diff options
author | zsloan | 2021-08-05 20:51:46 +0000 |
---|---|---|
committer | zsloan | 2021-08-05 20:51:46 +0000 |
commit | b8c88668dbf1bc96c2faf14c483e3af35babc3e5 (patch) | |
tree | 94a81d8729d455ed736a236449121c5a7054f72b /wqflask | |
parent | 457296f004e576a632d2e77c656d41d9ceed0c47 (diff) | |
download | genenetwork2-b8c88668dbf1bc96c2faf14c483e3af35babc3e5.tar.gz |
Generate phenotype and covariate filenames for GEMMA from a hash of values and dataset name - previously the covariate filename was static and the phenotype filename was just random
Diffstat (limited to 'wqflask')
-rw-r--r-- | wqflask/wqflask/marker_regression/gemma_mapping.py | 35 |
1 files changed, 20 insertions, 15 deletions
diff --git a/wqflask/wqflask/marker_regression/gemma_mapping.py b/wqflask/wqflask/marker_regression/gemma_mapping.py index f88c5ac8..efd8bba8 100644 --- a/wqflask/wqflask/marker_regression/gemma_mapping.py +++ b/wqflask/wqflask/marker_regression/gemma_mapping.py @@ -11,6 +11,7 @@ from utility.tools import flat_files from utility.tools import GEMMA_WRAPPER_COMMAND from utility.tools import TEMPDIR from utility.tools import WEBSERVER_MODE +from gn3.computations.gemma import generate_hash_of_string import utility.logger logger = utility.logger.getLogger(__name__) @@ -34,10 +35,7 @@ def run_gemma(this_trait, this_dataset, samples, vals, covariates, use_loco, genofile_name = this_dataset.group.name if first_run: - trait_filename = (f"{str(this_dataset.group.name)}_" - f"{str(this_trait.name)}_" - f"{generate_random_n_string(6)}") - gen_pheno_txt_file(this_dataset, genofile_name, vals, trait_filename) + pheno_filename = gen_pheno_txt_file(this_dataset, genofile_name, vals) if not os.path.isfile(f"{webqtlConfig.GENERATED_IMAGE_DIR}" f"{genofile_name}_output.assoc.txt"): @@ -56,13 +54,13 @@ def run_gemma(this_trait, this_dataset, samples, vals, covariates, use_loco, chr_list_string = ",".join(this_chromosomes_name) if covariates != "": - gen_covariates_file(this_dataset, covariates, samples) + covar_filename = gen_covariates_file(this_dataset, covariates, samples) if use_loco == "True": generate_k_command = (f"{GEMMA_WRAPPER_COMMAND} --json --loco " f"{chr_list_string} -- {GEMMAOPTS} " f"-g {flat_files('genotype/bimbam')}/" f"{genofile_name}_geno.txt -p " - f"{TEMPDIR}/gn2/{trait_filename}.txt -a " + f"{TEMPDIR}/gn2/{pheno_filename}.txt -a " f"{flat_files('genotype/bimbam')}/" f"{genofile_name}_snps.txt -gk > " f"{TEMPDIR}/gn2/{k_output_filename}.json") @@ -73,10 +71,10 @@ def run_gemma(this_trait, this_dataset, samples, vals, covariates, use_loco, f"-- {GEMMAOPTS} " f"-g {flat_files('genotype/bimbam')}/" f"{genofile_name}_geno.txt " - f"-p {TEMPDIR}/gn2/{trait_filename}.txt ") + f"-p {TEMPDIR}/gn2/{pheno_filename}.txt ") if covariates != "": gemma_command += (f"-c {flat_files('mapping')}/" - f"{this_dataset.group.name}_covariates.txt " + f"{covar_filename}.txt " f"-a {flat_files('genotype/bimbam')}/" f"{genofile_name}_snps.txt " f"-lmm 9 -maf {maf} > {TEMPDIR}/gn2/" @@ -92,7 +90,7 @@ def run_gemma(this_trait, this_dataset, samples, vals, covariates, use_loco, f"{GEMMAOPTS} " f" -g {flat_files('genotype/bimbam')}/" f"{genofile_name}_geno.txt -p " - f"{TEMPDIR}/gn2/{trait_filename}.txt -a " + f"{TEMPDIR}/gn2/{pheno_filename}.txt -a " f"{flat_files('genotype/bimbam')}/" f"{genofile_name}_snps.txt -gk > " f"{TEMPDIR}/gn2/{k_output_filename}.json") @@ -106,12 +104,11 @@ def run_gemma(this_trait, this_dataset, samples, vals, covariates, use_loco, f"{genofile_name}_snps.txt " f"-lmm 9 -g {flat_files('genotype/bimbam')}/" f"{genofile_name}_geno.txt -p " - f"{TEMPDIR}/gn2/{trait_filename}.txt ") + f"{TEMPDIR}/gn2/{pheno_filename}.txt ") if covariates != "": gemma_command += (f" -c {flat_files('mapping')}/" - f"{this_dataset.group.name}" - f"_covariates.txt > " + f"{covar_filename}.txt > " f"{TEMPDIR}/gn2/{gwa_output_filename}.json") else: gemma_command += f" > {TEMPDIR}/gn2/{gwa_output_filename}.json" @@ -129,16 +126,20 @@ def run_gemma(this_trait, this_dataset, samples, vals, covariates, use_loco, return marker_obs, gwa_output_filename -def gen_pheno_txt_file(this_dataset, genofile_name, vals, trait_filename): +def gen_pheno_txt_file(this_dataset, genofile_name, vals): """Generates phenotype file for GEMMA""" - with open(f"{TEMPDIR}/gn2/{trait_filename}.txt", "w") as outfile: + filename = "PHENO_" + generate_hash_of_string(this_dataset.name + str(vals)) + + with open(f"{TEMPDIR}/gn2/{filename}.txt", "w") as outfile: for value in vals: if value == "x": outfile.write("NA\n") else: outfile.write(value + "\n") + return filename + def gen_covariates_file(this_dataset, covariates, samples): covariate_list = covariates.split(",") @@ -168,14 +169,18 @@ def gen_covariates_file(this_dataset, covariates, samples): this_covariate_data.append("-9") covariate_data_object.append(this_covariate_data) + filename = "COVAR_" + generate_hash_of_string(this_dataset.name + str(covariate_data_object)) + with open((f"{flat_files('mapping')}/" - f"{this_dataset.group.name}_covariates.txt"), + f"{filename}.txt"), "w") as outfile: for i in range(len(covariate_data_object[0])): for this_covariate in covariate_data_object: outfile.write(str(this_covariate[i]) + "\t") outfile.write("\n") + return filename + def parse_loco_output(this_dataset, gwa_output_filename, loco="True"): |