on standard errors are shown for each estimate (estimation +/− 1.96*SE). p-values and standard errors were calculated using a block jackknife procedure. See Supplementary Data 3 for the significance of each enrichment estimate.
