Polygenic risk prediction: why and when out-of-sample prediction R2 can exceed SNP-based heritability

Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium

doi:10.1016/j.ajhg.2023.06.006

Polygenic risk prediction: why and when out-of-sample prediction R² can exceed SNP-based heritability

Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium

Vagelos College of Physicians and Surgeons

Producción científica › revisión exhaustiva

5 Citas (Scopus)

Resumen

In polygenic score (PGS) analysis, the coefficient of determination (R²) is a key statistic to evaluate efficacy. R² is the proportion of phenotypic variance explained by the PGS, calculated in a cohort that is independent of the genome-wide association study (GWAS) that provided estimates of allelic effect sizes. The SNP-based heritability (h_SNP², the proportion of total phenotypic variances attributable to all common SNPs) is the theoretical upper limit of the out-of-sample prediction R². However, in real data analyses R² has been reported to exceed h_SNP², which occurs in parallel with the observation that h_SNP² estimates tend to decline as the number of cohorts being meta-analyzed increases. Here, we quantify why and when these observations are expected. Using theory and simulation, we show that if heterogeneities in cohort-specific h_SNP² exist, or if genetic correlations between cohorts are less than one, h_SNP² estimates can decrease as the number of cohorts being meta-analyzed increases. We derive conditions when the out-of-sample prediction R² will be greater than h_SNP² and show the validity of our derivations with real data from a binary trait (major depression) and a continuous trait (educational attainment). Our research calls for a better approach to integrating information from multiple cohorts to address issues of between-cohort heterogeneity.

Idioma original	English
Páginas (desde-hasta)	1207-1215
Número de páginas	9
Publicación	American Journal of Human Genetics
Volumen	110
N.º	7
DOI	https://doi.org/10.1016/j.ajhg.2023.06.006
Estado	Published - jul. 6 2023

Financiación

We acknowledge funding from the Australian National Health & Medical Research Council ( 1173790 , 1113400 ), Australian Research Council ( FL180100072 ), and the National Institute of Mental Health ( R01MH124871 , R01MH121545 ). This work would not have been possible without the contributions of the investigators who comprise the PGC-MDD working group. The procedures followed in the PGC-MDD working group were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and that proper informed consent was obtained. Data analysis was conducted under the University of Queensland Human Research Ethics Committee approval HE002938 . For a full list of acknowledgments and ethical statements of all individual cohorts, please see the original publications. The PGC has received major funding from the US National Institute of Mental Health and the US National Institute on Drug Abuse ( U01 MH109528 and U01 MH1095320 ). Some statistical analyses were carried out on the NL Genetic Cluster Computer ( http://www.geneticcluster.org/ ) hosted by SURFsara who support the PGC through grants to Danielle Posthuma. GWAS summary statistics from 23andMe were included in the meta-analyzed GWAS summary statistics. We thank the customers, research participants, and employees of 23andMe for making this work possible. The study protocol used by 23andMe was approved by an external AAHRPP-accredited institutional review board. The graphical abstract was created with BioRender.com . We acknowledge funding from the Australian National Health & Medical Research Council (1173790, 1113400), Australian Research Council (FL180100072), and the National Institute of Mental Health (R01MH124871, R01MH121545). This work would not have been possible without the contributions of the investigators who comprise the PGC-MDD working group. The procedures followed in the PGC-MDD working group were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and that proper informed consent was obtained. Data analysis was conducted under the University of Queensland Human Research Ethics Committee approval HE002938. For a full list of acknowledgments and ethical statements of all individual cohorts, please see the original publications. The PGC has received major funding from the US National Institute of Mental Health and the US National Institute on Drug Abuse (U01 MH109528 and U01 MH1095320). Some statistical analyses were carried out on the NL Genetic Cluster Computer (http://www.geneticcluster.org/) hosted by SURFsara who support the PGC through grants to Danielle Posthuma. GWAS summary statistics from 23andMe were included in the meta-analyzed GWAS summary statistics. We thank the customers, research participants, and employees of 23andMe for making this work possible. The study protocol used by 23andMe was approved by an external AAHRPP-accredited institutional review board. The graphical abstract was created with BioRender.com. Study motivation, N.R.W. A.M.McI.; theory, X.W. P.M.V. N.R.W.; simulations & analyses, X.W.; data preparation, A.W. J.A.R. G.N. M.J.A.; first draft, X.W. N.R.W. P.M.V.; final draft, all authors read and approved the manuscript. The authors declare no competing interests.

Financiadores	Número del financiador
National Institute of Mental Health	R01MH121545, R01MH124871, HE002938
National Institute on Drug Abuse	U01 MH1095320, U01 MH109528
Pennsylvania Game Commission
Australian Research Council	FL180100072
National Health and Medical Research Council	1113400, 1173790

ASJC Scopus Subject Areas

Genetics
Genetics(clinical)

Acceder al documento

10.1016/j.ajhg.2023.06.006

Otros archivos y enlaces

Citar esto

@article{850351cfecfb4c298c553148d9c638aa,

title = "Polygenic risk prediction: why and when out-of-sample prediction R2 can exceed SNP-based heritability",

abstract = "In polygenic score (PGS) analysis, the coefficient of determination (R2) is a key statistic to evaluate efficacy. R2 is the proportion of phenotypic variance explained by the PGS, calculated in a cohort that is independent of the genome-wide association study (GWAS) that provided estimates of allelic effect sizes. The SNP-based heritability (hSNP2, the proportion of total phenotypic variances attributable to all common SNPs) is the theoretical upper limit of the out-of-sample prediction R2. However, in real data analyses R2 has been reported to exceed hSNP2, which occurs in parallel with the observation that hSNP2 estimates tend to decline as the number of cohorts being meta-analyzed increases. Here, we quantify why and when these observations are expected. Using theory and simulation, we show that if heterogeneities in cohort-specific hSNP2 exist, or if genetic correlations between cohorts are less than one, hSNP2 estimates can decrease as the number of cohorts being meta-analyzed increases. We derive conditions when the out-of-sample prediction R2 will be greater than hSNP2 and show the validity of our derivations with real data from a binary trait (major depression) and a continuous trait (educational attainment). Our research calls for a better approach to integrating information from multiple cohorts to address issues of between-cohort heterogeneity.",

author = "{Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium} and Xiaotong Wang and Alicia Walker and Revez, {Joana A.} and Guiyan Ni and Adams, {Mark J.} and McIntosh, {Andrew M.} and Wray, {Naomi R.} and Stephan Ripke and Manuel Mattheisen and Maciej Trzaskowski and Byrne, {Enda M.} and Abdel Abdellaoui and Esben Agerbo and Air, {Tracy M.} and Andlauer, {Till F.M.} and Bacanu, {Silviu Alin} and Marie B{\ae}kvad-Hansen and Beekman, {Aartjan T.F.} and Bigdeli, {Tim B.} and Binder, {Elisabeth B.} and Julien Bryois and Buttensch{\o}n, {Henriette N.} and Jonas Bybjerg-Grauholm and Na Cai and Enrique Castelao and Christensen, {Jane Hvarregaard} and Clarke, {Toni Kim} and Coleman, {Jonathan R.I.} and Luc{\'i}a Colodro-Conde and Baptiste Couvy-Duchesne and Nick Craddock and Crawford, {Gregory E.} and Gail Davies and Franziska Degenhardt and Derks, {Eske M.} and Nese Direk and Dolan, {Conor V.} and Dunn, {Erin C.} and Eley, {Thalia C.} and Valentina Escott-Price and Kiadeh, {Farnush Farhadi Hassan} and Finucane, {Hilary K.} and Foo, {Jerome C.} and Forstner, {Andreas J.} and Josef Frank and Gaspar, {H{\'e}l{\'e}na A.} and Michael Gill and Goes, {Fernando S.} and Gordon, {Scott D.} and Weissman, {Myrna M.}",

note = "Publisher Copyright: {\textcopyright} 2023 American Society of Human Genetics",

year = "2023",

month = jul,

day = "6",

doi = "10.1016/j.ajhg.2023.06.006",

language = "English",

volume = "110",

pages = "1207--1215",

journal = "American Journal of Human Genetics",

issn = "0002-9297",

publisher = "Cell Press",

number = "7",

}

TY - JOUR

T1 - Polygenic risk prediction

T2 - why and when out-of-sample prediction R2 can exceed SNP-based heritability

AU - Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium

AU - Wang, Xiaotong

AU - Walker, Alicia

AU - Revez, Joana A.

AU - Ni, Guiyan

AU - Adams, Mark J.

AU - McIntosh, Andrew M.

AU - Wray, Naomi R.

AU - Ripke, Stephan

AU - Mattheisen, Manuel

AU - Trzaskowski, Maciej

AU - Byrne, Enda M.

AU - Abdellaoui, Abdel

AU - Agerbo, Esben

AU - Air, Tracy M.

AU - Andlauer, Till F.M.

AU - Bacanu, Silviu Alin

AU - Bækvad-Hansen, Marie

AU - Beekman, Aartjan T.F.

AU - Bigdeli, Tim B.

AU - Binder, Elisabeth B.

AU - Bryois, Julien

AU - Buttenschøn, Henriette N.

AU - Bybjerg-Grauholm, Jonas

AU - Cai, Na

AU - Castelao, Enrique

AU - Christensen, Jane Hvarregaard

AU - Clarke, Toni Kim

AU - Coleman, Jonathan R.I.

AU - Colodro-Conde, Lucía

AU - Couvy-Duchesne, Baptiste

AU - Craddock, Nick

AU - Crawford, Gregory E.

AU - Davies, Gail

AU - Degenhardt, Franziska

AU - Derks, Eske M.

AU - Direk, Nese

AU - Dolan, Conor V.

AU - Dunn, Erin C.

AU - Eley, Thalia C.

AU - Escott-Price, Valentina

AU - Kiadeh, Farnush Farhadi Hassan

AU - Finucane, Hilary K.

AU - Foo, Jerome C.

AU - Forstner, Andreas J.

AU - Frank, Josef

AU - Gaspar, Héléna A.

AU - Gill, Michael

AU - Goes, Fernando S.

AU - Gordon, Scott D.

AU - Weissman, Myrna M.

PY - 2023/7/6

Y1 - 2023/7/6

N2 - In polygenic score (PGS) analysis, the coefficient of determination (R2) is a key statistic to evaluate efficacy. R2 is the proportion of phenotypic variance explained by the PGS, calculated in a cohort that is independent of the genome-wide association study (GWAS) that provided estimates of allelic effect sizes. The SNP-based heritability (hSNP2, the proportion of total phenotypic variances attributable to all common SNPs) is the theoretical upper limit of the out-of-sample prediction R2. However, in real data analyses R2 has been reported to exceed hSNP2, which occurs in parallel with the observation that hSNP2 estimates tend to decline as the number of cohorts being meta-analyzed increases. Here, we quantify why and when these observations are expected. Using theory and simulation, we show that if heterogeneities in cohort-specific hSNP2 exist, or if genetic correlations between cohorts are less than one, hSNP2 estimates can decrease as the number of cohorts being meta-analyzed increases. We derive conditions when the out-of-sample prediction R2 will be greater than hSNP2 and show the validity of our derivations with real data from a binary trait (major depression) and a continuous trait (educational attainment). Our research calls for a better approach to integrating information from multiple cohorts to address issues of between-cohort heterogeneity.

AB - In polygenic score (PGS) analysis, the coefficient of determination (R2) is a key statistic to evaluate efficacy. R2 is the proportion of phenotypic variance explained by the PGS, calculated in a cohort that is independent of the genome-wide association study (GWAS) that provided estimates of allelic effect sizes. The SNP-based heritability (hSNP2, the proportion of total phenotypic variances attributable to all common SNPs) is the theoretical upper limit of the out-of-sample prediction R2. However, in real data analyses R2 has been reported to exceed hSNP2, which occurs in parallel with the observation that hSNP2 estimates tend to decline as the number of cohorts being meta-analyzed increases. Here, we quantify why and when these observations are expected. Using theory and simulation, we show that if heterogeneities in cohort-specific hSNP2 exist, or if genetic correlations between cohorts are less than one, hSNP2 estimates can decrease as the number of cohorts being meta-analyzed increases. We derive conditions when the out-of-sample prediction R2 will be greater than hSNP2 and show the validity of our derivations with real data from a binary trait (major depression) and a continuous trait (educational attainment). Our research calls for a better approach to integrating information from multiple cohorts to address issues of between-cohort heterogeneity.

UR - http://www.scopus.com/inward/record.url?scp=85164270154&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85164270154&partnerID=8YFLogxK

U2 - 10.1016/j.ajhg.2023.06.006

DO - 10.1016/j.ajhg.2023.06.006

M3 - Article

C2 - 37379836

AN - SCOPUS:85164270154

SN - 0002-9297

VL - 110

SP - 1207

EP - 1215

JO - American Journal of Human Genetics

JF - American Journal of Human Genetics

IS - 7

ER -