Post by Jian Yang on Jun 11, 2015 0:56:58 GMT
--cojo-file test.ma
Input the summary-level statistics from a meta-analysis GWAS (or a single GWAS).
Input file format
test.ma
Columns are SNP, the effect allele, the other allele, frequency of the effect allele, effect size, standard error, p-value and sample size. The headers are not keywords and will be omitted by the program. Important: “A1” needs to be the effect allele with “A2” being the other allele and “freq” should be the frequency of “A1”.
NOTE: 1) For a case-control study, the effect size should be log(odds ratio) with its corresponding standard error. 2) Please always input the summary statistics of all the SNPs even if your analysis only focuses on a subset of SNPs because the program needs the summary data of all SNPs to calculate the phenotypic variance.
--cojo-slct
Perform a stepwise model selection procedure to select independently associated SNPs. Results will be saved in a *.jma file with additional file
*.jma.ldr showing the LD correlations between the SNPs.
--cojo-top-SNPs 10
Perform a stepwise model selection procedure to select a fixed number of independently associated SNPs without a p-value threshold. The output format is the same as that from --cojo-slct.
--cojo-joint
Fit all the included SNPs to estimate their joint effects without model selection. Results will be saved in a *.jma file with additional file *.jma.ldr showing the LD correlations between the SNPs.
--cojo-cond cond.snplist
Perform association analysis of the included SNPs conditional on the given list of SNPs. Results will be saved in a *.cma.
Input file format
cond.snplist
--cojo-p 5e-8
Threshold p-value to declare a genome-wide significant hit. The default value is 5e-8 if not specified. This option is only valid in conjunction with the option --cojo-slct. NOTE: it will be extremely time-consuming if you set a very low significance level, e.g. 5e-3.
--cojo-wind 10000
Specify a distance d (in Kb unit). It is assumed that SNPs more than d Kb away from each other are in complete linkage equilibrium. The default value is 10000 Kb (i.e. 10 Mb) if not specified.
--cojo-collinear 0.9
During the model selection procedure, the program will check the collinearity between the SNPs that have already been selected and a SNP to be tested. The testing SNP will not be selected if its multiple regression R2 on the selected SNPs is greater than the cutoff value. By default, the cutoff value is 0.9 if not specified.
--cojo-gc
If this option is specified, p-values will be adjusted by the genomic control method. By default, the genomic inflation factor will be calculated from the summary-level statistics of all the SNPs unless you specify a value, e.g. --cojo-gc 1.05.
--cojo-actual-geno
If the individual-level genotype data of the discovery set are available (e.g. a single-cohort GWAS), you can use the discovery set as the reference sample. In this case, the analysis will be equivalent to a multiple regression analysis with the actual genotype and phenotype data. Once this option is specified, GCTA will take all pairwise LD correlations between all SNPs into account, which overrides the –cojo-wind option. This option also allows GCTA to calculate the variance taken out from the residual variance by all the significant SNPs in the model, otherwise the residual variance will be fixed constant at the same level of the phenotypic variance.
Examples (Individual-level genotype data of the discovery set is NOT available) - Robust and recommended
# Select multiple associated SNPs through a stepwise selection procedure
# Select a fixed number of of top associated SNPs through a stepwise selection procedure
# Estimate the joint effects of a subset of SNPs (given in the file test.snplist) without model selection
# Perform single-SNP association analyses conditional on a set of SNPs (given in the file cond.snplist) without model selection
It should be more efficient to separate the analysis onto individual chromosomes or even some particular genomic regions. Please refer to the Data management section for some other options, e.g. including or excluding a list of SNPs and individuals or filtering SNPs based on the imputation quality score.
Examples (Individual-level genotype data of the discovery set is available)
# Select multiple associated SNPs through a stepwise selection procedure
In this case, it is recommended to perform the analysis using the data of all the genome-wide SNPs rather than separate the analysis onto individual chromosomes because GCTA needs to calculate the variance taken out from the residual variance by all the significant SNPs in the model, which could give you a bit more power.
# Estimate the joint effects of a subset of SNPs (given in the file test.snplist) without model selection
# Perform single-SNP association analyses conditional on a set of SNPs (given in the file cond.snplist) without model selection
Output file format
test.jma (generate by the option --cojo-slct or --cojo-joint)
Columns are chromosome; SNP; physical position; frequency of the effect allele in the original data; the effect allele; effect size, standard error and p-value from the original GWAS or meta-analysis; estimated effective sample size; frequency of the effect allele in the reference sample; effect size, standard error and p-value from a joint analysis of all the selected SNPs; LD correlation between the SNP i and SNP i + 1 for the SNPs on the list.
LD correlation matrix between all pairwise SNPs listed in test.jma.
test.jma.ldr (generate by the option --cojo-slct or --cojo-joint)
test.cma (generate by the option --cojo-slct or --cojo-cond)
Columns are chromosome; SNP; physical position; frequency of the effect allele in the original data; the effect allele; effect size, standard error and p-value from the original GWAS or meta-analysis; estimated effective sample size; frequency of the effect allele in the reference sample; effect size, standard error and p-value from conditional analyses.
References
Conditional and joint analysis method: Yang et al. (2012) Conditional and joint multiple-SNP analysis of GWAS summary statistics identifies additional variants influencing complex traits. Nat Genet 44(4):369-375. [PubMed ID: 22426310]
GCTA software: Yang J, Lee SH, Goddard ME and Visscher PM. GCTA: a tool for Genome-wide Complex Trait Analysis. Am J Hum Genet. 2011 Jan 88(1): 76-82. [PubMed ID: 21167468]
Input the summary-level statistics from a meta-analysis GWAS (or a single GWAS).
Input file format
test.ma
SNP A1 A2 freq b se p N
rs1001 A G 0.8493 0.0024 0.0055 0.6653 129850
rs1002 C G 0.0306 0.0034 0.0115 0.7659 129799
rs1003 A C 0.5128 0.0045 0.0038 0.2319 129830
...
Columns are SNP, the effect allele, the other allele, frequency of the effect allele, effect size, standard error, p-value and sample size. The headers are not keywords and will be omitted by the program. Important: “A1” needs to be the effect allele with “A2” being the other allele and “freq” should be the frequency of “A1”.
NOTE: 1) For a case-control study, the effect size should be log(odds ratio) with its corresponding standard error. 2) Please always input the summary statistics of all the SNPs even if your analysis only focuses on a subset of SNPs because the program needs the summary data of all SNPs to calculate the phenotypic variance.
--cojo-slct
Perform a stepwise model selection procedure to select independently associated SNPs. Results will be saved in a *.jma file with additional file
*.jma.ldr showing the LD correlations between the SNPs.
--cojo-top-SNPs 10
Perform a stepwise model selection procedure to select a fixed number of independently associated SNPs without a p-value threshold. The output format is the same as that from --cojo-slct.
--cojo-joint
Fit all the included SNPs to estimate their joint effects without model selection. Results will be saved in a *.jma file with additional file *.jma.ldr showing the LD correlations between the SNPs.
--cojo-cond cond.snplist
Perform association analysis of the included SNPs conditional on the given list of SNPs. Results will be saved in a *.cma.
Input file format
cond.snplist
rs1001
rs1002
...
--cojo-p 5e-8
Threshold p-value to declare a genome-wide significant hit. The default value is 5e-8 if not specified. This option is only valid in conjunction with the option --cojo-slct. NOTE: it will be extremely time-consuming if you set a very low significance level, e.g. 5e-3.
--cojo-wind 10000
Specify a distance d (in Kb unit). It is assumed that SNPs more than d Kb away from each other are in complete linkage equilibrium. The default value is 10000 Kb (i.e. 10 Mb) if not specified.
--cojo-collinear 0.9
During the model selection procedure, the program will check the collinearity between the SNPs that have already been selected and a SNP to be tested. The testing SNP will not be selected if its multiple regression R2 on the selected SNPs is greater than the cutoff value. By default, the cutoff value is 0.9 if not specified.
--cojo-gc
If this option is specified, p-values will be adjusted by the genomic control method. By default, the genomic inflation factor will be calculated from the summary-level statistics of all the SNPs unless you specify a value, e.g. --cojo-gc 1.05.
--cojo-actual-geno
If the individual-level genotype data of the discovery set are available (e.g. a single-cohort GWAS), you can use the discovery set as the reference sample. In this case, the analysis will be equivalent to a multiple regression analysis with the actual genotype and phenotype data. Once this option is specified, GCTA will take all pairwise LD correlations between all SNPs into account, which overrides the –cojo-wind option. This option also allows GCTA to calculate the variance taken out from the residual variance by all the significant SNPs in the model, otherwise the residual variance will be fixed constant at the same level of the phenotypic variance.
Examples (Individual-level genotype data of the discovery set is NOT available) - Robust and recommended
# Select multiple associated SNPs through a stepwise selection procedure
gcta64 --bfile test --chr 1 --maf 0.01 --cojo-file test.ma --cojo-slct --out test_chr1
# Select a fixed number of of top associated SNPs through a stepwise selection procedure
gcta64 --bfile test --chr 1 --maf 0.01 --cojo-file test.ma --cojo-top-SNPs 10 --out test_chr1
# Estimate the joint effects of a subset of SNPs (given in the file test.snplist) without model selection
gcta64 --bfile test --chr 1 --extract test.snplist --cojo-file test.ma --cojo-joint --out test_chr1
# Perform single-SNP association analyses conditional on a set of SNPs (given in the file cond.snplist) without model selection
gcta64 --bfile test --chr 1 --maf 0.01 --cojo-file test.ma --cojo-cond cond.snplist --out test_chr1
It should be more efficient to separate the analysis onto individual chromosomes or even some particular genomic regions. Please refer to the Data management section for some other options, e.g. including or excluding a list of SNPs and individuals or filtering SNPs based on the imputation quality score.
Examples (Individual-level genotype data of the discovery set is available)
# Select multiple associated SNPs through a stepwise selection procedure
gcta64 --bfile test --maf 0.01 --cojo-file test.ma --cojo-slct --cojo-actual-geno --out test
In this case, it is recommended to perform the analysis using the data of all the genome-wide SNPs rather than separate the analysis onto individual chromosomes because GCTA needs to calculate the variance taken out from the residual variance by all the significant SNPs in the model, which could give you a bit more power.
# Estimate the joint effects of a subset of SNPs (given in the file test.snplist) without model selection
gcta64 --bfile test --extract test.snplist --cojo-file test.ma --cojo-actual-geno --cojo-joint --out test
# Perform single-SNP association analyses conditional on a set of SNPs (given in the file cond.snplist) without model selection
gcta64 --bfile test --maf 0.01 --cojo-file test.ma --cojo-actual-geno --cojo-cond cond.snplist --out test
Output file format
test.jma (generate by the option --cojo-slct or --cojo-joint)
Chr SNP bp freq refA b se p n freq_geno bJ bJ_se pJ LD_r
1 rs2001 172585028 0.6105 A 0.0377 0.0042 6.38e-19 121056 0.614 0.0379 0.0042 1.74e-19 -0.345
1 rs2002 174763990 0.4294 C 0.0287 0.0041 3.65e-12 124061 0.418 0.0289 0.0041 1.58e-12 0.012
1 rs2003 196696685 0.5863 T 0.0237 0.0042 1.38e-08 116314 0.589 0.0237 0.0042 1.67e-08 0.0
...
Columns are chromosome; SNP; physical position; frequency of the effect allele in the original data; the effect allele; effect size, standard error and p-value from the original GWAS or meta-analysis; estimated effective sample size; frequency of the effect allele in the reference sample; effect size, standard error and p-value from a joint analysis of all the selected SNPs; LD correlation between the SNP i and SNP i + 1 for the SNPs on the list.
LD correlation matrix between all pairwise SNPs listed in test.jma.
test.jma.ldr (generate by the option --cojo-slct or --cojo-joint)
SNP rs2001 rs2002 rs2003 ...
rs2001 1 0.0525 -0.0672 ...
rs2002 0.0525 1 0.0045 ...
rs2003 -0.0672 0.0045 1 ...
...
test.cma (generate by the option --cojo-slct or --cojo-cond)
Chr SNP bp freq refA b se p n freq_geno bC bC_se pC
1 rs2001 172585028 0.6105 A 0.0377 0.0042 6.38e-19 121056 0.614 0.0379 0.0042 1.74e-19
1 rs2002 174763990 0.4294 C 0.0287 0.0041 3.65e-12 124061 0.418 0.0289 0.0041 1.58e-12
1 rs2003 196696685 0.5863 T 0.0237 0.0042 1.38e-08 116314 0.589 0.0237 0.0042 1.67e-08
...
Columns are chromosome; SNP; physical position; frequency of the effect allele in the original data; the effect allele; effect size, standard error and p-value from the original GWAS or meta-analysis; estimated effective sample size; frequency of the effect allele in the reference sample; effect size, standard error and p-value from conditional analyses.
References
Conditional and joint analysis method: Yang et al. (2012) Conditional and joint multiple-SNP analysis of GWAS summary statistics identifies additional variants influencing complex traits. Nat Genet 44(4):369-375. [PubMed ID: 22426310]
GCTA software: Yang J, Lee SH, Goddard ME and Visscher PM. GCTA: a tool for Genome-wide Complex Trait Analysis. Am J Hum Genet. 2011 Jan 88(1): 76-82. [PubMed ID: 21167468]