Genome Wide Association Neural Networks (GWANN)

Manuscript

Genome-wide association neural networks identify genes linked to family history of Alzheimer’s disease

Prerequisites

• SNP data in PGEN files
• Covariates in a csv file

Note: The code has been written to analyse UK Biobank data and may have some data fields such as age and sex hard-coded in certain parts of the code.

Analysis parameters

The analysis requires parameters and covariates in a yaml file. It also needs the participant ids along with the phenotype in csv files. These participants will be extracted from the pgen file (genetic data) and the covariate file.

They need to be within the following folder structure:

GWANN
  |__Code_AD
    |__<params>
      |__<exp_name>
        |-- params_FH_AD.yaml
        |-- covs_FH_AD.yaml
        |-- all_ids_FH_AD.yaml
        |-- train_ids_FH_AD.yaml
        |-- test_ids_FH_AD.yaml

• Currently the code expects the folder to be called 'params' for the White-British analysis and 'params_non_white' for the Asian and Black analysis. The <exp_name> is set as 'Sens8'. If you wish to modify the names of these folders, please change the path in the following folders:

  • Code_AD/run_genes.py
  • Code_AD/cov_model.py
  • Code_AD/dummy_genes.py
  • Code_AD/non_white_validation.py
  • Code_AD/non_white_dummy_genes.py

• 'FH_AD' is the phenotype of interest. Please change it if your phenotype is called something else.

Parameters

DATA_BASE_FOLDER: <Path to folder where csv data files should be saved>
DATA_LOGS: <Path to folder where data generation logs should be saved>
LOGS_BASE_FOLDER: <Path to folder where training plots and models should be saved>
RUNS_BASE_FOLDER: <Path to folder where training info and logs should be written>
PARAMS_PATH: <Path to this yaml file>
PHEN_COV_PATH: <Path to file with participant IDs, phenotype and covariate information>
RAW_BASE_FOLDER:
  '1': <Path to folder containing pgen file for chromosome>
  '10': <Path to folder containing pgen file for chromosome>
  '11': <Path to folder containing pgen file for chromosome>
  '12': <Path to folder containing pgen file for chromosome>
  '13': <Path to folder containing pgen file for chromosome>
  '14': <Path to folder containing pgen file for chromosome>
  '15': <Path to folder containing pgen file for chromosome>
  '16': <Path to folder containing pgen file for chromosome>
  '17': <Path to folder containing pgen file for chromosome>
  '18': <Path to folder containing pgen file for chromosome>
  '19': <Path to folder containing pgen file for chromosome>
  '2': <Path to folder containing pgen file for chromosome>
  '20': <Path to folder containing pgen file for chromosome>
  '21': <Path to folder containing pgen file for chromosome>
  '22': <Path to folder containing pgen file for chromosome>
  '3': <Path to folder containing pgen file for chromosome>
  '4': <Path to folder containing pgen file for chromosome>
  '5': <Path to folder containing pgen file for chromosome>
  '6': <Path to folder containing pgen file for chromosome>
  '7': <Path to folder containing pgen file for chromosome>
  '8': <Path to folder containing pgen file for chromosome>
  '9': <Path to folder containing pgen file for chromosome>
  'Dummy': <Path to folder containing pgen file with dummy data>

Covariates

List of covariates to use for the analysis. The ones given below were used in the analysis described in the manuscript.

COVARIATES : 
- 'f.31.0.0'
- 'f.21003.0.0'
- 'f.6138_ISCED'
- 'f.22009.0.1'
- 'f.22009.0.2'
- 'f.22009.0.3'
- 'f.22009.0.4'
- 'f.22009.0.5'
- 'f.22009.0.6'

White-British Population

The pipeline in Code_AD/run_pipeline.sh was used to run the analysis genome wide.

• It first trains the covariate model
• Then trains the models for each gene window
• Finally, trains models for dummy gene windows

Note: Please modify paths before running

Non-white Populations

After training the models on the White-British Population, the Code_AD/non_white_validation.sh was used to run the models in inference mode on the Asian and Black populations in the UK Biobank.

• It obtains the metrics for all random seeds and both populations for the top 100 gene windows from the white-british population
• Finally, it generates dummy SNPs for each population and obtains the metrics for the dummy windows

Note: Please modify paths before running

P-value calculations

The script Code_AD/post_hoc_scripts/calc_pvalue.R is used to obtain the P-value from a single file containing the metrics of every gene window for all random seeds, and a similar file for all dummy gene windows.

Note: Please modify paths before running

Post-hoc analyses

The scripts in the folder Code_AD/post_hoc_scripts were used to perform the series of post hoc analyses discussed in the manuscript.

Note: Please modify paths before running