LDpred2: better, faster, stronger

Privé, Florian; Arbel, Julyan; Vilhjálmsson, Bjarni J.

doi:10.1093/bioinformatics/btaa1029

Cited by 490 publications

(653 citation statements)

References 42 publications

Supporting

Mentioning

646

Contrasting

Unclassified

Order By: Relevance

“…Shrinkage methods for polygenic scoring can be separated into frequentist penalty-based methods (e.g. lasso regression-based lassosum [ 10 ], summary-based best linear unbiased prediction (SBLUP) [ 9 ]) and Bayesian methods that shrink estimates to fit a prior distribution of effect sizes, such as LDpred1 [ 8 ], LDpred2 [ 13 ], PRScs [ 11 ], SBayesR [ 12 ], and DBSLMM [ 14 ]. Each of these methods have been shown to improve the predictive utility of polygenic scores over those derived using the pT+clump approach.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Evaluation of polygenic prediction methodology within a reference-standardized framework

et al. 2021

View full text Add to dashboard Cite

The predictive utility of polygenic scores is increasing, and many polygenic scoring methods are available, but it is unclear which method performs best. This study evaluates the predictive utility of polygenic scoring methods within a reference-standardized framework, which uses a common set of variants and reference-based estimates of linkage disequilibrium and allele frequencies to construct scores. Eight polygenic score methods were tested: p-value thresholding and clumping (pT+clump), SBLUP, lassosum, LDpred1, LDpred2, PRScs, DBSLMM and SBayesR, evaluating their performance to predict outcomes in UK Biobank and the Twins Early Development Study (TEDS). Strategies to identify optimal p-value thresholds and shrinkage parameters were compared, including 10-fold cross validation, pseudovalidation and infinitesimal models (with no validation sample), and multi-polygenic score elastic net models. LDpred2, lassosum and PRScs performed strongly using 10-fold cross-validation to identify the most predictive p-value threshold or shrinkage parameter, giving a relative improvement of 16–18% over pT+clump in the correlation between observed and predicted outcome values. Using pseudovalidation, the best methods were PRScs, DBSLMM and SBayesR. PRScs pseudovalidation was only 3% worse than the best polygenic score identified by 10-fold cross validation. Elastic net models containing polygenic scores based on a range of parameters consistently improved prediction over any single polygenic score. Within a reference-standardized framework, the best polygenic prediction was achieved using LDpred2, lassosum and PRScs, modeling multiple polygenic scores derived using multiple parameters. This study will help researchers performing polygenic score studies to select the most powerful and predictive analysis methods.

show abstract

Section: Introductionmentioning

confidence: 99%

“…Use of a reference-standardized framework also offers advantages by improving the comparability of polygenic scores across cohorts. Several polygenic scoring methods now recommend the use of HapMap3 SNPs and precomputed external LD estimate references [ 11 – 13 ], in line with a reference-standardized approach.…”

Section: Introductionmentioning

confidence: 99%

Evaluation of polygenic prediction methodology within a reference-standardized framework

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Methods differ in the choice of the prior distribution. For example, LDpred and LDpred2 assumes a point-normal mixture distribution or a single normal distribution [3,4]. SBayesR assumes a mixture of three normal distributions with a point mass at zero [5].…”

Section: Introductionmentioning

confidence: 99%

A fast and robust Bayesian nonparametric method for prediction of complex traits using summary statistics

Zhou

Zhao

2020

Preprint

View full text Add to dashboard Cite

Genetic prediction of complex traits has great promise for disease prevention, monitoring, and treatment. The development of accurate risk prediction models is hindered by the wide diversity of genetic architecture across different traits, limited access to individual level data for training and parameter tuning, and the demand for computational resources. To overcome the limitations of the most existing methods that make explicit assumptions on the underlying genetic architecture and need a separate validation data set for parameter tuning, we develop a summary statistics-based nonparametric method that does not rely on specific parametric assumptions about genetic architecture and validation datasets to tune parameters. In our implementation, we refine the commonly used likelihood assumption to deal with the discrepancy between summary statistics and external reference panel. We also leverage the block structure of the reference linkage disequilibrium matrix for implementation of a parallel algorithm. Through simulations and applications to twelve traits, we show that our method is adaptive to different genetic architectures, statistically robust, and computationally efficient. Our method is available at https://github.com/eldronzhou/SDPR.

show abstract

“…1). We considered two alternative methods to select the SNP effects to include in the PGS calculation: pruning and thresholding (P&T) and LDpred2 28,29 . When we included the PGS obtained using P&T as a fixed effect with fastGWA (which we refer to as fastGWA-PGS-PT) we recovered 82 additional causal variants, on average, below the conventional P-value threshold of 5x10 −8 compared to fastGWA (corresponding to a relative increase in power of 18.4%; p = 3.0 × 10 −32 from a paired T-test; Table S1-S3).…”

Section: /17mentioning

confidence: 99%

“…Association testing was then performed using fastGWA in a chromosome-wise manner, with the corresponding LOCO PGS included as a fixed effect. The bigsnpr R package was used to calculate the LOCO PGS-LDpred2 fixed effects 29 . To reduce computation time, 22 LOCO genotype objects containing the SNP correlations were precomputed.…”

Section: Simulation Association Testsmentioning

confidence: 99%

Controlling for background genetic effects using polygenic scores improves the power of genome-wide association studies

Bennett

Morris

Seoighe

2020

Preprint

View full text Add to dashboard Cite

Ongoing increases in the size of human genotype and phenotype collections offer the promise of improved understanding of the genetics of complex diseases.In addition to the biological insights that can be gained from the nature of the variants that contribute to the genetic component of complex trait variability, these data have brought forward the prospect of predicting complex traits and the risk of complex genetic diseases from genotype data. Optimal realization of these objectives requires ongoing methodological developments, designed to identify true trait-associated variants and accurately predict phenotype from genotype. In addition to optimal accuracy, these methods must also be computationally efficient, in order to remain tractable in the context of high variant densities and very large sample sizes. Here we show that the power of linear mixed models that are in widespread use for GWAS can be increased significantly by modeling off target genetic effects using polygenic scores derived from SNPs that are not on the same chromosome as the target SNP. Applying this method to simulated and real data we found that it can result in a substantial increase in the number of variants passing genome-wide significance thresholds. This increase in power to detect trait-associated variants also translates into an increase in the accuracy with which the resulting polygenic score predicts the phenotype from genotype data. Bachground 1 Linear mixed effects models (LMMs) are routinely applied to detect associations 2 between SNPs and phenotypes in genome-wide association studies (GWAS). The 3 Bennett et al. Page 2 of 16 contribution of a given SNP to the phenotype is modelled as a fixed effect, while 4 the contribution of all other SNPs is modelled as a random effect, with the co-5 variance structure of this random effect corresponding to the genetic relationships 6 between the individuals in the study. In order to avoid including the component 7 of the genetic relatedness contributed by variants in linkage disequilibrium (LD) 8 with the tested variant, the genetic relationship matrix (GRM) is calculated using 9 an LD-pruned set of variants [1, 2]. Originally these methods assumed that a large 10 number of SNPs may each make small contributions to the phenotypic variation, 11 such that the phenotypic contribution of the genetic background could be mod-12 elled as a multivariate normal (with covariance matrix proportional to the GRM) 13 [2, 3, 4, 5]. Newer implementations of these models have allowed for the possibility 14 of a combination of SNPs with effect sizes close to zero as well as some SNPs with 15 large effect sizes. This can be done by modelling the effect sizes of background SNPs 16 as a mixture of normal distributions, with a proportion, p << 1, of SNPs with large 17 effect sizes and the remainder with effect sizes close to zero. This results in a 'spike 18 and slab' mixture model for the SNP effect sizes [6] (the spike corresponds to a nor-19 mal component with small variance and the slab corresponding to a ...

show abstract

LDpred2: better, faster, stronger

Cited by 490 publications

References 42 publications

Evaluation of polygenic prediction methodology within a reference-standardized framework

Evaluation of polygenic prediction methodology within a reference-standardized framework

A fast and robust Bayesian nonparametric method for prediction of complex traits using summary statistics

Controlling for background genetic effects using polygenic scores improves the power of genome-wide association studies

Contact Info

Product

Resources

About