This is a GWAS pipeline that is intended to be run with biobank size data

The workflow is designed to be modular with a main workflow file that calls the preprocessing subworkflow and (optionally) the GWAS tools. Currently it supports Bolt-LMM and SAIGE.

Inputs for the workflow

genotype files (required)

The pipeline requires both directly genotyped and imputed genotype files. Currently, the pipeline supports directly genotype files in binary PLINK format, and imputed genotype files in either VCF or bgen format. The directly genotype files and the imputed genotypes do not have to be with the same genome build. For example, the directly genotypes can be of hg19 and the imputed genotypes can be of hg38. When the genome builds differ, the directly genotyped files will be converted to match with the imputed genotypes, and the user has to provide the chain_file for liftover.

More spefically,

• genotype_bed, genotype_bim and genotype_fam are directly genotype files in Plink format
• genotype_samples_to_keep_file: subset of samples to be used from the genotype file. Need to have FID and IID columns.
• chain_file (optional): lifting the genotype files to the same build as the imputed files.

The first 10 principal components are estimated from the genotype plink files and will be added as covariates for the GWAS analysis.

Imputed files (required)

The workflow can handle either VCF or bgen files as inputs for the imputed genotype data. The list of the file and their location can be specified as imputed_list_of_vcf_file or imputed_list_of_bgen_file.

In case that the VCF file is used, they will be converted to bgen format for the association study. The user can provide the id_delim string if FID and IID are concat into one ID, if id_delim is not provided, double ID is assumed - both FID and IID are the same as sample ID in VCF.

Finally, if the imputed data and the genotype data are of different genome build, the workflow can liftover the genotype files to match with the imputed data if the user provides the chain_file.

• imputed_list_of_vcf_file or imputed_list_of_bgen_file: list of imputed genotype files (they are usually divided by chromosome)
• imputed_samples_to_keep_file: subset of samples to be used from the imputed genotype file. Need to have FID and IID columns.

Defining phenotype, covariates and sample sets (required)

The phenotypes, covariates and cohorts are defined by 3 files:

• covariate_tsv_file: File that contains the sample ID, phenotype, and the covariates used in the association study.
• variable_info_tsv_file: File that defines the columns in the covariate_tsv_file.
• sample_sets_json_file: File that defines the subsets used for the association study.

More specifically: The *covariate_tsv_file" is a tab delimited file that includes the id, phenotype and covariates of each subject, here is an example:

id is_case  sex age
id1 1 1 56
id2 0 0 60

The variable_info_tsv_file describes each of the columns that will be used in the analysis, which includes sampleID, phenotype and covarites. These columns/variable names should match with those in the covariate_tsv_file. The columns that are not used in the covariate_tsv_file do not have to be listed here. The variable_info_tsv_file has 4 columns as shown in the example file below. Currently, only variableName and variableType are used by the pipeline.

• variableName is the column name in the covariate_tsv_file
• variableType could be:
  • sampleID: this variable is the sample id
  • phenotype_quantitative: this variable is a phenotype/target variable and it's a quantitative trait
  • phenotype_binary: this variable is a phenotype/target variable and it's a binary trait
  • quantitative: this is a quantitative covaraite
  • binary: this is a binary covariate
  • categorical: this is a categorical covariate. The workflow will perform one-hot encoding to turn this variable into N-1 binary variables, where N is the number of categories

variableName    variableType    na_values       transformation
bmi  phenotype_quantitative  NA      no
sex     binary  NA      no

The sample_sets_json_file defines the different subset of the samples you would like to use as the cohort of the study. For example, this json file defines 3 sets: the complete_set which has no selection criteria, setMale which includes those subjects in the sex column with value equals to 1, and setFemale which includes those subjects in the xex column with value equals to 2. Also note that the set name (e.g. complete_set in the example below) will be the appended as the suffix for the gwas result files.

{
   "complete_set":[],
   "setMale": [
    {"varname":"sex", "criteria":"include", "value":"1"}
   ],
   "setFemale": [
    {"varname":"sex", "criteria":"include", "value":"2"}
   ]
}

Which GWAS tool to use

• useBOLT: true or false
  • if bolt is used, it requires the genetic_map_file and ld_scores_file
• useSAIGE: true or false

association test programs specific options

*minMAC: minimum minor allele count *minMAF: minimum minor allele frequency

workflow steps

• Determine the sample subsets and create covar files for each subset
• The tasks are scattered for each sample subset:
  • The preprocessing steps are summarized in the DAG.
  • Run on each of the association tool specified
  • Gather the results
  • Make Manhattan and QQ plots

Limitations

• Right now this pipeline only works on the 22 human autosomes
• Variable transformation is not yet handled
• Currently, different batches are analyzed together with batch as a covariate, future versions will support meta-analysis of different batches
• The workflow currently points to the latest version of subworkflow on GitHub, this needs to be changed to a particular release instead.