Fine-tuning Polygenic Risk Scores with GWAS Summary Statistics

Zhao, Zijie; Yi, Yanyao; Wu, Yuchang; Zhong, Xiaoyuan; Lin, Yunzhi; Hohman, Timothy J.; Fletcher, Jason M.; Lu, Qiongshi

doi:10.1101/810713

Cited by 15 publications

(22 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given X T Y/n, the estimate of γ from all n training samples, our aim is to generate X A T Y A /n A an estimate of γ from n A samples (where n A =n-n E ). Our approach is similar to that of Zhao et al, 43 who propose sampling X A T Y A /n A from N(X T Y/n, n E /n A V/n), where V is the variance of X T Y. However, while Zhao et al restrict to independent SNPs, and subsequently derive V = I + X T YY T X/n 2 , where I is an identity matrix, we instead use V = X T X, as proposed by Zhu and Stephens.…”

Section: Methodsmentioning

confidence: 94%

“…Therefore, we instead create "pseudo" partial summary statistics. 43 Let γ = (γ 1 , γ 2 , …, γ m ) T denote the vector of true SNP effect sizes from single-SNP analysis (note that γ j will usually differ from β j , because β j reflects how much SNP j contributes directly to the phenotype, whereas γ j reflects how much contribution it tags). Given X T Y/n, the estimate of γ from all n training samples, our aim is to generate X A T Y A /n A an estimate of γ from n A samples (where n A =n-n E ).…”

Section: Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Improved genetic prediction of complex traits from individual-level data or summary statistics

Zhang

Privé

Vilhjálmsson

et al. 2020

Preprint

View full text Add to dashboard Cite

At present, most tools for constructing genetic prediction models begin with the assumption that all genetic variants contribute equally towards the phenotype. However, this represents a sub-optimal model for how heritability is distributed across the genome. Here we construct prediction models for 14 phenotypes from the UK Biobank (200,000 individuals per phenotype) using four of the most popular prediction tools: lasso, ridge regression, Bolt-LMM and BayesR. When we improve the assumed heritability model, prediction accuracy always improves (i.e., for all four tools and for all 14 phenotypes). When we construct prediction models using individual-level data, the best-performing tool is Bolt-LMM; if we replace its default heritability model with the most realistic model currently available, the average proportion of phenotypic variance explained increases by 19% (s.d. 2), equivalent to increasing the sample size by about a quarter. When we construct prediction models using summary statistics, the best tool depends on the phenotype. Therefore, we develop MegaPRS, a summary statistic prediction tool for constructing lasso, ridge regression, Bolt-LMM and BayesR prediction models, that allows the user to specify the heritability model.

show abstract

Section: Methodsmentioning

confidence: 94%

Section: Methodsmentioning

confidence: 99%

Improved genetic prediction of complex traits from individual-level data or summary statistics

Zhang

Privé

Vilhjálmsson

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…The phenotypes for the association analysis were chosen based on the feasibility of the CSF metabolome being relevant to the phenotype and the availability of GWAS summary statistics for the phenotype. The only exception was the GWAS for the AD proxy phenotype, which was developed in-house on the UK Biobank data set as a surrogate measure for AD risk based on parental diagnosis and age at diagnosis, following previous research 100,101 . The CNS phenotypes and sources of the GWAS summary statistics are listed in Supplementary Table 9.…”

Section: Metabolite-phenotype Association Testingmentioning

confidence: 99%

Cerebrospinal fluid metabolomics identifies 19 brain-related phenotype associations

Panyard

Kim

Darst

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Advances in technology have allowed for the study of metabolomics in the context of disease, enabling the discovery of new potential risk factors, diagnostic markers, and drug targets. For neurological and psychiatric phenotypes, the cerebrospinal fluid (CSF) is of particular biomedical importance as it is in direct contact with the brain and spinal cord. However, the CSF metabolome is difficult to study on a large scale due to the relative complexity of the procedure needed to collect the fluid compared to blood or urine studies. Here, we present a metabolome-wide association study (MWAS), an analysis using individual-level genetic and metabolomic data from two cohorts to impute metabolites into large samples with genome-wide association summary statistics. We conducted a metabolome-wide genome-wide association analysis with 338 CSF metabolites, identifying 16 genotype-metabolite associations, 6 of which were novel. Using these results, we then built prediction models for all available CSF metabolites and tested for associations with 27 neurological and psychiatric phenotypes in large cohorts, identifying 19 significant CSF metabolite-phenotype associations. Our results demonstrate the potential of MWAS to overcome the logistic challenges inherent in cerebrospinal fluid research to study the role of metabolomics in brain-related phenotypes and the feasibility of this framework for similar studies of omic data in scarce sample types.

show abstract

“…We calculated PRS of ASD and other 67 complex traits ( Table S1 ) for all parents and simulated offspring. Each PRS was statistically fine-tuned from GWAS summary statistics using PUMAS [10] ( Methods ) and standardized by cohort using the mean and standard deviation of parental scores.…”

Section: Resultsmentioning

confidence: 99%

Gamete simulation improves polygenic transmission disequilibrium analysis

Chen

You

Zhao

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

Polygenic risk scores (PRS) derived from summary statistics of genome-wide association studies (GWAS) have enjoyed great popularity in human genetics research. Applied to population cohorts, PRS can effectively stratify individuals by risk group and has promising applications in early diagnosis and clinical intervention. However, our understanding of within-family polygenic risk is incomplete, in part because the small samples per family significantly limits power. Here, to address this challenge, we introduce ORIGAMI, a computational framework that uses parental genotype data to simulate offspring genomes. ORIGAMI uses state-of-the-art genetic maps to simulate realistic recombination events on phased parental genomes and allows quantifying the prospective PRS variability within each family. We quantify and showcase the substantially reduced yet highly heterogeneous PRS variation within families for numerous complex traits. Further, we incorporate within-family PRS variability to improve polygenic transmission disequilibrium test (pTDT). Through simulations, we demonstrate that modeling within-family risk substantially improves the statistical power of pTDT. Applied to 7,805 trios of autism spectrum disorder (ASD) probands and healthy parents, we successfully replicated previously reported over-transmission of ASD, educational attainment, and schizophrenia risk, and identified multiple novel traits with significant transmission disequilibrium. These results provided novel etiologic insights into the shared genetic basis of various complex traits and ASD.

show abstract

Fine-tuning Polygenic Risk Scores with GWAS Summary Statistics

Cited by 15 publications

References 43 publications

Improved genetic prediction of complex traits from individual-level data or summary statistics

Improved genetic prediction of complex traits from individual-level data or summary statistics

Cerebrospinal fluid metabolomics identifies 19 brain-related phenotype associations

Gamete simulation improves polygenic transmission disequilibrium analysis

Contact Info

Product

Resources

About