GWAS

Read First

Make sure you have read the Setup (Read First) section.

Perhaps the most common study design in population genetics is the Genome-Wide Association Study (GWAS). In TarGene, for computational reasons, a GWAS can only be performed across plink BED files. For our GWAS example, let's save the following snippet to a my_gwas.conf file:

params {
    BED_FILES = "test/assets/unphased_bed/ukb_chr{1,2,3}.{bed,bim,fam}"
    ESTIMATORS_CONFIG = "wtmle--glm"
    TRAITS_DATASET = "test/assets/dataset.csv"
    UKB_WITHDRAWAL_LIST = "test/assets/withdrawal_list.txt"
    ESTIMANDS_CONFIG = "test/assets/gwas_config.yaml"
    UKB_CONFIG = "test/assets/ukbconfig_gwas.yaml"
}

Let's now unpack what this configuration file is saying. Since we have already described most parameters in the Setup (Read First) section, we focus on the two remaining parameters:

  • ESTIMATORS_CONFIG: Defines the estimation strategy. We will be using Targeted Minimum-Loss Estimator with a simple Generalized Linear Model (glm) to learn the outcome models ($Q_Y$)and propensity scores ($G$). For small studies, more complex machine-learning models can be used but for a large GWAS this might be computationally too intensive. See Specifying a Targeted Estimator for more information on this.
  • ESTIMANDS_CONFIG: Since we are performing a GWAS, we are interested in the effect of all variants, across all phenotypes in the UKB_CONFIG file. The ESTIMANDS_CONFIG is a pretty succinct YAML file for a GWAS, it could contain only one line. Here we will be a little more fancy and add extra outcome predictors to improve the precision of inference. The test/assets/gwas_config.yaml contains the following which is self explanatory.
type: gwas

outcome_extra_covariates:
  - "Number of vehicles in household"
  - "Cheese intake"

The optional outcome_extra_covariates are variables to be used as extra predictors of the outcome (but not as confounders). In traditional GWAS these are typically age and sex.

The GWAS can then be run as follows:

nextflow run https://github.com/TARGENE/targene-pipeline -r v0.13.0 -profile local -c my_gwas.config -resume

The workflow should start running and download the first container image. Once successfully completed, the workflow results will be found in the results folder. The main output is the summary stats YAML file (results/results.summary.yaml) which contains entries of the form:

- EFFECT_TYPE: "ATE"
  OUTCOME: Body mass index (BMI)
  TREATMENTS:
    - 1:235339932:G:A: "GG => GA"
    - 1:235339932:G:A: "GA => AA"
  WTMLE_GLM_GLM:
    PVALUE: 0.67290478200139
    COMPONENTS:
      - PVALUE: 0.44667770202884105
        EFFECT_SIZE: -0.5283205242017841
      - PVALUE: 0.5038425323927094
        EFFECT_SIZE: 0.3319580826783738

Remember that in TarGene we estimate the effect of allelic changes and we do not assume that the effect of adding one minor allele is the same as the effect of adding the second minor allele. Hence each variant as up to two distinct effect sizes each with their own p-value. The overall p-value (under the WTMLE_GLM_GLM section) comes from a multidimensional test asking whether any of the allelic changes is non zero. In some cases, only one of the two changes is estimated. This is because there wasn't sufficient data to be confident about assessing the second allelic change (see the POSITIVITY_CONSTRAINT in the workflow parameters).

For more on TarGene's outputs, see Understanding TarGene's Outputs.

GWIS

One may concern themselves with interaction effects at genome-wide scale, also known as a genome-wide interaction study (GWIS). Whether this query concerns how variants interact with other variants (epistasis) or gene-by-environment interaction, both can be addressed using the gwas configuration. All that is required for this that differs from the traditional GWAS configuation is some minor additions to the ESTIMANDS_CONFIG yaml file.

We can begin by expanding upon our intial estimands configuration. In the first example, we did not need to specify the estimands field as they are by default set to be the single order Average Treatment Effects (ATE). Now, that we want more complex estimands we must define the field estimands with the subfields type and orders. Here, we specify that we want to estimate the Average Interaction Effect (AIE) of the second order between each variant and some treatment variable on outcome. These treatment variables are defined under the extra_treatments field where we have specified both the genetic variant 1:235342235:C:T and the environmental variable Hot drink temperature. This configuration defines the pairwise interaction between each genetic variant in the provided BED_FILES and both 1:235342235:C:T and Hot drink temperature.

type: gwas

estimands:
  - type: AIE
    orders: [2]

extra_treatments:
  - Hot drink temperature
  - 1:235342235:C:T

outcome_extra_covariates:
  - "Number of vehicles in household"
  - "Cheese intake"

Futhermore, if we were to add and additional order, say 3:

type: gwas

estimands:
  - type: AIE
    orders: [2,3]

extra_treatments:
  - Hot drink temperature
  - 1:235342235:C:T

outcome_extra_covariates:
  - "Number of vehicles in household"
  - "Cheese intake"

We would then have the previously defined estimands as well as a three-point interaction between each variant and both variables in extra_treatments.

The GWAS can then be run as before:

nextflow run https://github.com/TARGENE/targene-pipeline -r v0.13.0 -profile local -c my_gwas.config -resume

And the output would be similar, however the EFFECT_TYPE and TREATMENTS will, of course, be different. Here is an example of output for the two point interaction between two variants:

- EFFECT_TYPE: "AIE"
  OUTCOME: Body mass index (BMI)
  TREATMENTS:
    - 1:235339932:G:A: "GG => GA"
      1:235342235:C:T: "CC => CT"
    - 1:235339932:G:A: "GA => AA"
      1:235342235:C:T: "CC => CT"
    - 1:235339932:G:A: "GG => GA"
      1:235342235:C:T: "CT => TT"
    - 1:235339932:G:A: "GA => AA"
      1:235342235:C:T: "CT => TT"
  WTMLE_GLM_GLM:
    PVALUE: 0.19908549213021756
    COMPONENTS:
      - PVALUE: 0.09118230780188791
        EFFECT_SIZE: -0.0032775901441661776
      - PVALUE: 0.26941460025816427
        EFFECT_SIZE: 0.0015468229348041384
      - PVALUE: 0.54123876102934512
        EFFECT_SIZE: 0.0008921456782310245
      - PVALUE: 0.71034521098765432
        EFFECT_SIZE: -0.0004356789012345678