User Manual - Multi-Trait GWAS Simulation Software
Contents
1. below In the simplest case the parameters you will need to specify are e sample size n e number of SNP replicates m minor allele frequency of the SNP MAF e number of phenotypes k e genetic effects as a percentage of phenotypic variance explained v e phenotype correlation matrix cor mat The simulated SNPs are independent the m parameter specifies the number of replicates of a SNP to simulate The more replicates the less variance in the power estimates The phenotype data are simulated multivariate normal variables with multivariate normal error term whose variance is the phenotype correlation matrix This means that the simulated phenotype data correlation is not exactly as specified but the same to within 2 d p To simulate data on 2 phenotypes with correlation 0 45 and phenotypic variance explained of 0 2 on phenotype 1 and 0 3 of phenotype 2 for 1000 individuals and 100 replicates of a SNP with minor allele frequency 0 1 use the following command R file MultiTraitGWAS R q args n 1000 m 100 k 2 MAF 0 1 v 0 002 0 003 cor mat 0 45 strand data A folder data_output_date_time is created containing 100 SNP data files snp_data_i and 100 phenotype data files pheno_data_i There is also the data_output_log file that details the parameters and the run time for the simulation Example first 10 lines of a SNP data file rsi OOrRrOCCOrR OF WN Example first 10 lines of a phenotype data file Phenol1 P2. of two phenotypes from the 12 of which there are 66 combinations N B Since S4b uses real SNP data comparisons may not be able to be performed on the exact number of SNPs specified without removing SNPs Instead comparisons are performed on the closest possible number of SNPs 6 7 Additional Options If you would like to perform comparisons across multiple scenarios you should provide the strand argument with more than one scenario For example to run 1 4a and S4b specify strand S1 S54a S4b Make sure that you have specified the required parameters for all scenarios In addition to obtaining power estimates and P values log p Bayes factors you can get the software to output the simulated data used in the analysis To do this you should specify data out TRUE By default the P value threshold is set to be the genome wide significance threshold P 5 x 1078 and the logio Bayes factor threshold is set to be 6 These thresholds can be changed by defining the p val threshold and log10 bf threshold parameters for example p val threshold 0 0000005 logi0 bf threshold 4 11 If you want to run the comparisons for all methods except two say MANOVA and min P then you don t need to list all eight other methods for the methods parameter Instead if you specify methods MANOVA min P the comparison will be performed for all methods except MANOVA and min P 6 8 Output A folder with name in the format strand_output_dat
3. phenotypes k phenotypic variance explained v and the methods to compare methods Example R file MultiTraitGWAS R q args n 5000 m 100 k 2 MAF 0 3 v 0 005 0 001 methods min P CCA strand S4a The above command compares the min P and CCA methods for two phenotypes 100 SNPs 5000 individuals minor allele frequency 0 3 and phenotypic variance explained by SNP of 0 5 and 0 1 respectively 10 6 6 S4b real data informed genetic effects and phenotypic correlations Scenario S4b in addition uses real data to inform the SNP effects variance explained to run comparisons for S4b you should provide the sample size n number of SNPs m number of phenotypes k and the methods to compare methods N B Currently the maximum number of phenotypes for this scenario is 12 but this will likely increase in the future when more summary data is incorporated Example R file MultiTraitGWAS R q args n 5000 m 1000 k 12 methods min P CCA strand S4b The above command compares the min P and CCA methods for 12 phenotypes 5000 individ uals and 1000 SNPs under the S4b framework for more details see our paper In addition S4b allows methods to be compared on combinations of the 12 phenotypes To do this set k equal to any positive integer less than 12 Example R file MultiTraitGWAS R q args n 5000 m 1000 k 2 methods min P CCA strand S4b The above command performs the S4b comparison on every combination
4. Multi trait GWAS Simulator User Manual Heather F Porter amp Paul F O Reilly multitraitgwasQ gmail com MRC Social Genetic and Developmental Psychiatry Centre Institute of Psychiatry Psychology amp Neuroscience King s College London UK Contents Background 3 Software program R packages 4 Quick start 5 A Simulating multivariate data 7 5 1 Phenotypes with direct SNP effects 2 0 0 0 0 00 ee ee 7 5 2 Phenotypes with indirect SNP effects 0 0 0 0 0 00000 8 5 3 Case control phenotypes a 8 5 4 Output 2 444 8 eb bee ee ee eee dae Eee Dee Re eR Es 8 6 Multivariate method comparison 9 6 1 Methods 9 peana ai 9 ph tein shee St eae 9 eA hn aoe eh a 10 eee ee 10 ores 11 e A 11 6 8 Output cx seeds aa ek e eS 12 1 Background Many genetic analyses are now performed on multiple phenotypes jointly e g to increase discov ery power in genome wide association studies GWAS for sub phenotype and cross phenotype prediction using polygenic risk scores or to estimate genetic correlations among traits However estimating the power of the related methods and comparing their performance is challenging due to the complexity of the biological network introduced when considering multiple pheno types We have produced a simulation framework in order to address this the framework summarised below models this network explicitly enabling the vast model
5. e combination number Again when m is larger than the number of SNPs for that combination multiple iterations will be performed to estimate the power and for each combination there will be a file for every iteration performed SNP_b_MAF_k_n comb_it If you requested the simulated data to be output then a folder called data will be created within this directory this folder will contain m SNP data files snp_data_i for i 1 m and m phenotype data files pheno_data_i for i 1 m For S4b when k 12 the SNP and phenotype data filenames instead end with _it_n snp where n snp ranges from 1 to the total number of SNPs for the 12 phenotypes for that iteration When k 12 the SNP and phenotype data filenames end with _k_n comb_it_n snp where n snp ranges from 1 to the total number of SNPs for that combination of phenotypes for that iteration N B SNP numbers can vary between iterations as they are sampled randomly each time there is a limit imposed so that there are no more than 20 SNPs associated with any one phenotype in order to not bias the results toward phenotypes with lots of SNP hits N B For some phenotypes there are as little as two SNPs associated and so you may find that for some of the combinations of phenotypes there are no SNPs for which there is complete SNP effect data In this case there will be no SNP effect data file for that combination 12
6. e_time is created containing the output files for each comparison analysis performed if multiple scenarios are selected a separate folder will be created for each In each folder there will be an output log file strand_output_log which details the parameter settings as well as the run time for the analysis and the power cal culation results file with one column per method strand_power For each method compared there will be a separate file containing the P values or log Bayes factors BFs for that method where rows represent SNPs strand_p_values_method or strand_log10_bayes_factors method For S1 the P value and log g BF files have multiple columns representing the different phenotypic correlations and the rows in the power results file correspond to these correlations In addition for scenario S4b and when k 12 a file is created which details the SNPs used in the comparison as well as the beta coefficients across all 12 phenotypes and the MAF SNP_b_MAF _12 If mis larger than the number of SNPs for these 12 phenotypes multiple iterations of this comparison will be performed and there will be a file for each iteration with the iteration number at the end of the filename SNP_b_MAF_12_it where it is the itera tion number When k 12 a file is created which details the combinations of phenotypes combinations_k and one file per combination is created detailing which SNPs were included SNP_b MAF k n comb where n comb is th
7. etic effects To run comparisons for scenario S3 of the simulation framework along with the sample size n number of SNPs m minor allele frequency MAF number of phenotypes k phenotypic variance explained v and methods methods you need to specify e genetic correlations default gen cor 0 6 0 2 0 05 Example R file MultiTraitGWAS R q args n 5000 m 100 k 2 MAF 0 3 v 0 005 0 001 gen cor 0 6 0 2 0 05 methods min P CCA strand 53 The above command compares the min P and CCA methods for two phenotypes 100 SNPs 5000 individuals minor allele frequency 0 3 and phenotypic variance explained by SNP of 0 5 and 0 1 respectively The genetic correlations are 0 6 0 2 and 0 05 meaning that if the phenotypic variance explained is the same for both phenotypes their correlation is 0 6 if the variance explained is different but neither are zero the pairwise phenotypic correlation is 0 2 and if one variance explained is zero the pairwise phenotypic correlation is 0 05 For ex ample ifk 4 and v 0 005 0 005 0 001 0 then the phenotypic correlation matrix is given by 1 0 6 0 2 0 05 0 6 1 0 2 0 05 0 2 0 2 1 0 05 0 05 0 05 0 05 1 6 5 S4a real data informed phenotypic correlations Scenario S4a samples the phenotypic correlations from a mixture Gaussian distribution informed by real data to run comparisons for S4a you need to provide the sample size n number of SNPs m minor allele frequency MAF number of
8. heno2 0 826669320195689 0 42083503810129 0 195993062417846 0 328777224652241 0 13776496562821 0 0259098766867293 0 456580144739775 1 2848771006169 1 19281347343584 0 572394773681447 0 173851560588671 0 245754462108122 0 224286770332517 1 47187854347651 1 79286516350429 1 80735748486297 1 1057183510766 0 183142822421996 0 37684498650171 0 229370868124437 Example log file HHTHHHHHHHHHHHHHHHHHHHHHEHHHRHHHHHE HEHEHE RHEE HHHA RHEE HHH HR AA Multi trait GWAS Software Heather F Porter Paul F O Reilly If you use this software in published work please cite HHPHHHHHHHHHHHHHHHHHHHHHEHHRHHHHHHHEHHHHHHHHHHHEHHHRHE RHEE HEHEHE AA Date 27 08 2015 Time 11 58 09 BST HHTHHHHHHEHHHHHHHHHHHEHHHEEHHHHHHHEEH RARA RARA RHEE RHE HHEE HHH RA RES HH Sample size 1000 SNPs 100 MAF 0 1 Phenotypes 2 Effect sizes 0 002 0 003 Correlation s 0 45 HHEHHHHHHHHHEHH HEHEHE HHHHHHHHHHHHHHEHAHAHHHAHEHEHEHHHEHEHEHAHHHHHH HEHEHE Run time 1 18 seconds HHHEHHHHHEHEHH HEHEHE HHHHHEHHHHHHEHEHAHHHHHEHHHEHEHRHAHEHEHHHHHE HEHEHE There are more available options for simulating multivariate data see the Simulating mul tivariate data section but in all cases the output structure remains the same The input parameters are always specified in the same way provided you know the name of the parameter and its input form which is detailed in the Simulating multivariate data section 5 Simulating multivariate data This software
9. n comparisons for scenario S1 of the simulation framework along with the sample size n number of SNPs m minor allele frequency MAF number of phenotypes k and phenotypic variance explained v you need to specify e methods to compare default methods ALL e phenotypic correlations default cor 0 9 0 8 0 9 Example R file MultiTraitGWAS R q args n 5000 m 100 k 2 MAF 0 3 v 0 005 0 001 cor 0 6 0 2 0 1 methods min P CCA strand S1 The above command performs the S1 comparison for the methods min P and CCA for two phe notypes 100 SNPs 5000 individuals minor allele frequency 0 3 phenotypic variance explained by SNP of 0 5 and 0 1 respectively and for phenotypic correlations ranging from 0 6 to 0 2 in increments of 0 1 6 3 S2 uniform genetic effects and phenotypic correlations To run comparisons for scenario S2 of the simulation framework you will need to specify the sample size n number of SNPs m minor allele frequency MAF number of phenotypes k and methods to compare methods The phenotypic variance explained and phenotypic correlations are sampled from the uniform distribution Example R file MultiTraitGWAS R q args n 5000 m 100 k 2 MAF 0 3 methods min P CCA strand S2 The above command runs the S2 comparison on the methods min P and CCA for two pheno types 100 SNPs 5000 individuals and minor allele frequency 0 3 6 4 S3 phenotypic correlations reflective of gen
10. ocation in order to run the software The data files are only required for the final simulation scenario S4b They contain effect size and phenotypic correlation data that is used to inform the simulations These files should be stored in the same location as the R scripts You will also need to download the software corresponding to four of the methods if you wish to perform comparisons with them For SHet and SHom download CPASSOC here make sure that the FunctionSet R script is in the same directory as MultiTraitGWAS R The executable files for BIMBAM download here and SNPTEST download here should be downloaded renamed bimbam and snptest respectively and stored in the same directory as MultiTraitGWAS R This software has been tested for compatibility with SNPTEST v2 5 1 This software is written in R but should be executed from the terminal window The examples in this user manual give the commands to execute from the terminal N B This software makes use of BASH scripting and can currently only be run on UNIX Linux machines 3 R packages This software requires the following R packages and their dependencies e abind e aod e batch e combinat e epitools e Imtest e meta e MultiPhen e Z00 Make sure that the packages are installed in the directory where the software package files are stored 4 Quick start To get a feel for the software program first try simulating some multivariate data using the examples
11. program can generate simulated SNP and multivariate phenotype data SNP and phenotype data can be generated according to the following parameters e sample size n e number of SNP replicates m e minor allele frequency of the SNP MAF e number of phenotypes k e genetic effect vector of phenotypic variance explained v e phenotype correlation matrix cor mat e direct effects default direct TRUE e number of indirect effects n indirect e case control default cc FALSE e case control phenotypes which cc e case prevalence prev 5 1 Phenotypes with direct SNP effects To generate phenotypes with direct SNP effects you need to specify the number of individu als n number of SNPs m and number of phenotypes k for which to generate the data as well as the minor allele frequency of the SNPs MAF the genetic effect vector v of phenotypic variance explained and the phenotypic correlation matrix cor mat in the form 01 2 C1 35 5C1ky Ck 2 k gt Ck 1 k for correlation matrix 1 amp 2 Clik C2 1 1 s C2k Ckl Ck2 gt gt 1 i e the upper triangular entries 4 1 of the correlation matrix from left to right from the first to the last row Example R file MultiTraitGWAS R q args n 5000 m 100 k 3 MAF 0 3 v 0 005 0 001 0 002 cor mat 0 25 0 3 0 01 strand data The above command generates SNP data for 100 SNPs 5000 individuals and minor allele fre quency 0 3 From these SNPs phenot
12. should be case control which cc and the prevalence of the cases prev Example R file MultiTraitGWAS R q args n 5000 m 100 k 3 MAF 0 3 v 0 005 0 001 0 cor mat 0 2 0 6 0 05 direct TRUE cc TRUE which cc 1 3 prev 0 01 0 05 strand data The above command generates data for 3 phenotypes 100 SNPs 5000 individuals minor allele frequency 0 3 and SNP variance explained of 0 5 0 1 and 0 for the three phenotypes respectively The first and third phenotypes are case control the second is quantitative with case prevalences of 1 and 5 respectively The phenotypic correlation matrix is 1 0 2 0 6 0 2 1 0 05 0 6 0 05 1 5 4 Output A folder with name in the format data_output_date_time is created containing m SNP data files snp_data_i for i 1 m and m phenotype data files pheno_data_i for i 1 m Each SNP data file contains a single column representing a single SNP and each phenotype data file contains k columns representing each phenotype 6 Multivariate method comparison 6 1 Methods The current list of methods included in this software program are e TATES e min P o SHet SHom e CCA e MANOVA e MultiPhen e Combined PC e BIMBAM e SNPTEST N B Use the exact names of the methods that appear above when selecting which methods to compare using the software program For Sy and SHom you should refer to the methods as SHet and SHom 6 2 Sl fixed genetic effects and phenotypic correlations To ru
13. space of different genetic effects on a set of multiple phenotypes of varying degrees of correlation to be explored systematically We present our framework as an R executable program for generating multi trait GWAS data corresponding to the simulation framework and comparing new and existing multi trait GWAS methods across a wide range of scenarios The simulation framework incorporates five different scenarios that explore different combina tions of genetic effects and phenotypic correlations S1 genetic effects and phenotypic correlations varied systematically S2 genetic effects and phenotypic correlations sampled from uniform distribution S3 genetic effects reflect phenotypic correlations S4a fixed genetic effects phenotypic correlations sampled from real data S4b genetic effects and phenotypic correlations sampled from real data This user manual describes how to run the software program that implements this simulation framework examples are provided for each function If you use this software in any published work please cite our comparison study paper For questions regarding this software or the comparison study please contact us multitraitgwasQgmail com 2 Software program The software program folder Multi_Trait_GWAS_Software contains four R scripts and 14 data files The main R script used for the comparisons is MultiTraitGWAS R the other R scripts are dependencies of this script and should be stored in the same l
14. ype data is generated for 3 phenotypes where the SNP explains 0 5 0 1 and 0 2 of the variance in phenotype respectively with phenotypic cor relation matrix 1 0 25 0 3 0 25 1 0 01 0 3 0 01 1 5 2 Phenotypes with indirect SNP effects In addition to the parameters specified above you will need to set direct equal to FALSE and specify n indirect the number of phenotypes to have indirect effects by default the first phenotype always has a direct SNP effect so n indirect 0 k 1 Example R file MultiTraitGWAS R q args n 5000 m 100 k 3 MAF 0 3 v 0 005 0 001 0 002 cor mat 0 25 0 3 0 01 direct FALSE n indirect 1 strand data The above command generates data for 3 phenotypes 5000 individuals 100 SNPs minor al lele frequency 0 3 0 5 variance explained from SNP to the first phenotype 0 1 variance explained from SNP to the second phenotype and 0 2 variance explained from the second phenotype to the third phenotype The phenotypic correlation matrix is the same as above N B The first k n indirect phenotypes have direct SNP effects and the last n indirect phenotypes have indirect SNP effects with the indirect SNP effect on phenotype i being induced by phenotype i 1 for i k n indirect 1 k 5 3 Case control phenotypes In addition to the parameters specified above for direct indirect effects to generate case control data you will need to set cc equal to TRUE specify which phenotypes
Download Pdf Manuals
Related Search