Skip to contents

DuoRank PRS (Dr.PRS)

dr.prs.Rd

Dr.PRS is designed to rank the importance for PRS inputs and futher optimazation. It takes two none-overlap stage plink file (bfile) for machine learning modeling:

  • lasso (to capture linear relationship)

  • catboost (to capture non-linear relationship) It also calculated PRS with plink using traditional additive model.

Usage

dr.prs(
  stage1_bfile = "./data/zuo/bed/SNP_for_PRS/VKH-zhonghua-for_prs_snplist",
  stage2_bfile = "./data/zuo/bed/SNP_for_PRS/VKH-ASA-for_prs_snplist",
  summary_file = "./output/part1-gwas/prs/snp_for_prs.txt",
  output_dir = "./output/part1-gwas/prs/b1t2",
  method = c(1, 2, 3),
  dr_optimazation = TRUE,
  snp_col = "SNP.meta",
  a1_col = "A1",
  weight_col = "OR_Final",
  weight_type = "OR",
  divide_ratio = 0.7,
  iLasso_iteration = 1000,
  nfolds = 10,
  plink_bin = plinkbinr::get_plink_exe(),
  clump = T,
  clump_include_hla = F,
  clump_param = NULL,
  seed = 725
)

combine_rank(rank1, rank2, auc1 = NULL, auc2 = NULL)

Arguments

stage1_bfile

Path to the stage1 PLINK binary files. Normally it is the one generates summary data. If only 1 source of individual data is available, you can split it into 2 non-overlap datasets.

stage2_bfile

Path to the stage2 PLINK binary files, i.e., the target dataset for PRS calculation.

summary_file

Path to the summary statistics file or a data frame containing SNPs, alleles, and weights. Note this summary only contains refined SNP for PRS calculation.

output_dir

Directory to save output files.

method

Integer vector indicating which methods to use: 1=PLINK, 2=CatBoost, 3=iLasso. Default is all three methods.

dr_optimazation

Logical, whether to perform greedy search to optimize PRS based on combined rank from CatBoost and iLasso (default: TRUE).

snp_col

Column name in the summary statistics for SNP IDs (default: "SNP.meta").

a1_col

Column name in the summary statistics for effect allele (default: "A1").

weight_col

Column name in the summary statistics for weights (default: "OR_Final").

weight_type

Type of weights, either "OR" (default) or "Beta". If "OR", it will calculate Beta values.

divide_ratio

Proportion of stage1 data in iLasso (default: 0.7).

iLasso_iteration

Number of iterations for iLasso model (default: 1000).

nfolds

Number of folds for cross-validation (default: 10).

Path to the PLINK binary executable (default: plinkbinr::get_plink_exe()).

clump

Place holder for future clumping function.

clump_include_hla

Place holder for future clumping function.

clump_param

Place holder for future clumping function.

seed

Random seed (Default: 725).

rank1

rank1

rank2

rank2

auc1

auc1

auc2

auc2

Value

A list with:

  • lasso_importance, catboost_importance, combined_rank

  • greedy_auc_curve (data.frame: kept_snps, auc)

  • final_subset (character vector of SNPs)

  • plink_prs_file (path), logs