Test compute LD matrix + write Impg model function

impute_jass/impute_jass/__init__.py

+import impute_jass.ld_matrix as LD

impute_jass/impute_jass/ld_matrix.py

+# coding: utf-8
 """
     Function set to compute LD correlation from a reference panel
     in predefined Region
@@ -8,10 +9,9 @@
 import scipy as sc
 import pandas as pd
 import subprocess as sub
-
-sub.check_output("pwd")
-
-LD_region = pd.read_csv('./impute_for_jass/Imputation_for_jass/impute_jass/data/Region_LD.csv')
+import pkg_resources
+import numpy as np
+import re
 
 
 def launch_plink_ld(startpos, endpos, chr, reffile, folder):
@@ -19,36 +19,50 @@ def launch_plink_ld(startpos, endpos, chr, reffile, folder):
     launch plink ld
     """
 
-    fo = "{0}/{1}_{2}_{3}".format(folder, chr, startpos, endpos)
+    fo = "{0}/chr{1}_{2}_{3}".format(folder, chr, startpos, endpos)
     cmd = "p-link --noweb --bfile {0} --r --ld-window-r2 0  --from-bp {1} --to-bp {2} --chr {3} --out {4}".format(reffile, startpos, endpos, chr, fo)
+    #print(cmd)
+
     sub.check_output(cmd, shell=True)
 
 
-def generate_sparse_matrix(plink_ld, path_ld_mat):
+def generate_sparse_matrix(plink_ld):
     """
     read plink results create a sparse dataframe LD-matrix
     then save it to a zipped pickle
     """
 
     plink_ld = pd.read_csv(plink_ld, sep = "\s+")
-    mat_ld = plink_ld.pivot(index='SNP_A', columns='SNP_B', values='R').to_sparse(fill_value=0)
-    mat_ld.to_pickle(path_ld_mat)
+    mat_ld = plink_ld.pivot(index='SNP_A', columns='SNP_B', values='R')
+    un_index = mat_ld.index.union(mat_ld.columns)
+    mat_ld = mat_ld.reindex(index=un_index, columns=un_index)
+    mat_ld.fillna(0, inplace=True)
+
+    sym = mat_ld.values + mat_ld.values.transpose()
+    np.fill_diagonal(sym, 1.01)
+    mat_ld = pd.DataFrame(sym, index=mat_ld.index, columns=mat_ld.columns)
+
+#    mat_ld = pd.DataFrame(np.maximum(mat_ld.values, mat_ld.values.transpose()), index=un_index, columns=un_index)
+    mat_ld = mat_ld.to_sparse()
+    return mat_ld
+    #mat_ld.to_pickle(path_ld_mat,, compression='gzip')
 
 def generate_genome_matrices(region_files, reffolder, folder_output):
     """
-
+    go through region files and compute LD matrix for each transform and
+    save the results in a pandas sparse dataframe
     """
-
     regions = pd.read_csv(region_files)
-    for reg in region_files.iterrows():
+    for reg in regions.iterrows():
         print(reg[0])
         # input reference panel file
         fi_ref = "{0}/{1}.eur.1pct".format(reffolder, reg[1]['chr'])
 
+        chr_int = re.search('([0-9]{1,2})', str(reg[1]['chr'])).group()
         # Compute the LD correlation with LD
-        launch_plink_ld(reg[1]['start'], reg[1]['stop'], reg[1]['chr'], fi_ref, folder_output)
+        launch_plink_ld(reg[1]['start'], reg[1]['stop'], chr_int, fi_ref, folder_output)
 
-        fi_plink = "{0}/{1}_{2}_{3}.ld".format(folder_output, reg[1]['chr'], reg[1]['startpos'], reg[1]["endpos"])
-        fo_mat = "{0}/{1}_{2}_{3}.mat".format(folder_output, reg[1]['chr'], reg[1]['startpos'], reg[1]["endpos"])
+        #fi_plink = "{0}/{1}_{2}_{3}.ld".format(folder_output, reg[1]['chr'], reg[1]['startpos'], reg[1]["endpos"])
+        #fo_mat = "{0}/{1}_{2}_{3}.mat".format(folder_output, reg[1]['chr'], reg[1]['startpos'], reg[1]["endpos"])
         #transform plink output to a compressed generate_sparse_matrix
-        generate_sparse_matrix(fi_plink, fo_mat)
+        #generate_sparse_matrix(fi_plink, fo_mat)

impute_jass/setup.py

 from setuptools import setup, find_packages
 
-setup(name='jass_preprocessing',
+setup(name='impute_jass',
       version='0.1',
       description='Preprocess GWAS summary statistic for JASS',
       url='http:https://gitlab.pasteur.fr/statistical-genetics/JASS_Pre-processing',
@@ -9,5 +9,5 @@ setup(name='jass_preprocessing',
       license='MIT',
       #package_dir = {'': 'jass_preprocessing'},
       packages= ['impute_jass'],
-      package_data = {'impute_jass':'data/*.csv'},
+      package_data = {'impute_jass':'./data/*.csv'},
       zip_safe=False)