TY - JOUR
T1 - A Comparison of Ten Polygenic Score Methods for Psychiatric Disorders Applied Across Multiple Cohorts
AU - Schizophrenia Working Group of the Psychiatric Genomics Consortium
AU - Major Depressive Disorder Working Group of the Psychiatric Genomics Consortium
AU - Ni, Guiyan
AU - Zeng, Jian
AU - Revez, Joana A.
AU - Wang, Ying
AU - Zheng, Zhili
AU - Ge, Tian
AU - Restuadi, Restuadi
AU - Kiewa, Jacqueline
AU - Nyholt, Dale R.
AU - Coleman, Jonathan R.I.
AU - Smoller, Jordan W.
AU - Ripke, Stephan
AU - Neale, Benjamin M.
AU - Corvin, Aiden
AU - Walters, James T.R.
AU - Farh, Kai How
AU - Holmans, Peter A.
AU - Lee, Phil
AU - Bulik-Sullivan, Brendan
AU - Collier, David A.
AU - Huang, Hailiang
AU - Pers, Tune H.
AU - Agartz, Ingrid
AU - Agerbo, Esben
AU - Albus, Margot
AU - Alexander, Madeline
AU - Amin, Farooq
AU - Bacanu, Silviu A.
AU - Begemann, Martin
AU - Belliveau, Richard A.
AU - Bene, Judit
AU - Bergen, Sarah E.
AU - Bevilacqua, Elizabeth
AU - Bigdeli, Tim B.
AU - Black, Donald W.
AU - Bruggeman, Richard
AU - Buccola, Nancy G.
AU - Buckner, Randy L.
AU - Byerley, William
AU - Cahn, Wiepke
AU - Cai, Guiqing
AU - Campion, Dominique
AU - Cantor, Rita M.
AU - Carr, Vaughan J.
AU - Carrera, Noa
AU - Catts, Stanley V.
AU - Svrakic, Dragan M.
AU - Rice, John P.
AU - Heath, Andrew C.
AU - Madden, Pamela A.F.
N1 - Publisher Copyright:
© 2021 Society of Biological Psychiatry
PY - 2021/11/1
Y1 - 2021/11/1
N2 - Background: Polygenic scores (PGSs), which assess the genetic risk of individuals for a disease, are calculated as a weighted count of risk alleles identified in genome-wide association studies. PGS methods differ in which DNA variants are included and the weights assigned to them; some require an independent tuning sample to help inform these choices. PGSs are evaluated in independent target cohorts with known disease status. Variability between target cohorts is observed in applications to real data sets, which could reflect a number of factors, e.g., phenotype definition or technical factors. Methods: The Psychiatric Genomics Consortium Working Groups for schizophrenia and major depressive disorder bring together many independently collected case-control cohorts. We used these resources (31,328 schizophrenia cases, 41,191 controls; 248,750 major depressive disorder cases, 563,184 controls) in repeated application of leave-one-cohort-out meta-analyses, each used to calculate and evaluate PGS in the left-out (target) cohort. Ten PGS methods (the baseline PC+T method and 9 methods that model genetic architecture more formally: SBLUP, LDpred2-Inf, LDpred-funct, LDpred2, Lassosum, PRS-CS, PRS-CS-auto, SBayesR, MegaPRS) were compared. Results: Compared with PC+T, the other 9 methods gave higher prediction statistics, MegaPRS, LDPred2, and SBayesR significantly so, explaining up to 9.2% variance in liability for schizophrenia across 30 target cohorts, an increase of 44%. For major depressive disorder across 26 target cohorts, these statistics were 3.5% and 59%, respectively. Conclusions: Although the methods that more formally model genetic architecture have similar performance, MegaPRS, LDpred2, and SBayesR rank highest in most comparisons and are recommended in applications to psychiatric disorders.
AB - Background: Polygenic scores (PGSs), which assess the genetic risk of individuals for a disease, are calculated as a weighted count of risk alleles identified in genome-wide association studies. PGS methods differ in which DNA variants are included and the weights assigned to them; some require an independent tuning sample to help inform these choices. PGSs are evaluated in independent target cohorts with known disease status. Variability between target cohorts is observed in applications to real data sets, which could reflect a number of factors, e.g., phenotype definition or technical factors. Methods: The Psychiatric Genomics Consortium Working Groups for schizophrenia and major depressive disorder bring together many independently collected case-control cohorts. We used these resources (31,328 schizophrenia cases, 41,191 controls; 248,750 major depressive disorder cases, 563,184 controls) in repeated application of leave-one-cohort-out meta-analyses, each used to calculate and evaluate PGS in the left-out (target) cohort. Ten PGS methods (the baseline PC+T method and 9 methods that model genetic architecture more formally: SBLUP, LDpred2-Inf, LDpred-funct, LDpred2, Lassosum, PRS-CS, PRS-CS-auto, SBayesR, MegaPRS) were compared. Results: Compared with PC+T, the other 9 methods gave higher prediction statistics, MegaPRS, LDPred2, and SBayesR significantly so, explaining up to 9.2% variance in liability for schizophrenia across 30 target cohorts, an increase of 44%. For major depressive disorder across 26 target cohorts, these statistics were 3.5% and 59%, respectively. Conclusions: Although the methods that more formally model genetic architecture have similar performance, MegaPRS, LDpred2, and SBayesR rank highest in most comparisons and are recommended in applications to psychiatric disorders.
KW - LDpred2
KW - Lassosum
KW - Major depressive disorder
KW - MegaPRS
KW - PRS-CS
KW - Polygenic scores
KW - Psychiatric disorders
KW - Risk prediction
KW - SBayesR
KW - Schizophrenia
UR - http://www.scopus.com/inward/record.url?scp=85107783986&partnerID=8YFLogxK
U2 - 10.1016/j.biopsych.2021.04.018
DO - 10.1016/j.biopsych.2021.04.018
M3 - Article
C2 - 34304866
AN - SCOPUS:85107783986
SN - 0006-3223
VL - 90
SP - 611
EP - 620
JO - Biological Psychiatry
JF - Biological Psychiatry
IS - 9
ER -