aboutsummaryrefslogtreecommitdiff
path: root/wqflask
diff options
context:
space:
mode:
Diffstat (limited to 'wqflask')
-rw-r--r--wqflask/wqflask/marker_regression/qtlreaper_mapping.py46
1 files changed, 28 insertions, 18 deletions
diff --git a/wqflask/wqflask/marker_regression/qtlreaper_mapping.py b/wqflask/wqflask/marker_regression/qtlreaper_mapping.py
index 78b1f7b0..505ae295 100644
--- a/wqflask/wqflask/marker_regression/qtlreaper_mapping.py
+++ b/wqflask/wqflask/marker_regression/qtlreaper_mapping.py
@@ -17,22 +17,29 @@ def run_reaper(this_trait, this_dataset, samples, vals, json_data, num_perm, boo
else:
genofile_name = this_dataset.group.name
- trait_filename = str(this_trait.name) + "_" + str(this_dataset.name) + "_pheno"
+ trait_filename =f"{str(this_trait.name)}_{str(this_dataset.name)}_pheno"
gen_pheno_txt_file(samples, vals, trait_filename)
- output_filename = this_dataset.group.name + "_GWA_" + ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(6))
+ output_filename = (f"{this_dataset.group.name}_GWA_"+
+ ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(6))
+ )
bootstrap_filename = None
permu_filename = None
opt_list = []
if boot_check and num_bootstrap > 0:
- bootstrap_filename = this_dataset.group.name + "_BOOTSTRAP_" + ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(6))
+ bootstrap_filename = (f"{this_dataset.group.name}_BOOTSTRAP_" +
+ ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(6))
+ )
opt_list.append("-b")
- opt_list.append("--n_bootstrap " + str(num_bootstrap))
- opt_list.append("--bootstrap_output " + webqtlConfig.GENERATED_IMAGE_DIR + bootstrap_filename + ".txt")
+ opt_list.append(f"--n_bootstrap{str(num_bootstrap)}")
+ opt_list.append(f"--bootstrap_output{webqtlConfig.GENERATED_IMAGE_DIR}{bootstrap_filename}.txt")
if num_perm > 0:
- permu_filename = this_dataset.group.name + "_PERM_" + ''.join(random.choice(string.ascii_uppercase + string.digits) for _ in range(6))
+ permu_filename =("{this_dataset.group.name}_PERM_" +
+ ''.join(random.choice(string.ascii_uppercase +
+ string.digits) for _ in range(6))
+ )
opt_list.append("-n " + str(num_perm))
opt_list.append("--permu_output " + webqtlConfig.GENERATED_IMAGE_DIR + permu_filename + ".txt")
if control_marker != "" and do_control == "true":
@@ -40,13 +47,15 @@ def run_reaper(this_trait, this_dataset, samples, vals, json_data, num_perm, boo
if manhattan_plot != True:
opt_list.append("--interval 1")
- reaper_command = REAPER_COMMAND + ' --geno {0}/{1}.geno --traits {2}/gn2/{3}.txt {4} -o {5}{6}.txt'.format(flat_files('genotype'),
- genofile_name,
- TEMPDIR,
- trait_filename,
- " ".join(opt_list),
- webqtlConfig.GENERATED_IMAGE_DIR,
- output_filename)
+ reaper_command = (REAPER_COMMAND +
+ ' --geno {0}/{1}.geno --traits {2}/gn2/{3}.txt {4} -o {5}{6}.txt'.format(flat_files('genotype'),
+
+ genofile_name,
+ TEMPDIR,
+ trait_filename,
+ " ".join(opt_list),
+ webqtlConfig.GENERATED_IMAGE_DIR,
+ output_filename))
logger.debug("reaper_command:" + reaper_command)
os.system(reaper_command)
@@ -61,12 +70,13 @@ def run_reaper(this_trait, this_dataset, samples, vals, json_data, num_perm, boo
suggestive = permu_vals[int(num_perm*0.37-1)]
significant = permu_vals[int(num_perm*0.95-1)]
- return marker_obs, permu_vals, suggestive, significant, bootstrap_vals, [output_filename, permu_filename, bootstrap_filename]
+ return (marker_obs, permu_vals, suggestive, significant, bootstrap_vals,
+ [output_filename, permu_filename, bootstrap_filename])
def gen_pheno_txt_file(samples, vals, trait_filename):
"""Generates phenotype file for GEMMA"""
- with open("{}/gn2/{}.txt".format(TEMPDIR, trait_filename), "w") as outfile:
+ with open(f"{TEMPDIR}/gn2/{trait_filename}.txt","w") as outfile:
outfile.write("Trait\t")
filtered_sample_list = []
@@ -90,7 +100,7 @@ def parse_reaper_output(gwa_filename, permu_filename, bootstrap_filename):
only_cm = False
only_mb = False
- with open("{}{}.txt".format(webqtlConfig.GENERATED_IMAGE_DIR, gwa_filename)) as output_file:
+ with open(f"{webqtlConfig.GENERATED_IMAGE_DIR}{gwa_filename}.txt") as output_file:
for line in output_file:
if line.startswith("ID\t"):
if len(line.split("\t")) < 8:
@@ -137,13 +147,13 @@ def parse_reaper_output(gwa_filename, permu_filename, bootstrap_filename):
permu_vals = []
if permu_filename:
- with open("{}{}.txt".format(webqtlConfig.GENERATED_IMAGE_DIR, permu_filename)) as permu_file:
+ with open(f"{webqtlConfig.GENERATED_IMAGE_DIR}{permu_filename}.txt") as permu_file:
for line in permu_file:
permu_vals.append(float(line))
bootstrap_vals = []
if bootstrap_filename:
- with open("{}{}.txt".format(webqtlConfig.GENERATED_IMAGE_DIR, bootstrap_filename)) as bootstrap_file:
+ with open(f"{webqtlConfig.GENERATED_IMAGE_DIR}{bootstrap_filename}.txt") as bootstrap_file:
for line in bootstrap_file:
bootstrap_vals.append(int(line))