LSHTM_analysis/scripts/or_kinship_link.py

478 lines
18 KiB
Python
Executable file

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Wed Jun 10 11:13:49 2020
@author: tanu
"""
#=======================================================================
#%% useful links
#https://chrisalbon.com/python/data_wrangling/pandas_join_merge_dataframe/
#https://kanoki.org/2019/11/12/how-to-use-regex-in-pandas/
#https://stackoverflow.com/questions/40348541/pandas-diff-with-string
#=======================================================================
import os, sys
import pandas as pd
import numpy as np
import re
import argparse
homedir = os.path.expanduser('~')
os.chdir(homedir + '/git/LSHTM_analysis/scripts')
# local import
from find_missense import find_missense
#=======================================================================
#%% command line args
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument('-d', '--drug', help = 'drug name', default = None)
arg_parser.add_argument('-g', '--gene', help = 'gene name (case sensitive)', default = None)
arg_parser.add_argument('--datadir', help = 'Data Directory. By default, it assmumes homedir + git/Data')
arg_parser.add_argument('-i', '--input_dir', help = 'Input dir containing pdb files. By default, it assmumes homedir + <drug> + input')
arg_parser.add_argument('-o', '--output_dir', help = 'Output dir for results. By default, it assmes homedir + <drug> + output')
# FIXME: remove defaults
arg_parser.add_argument('-sc', '--start_coord', help = 'start of coding region (cds) of gene', default = None, type = int) # pnca cds
arg_parser.add_argument('-ec', '--end_coord', help = 'end of coding region (cds) of gene', default = None, type = int) # pnca cds
arg_parser.add_argument('--debug', action ='store_true', help = 'Debug Mode')
args = arg_parser.parse_args()
#=======================================================================
#%% variables
#gene = 'pncA'
#drug = 'pyrazinamide'
#start_cds = 2288681
#end_cds = 2289241
#%%=====================================================================
# Command line options
gene = args.gene
drug = args.drug
gene_match = gene + '_p.'
datadir = args.datadir
indir = args.input_dir
outdir = args.output_dir
start_cds = args.start_coord
end_cds = args.end_coord
#%%=======================================================================
#==============
# directories
#==============
if not datadir:
datadir = homedir + '/' + 'git/Data'
if not indir:
indir = datadir + '/' + drug + '/input'
if not outdir:
outdir = datadir + '/' + drug + '/output'
#=======
# input
#=======
gene_info_filename = 'ns'+ gene.lower()+ '_snp_info.txt'
#gene_info_filename = 'ns'+ gene.lower()+ '_snp_info.csv'
gene_info = indir + '/' + gene_info_filename
print('gene info file: ', gene_info
, '\n============================================================')
in_filename_or = 'ns'+ gene.lower()+ '_assoc.txt'
gene_or = indir + '/' + in_filename_or
print('gene OR file: ', gene_or
, '\n============================================================')
#=======
# output
#=======
gene_or_filename = gene.lower() + '_af_or_kinship.csv' # other one is called AFandOR
outfile_or_kin = outdir + '/' + gene_or_filename
print('Output file: ', outfile_or_kin
, '\n============================================================')
#%% read files: preformatted using bash
# or file: '...assoc.txt'
# FIXME: call bash script from here
or_df = pd.read_csv(gene_or, sep = '\t', header = 0, index_col = False) # 182, 12 (without filtering for missense muts, it was 212 i.e we 30 muts weren't missense)
or_df.head()
or_df.columns
#%% snp_info file: master and gene specific ones
# gene info
info_df2 = pd.read_csv(gene_info, sep = '\t', header = 0) #447, 11
#info_df2 = pd.read_csv(gene_info, sep = ',', header = 0) #447 10
mis_mut_cover = (info_df2['chromosome_number'].nunique()/info_df2['chromosome_number'].count()) * 100
print('*****RESULT*****'
, '\nPercentage of missense mut in pncA:', mis_mut_cover
, '\n*****RESULT*****')
# v6: 61.07
# v4: 65.7%
#%% Extracting ref allele and alt allele as single letters
# info_df has some of these params as more than a single letter, which means that
# when you try to merge ONLY using chromosome_number, then it messes up... and is WRONG.
# Hence the merge needs to be performed on a unique set of attributes which in our case
# would be chromosome_number, ref_allele and alt_allele
df_ncols = len(or_df.columns)
print('Dim of df:',or_df.shape
, '\nExtracting missense muts as single letters from: find_missense()')
find_missense(or_df, ref_allele_column = 'ref_allele1', alt_allele_column = 'alt_allele0')
print('Dim of revised df:', or_df.shape, ' after extraction of missense muts')
# FIXME: import this from function
ncols_added_func = 4
if or_df.shape[1] == df_ncols + ncols_added_func:
print('PASS: Succesfuly extracted ref and alt alleles for missense muts')
else:
print('FAIL: No. of cols mismatch'
,'\nOriginal length:', df_ncols
, '\nExpected no. of cols:', df_ncols + ncols_added_func
, '\nGot:', or_df.shape[1]
, '\nCheck hardcoded value of ncols_add?')
if (or_df['tot_diff'] == 1).sum() == len(or_df) and (or_df['n_diff'] == 1).sum() == len(or_df) and or_df['n_diff'].equals(or_df['tot_diff']):
print('PASS: missene muts correctly extracted from source')
else:
print('FAIL: n_diff and tot_diff differ, check source data')
sys.exit()
del(df_ncols, ncols_added_func)
#%% check dtypes before merging
#or_df.dtypes
or_df.info()
#info_df2.dtypes
info_df2.info()
#%% perform merge: or_df and snp_info
print('Preparing dfs for merging... Finding common cols to merge')
# find common columns
#merging_cols = ['chromosome_number', 'ref_allele', 'alt_allele']
merging_cols = or_df.columns[or_df.columns.isin(info_df2.columns)].to_list()
print('No. of common cols identified:', len(merging_cols)
, '\nColumns to merge on:', merging_cols
, '\nChecking dtypes in merging_cols...'
, '\n=================================================')
# make sure chromosome_number dtypes are consisent
or_df[merging_cols].dtypes == info_df2[merging_cols].dtypes
# info_df2 contains multiple chromosome number in the column, so it is not
# possible to convert this to int. Therefore, converting to string in or_df column
if not (or_df[merging_cols].dtypes == info_df2[merging_cols].dtypes).all():
print('Data types not same, converting chromsome_number to str in or_df')
or_df['chromosome_number'] = or_df['chromosome_number'].astype(str)
print('Checking after converting dtype in or_df')
if (or_df[merging_cols].dtypes == info_df2[merging_cols].dtypes).all():
print('PASS: dfs ready to merge..')
else:
print('FAIL: unable to make dtypes consistent required for merging! Check dtypes')
sys.exit()
# %% sanity check and perform merge
#my_join = 'inner'
#my_join = 'outer'
my_join = 'left'
#my_join = 'right'
expected_cols = or_df.shape[1] + info_df2.shape[1] - len(merging_cols)
print('Merging 2 dfs: or_df and info_df'
, '\nJoin type:', my_join
, '\nColumns to merge on:', merging_cols
, '\nExpected cols after merging:', expected_cols
, '\n=================================================')
dfm2 = pd.merge(or_df, info_df2, on = merging_cols, how = my_join, indicator = True)
dfm2['_merge'].value_counts()
expected_cols = expected_cols + 1 # due to indicator = T
# count no. of nan in 'mut_type'. ideally should be 0
if not dfm2['mut_type'].isna().sum() > 0:
print('Good merging, no NA detected')
else:
print('OR detected without metadata'
, '\nNo. of NA in dfm2:', dfm2['mut_type'].isna().sum()
, '\nWriting these to output file to check later with jody!')
dfm2_missing_info = dfm2[dfm2['mut_type'].isna()]
missing_or_metadata = "or_kinship_missing_metadata.csv"
outfile_missing_or_metadata = outdir + '/' + str(dfm2['mut_type'].isna().sum()) + '_' + missing_or_metadata
print('\noutput file:', outfile_missing_or_metadata)
dfm2_missing_info.to_csv(outfile_missing_or_metadata, index = False)
print('Finsihed writing file'
, '\nDim:', dfm2_missing_info.shape)
#PENDING Jody's reply
# !!!!!!!!
# drop nan from dfm2_mis as these are not useful and JP confirmed the same
print('Dropping NAs before further processing...')
dfm2_mis = dfm2[dfm2['mut_type'].notnull()]
# !!!!!!!!
#%% extract mut info into three cols
df_ncols = len(dfm2_mis.columns)
print('Dim of df to add cols to:', dfm2_mis.shape)
# column names already present, wrap this in a if and perform sanity check
ncols_add = 0
if not 'wild_type' in dfm2_mis.columns:
print('Extracting and adding column: wild_type'
, '\n===============================================================')
dfm2_mis['wild_type'] = dfm2_mis['mut_info_f1'].str.extract('(\w{1})>')
ncols_add+=1
if not 'position' in dfm2_mis.columns:
print('Extracting and adding column: position'
, '\n===============================================================')
dfm2_mis['position'] = dfm2_mis['mut_info_f1'].str.extract('(\d+)')
#dfm2_mis['position'] = dfm2_mis[:,'mut_info_f1'].str.extract('(\d+)')
ncols_add+=1
if not 'mutant_type' in dfm2_mis.columns:
print('Extracting and adding column: mutant_type'
, '\n================================================================')
dfm2_mis['mutant_type'] = dfm2_mis['mut_info_f1'].str.extract('>\d+(\w{1})')
ncols_add+=1
if not 'mutationinformation' in dfm2_mis.columns:
print('combining to create column: mutationinformation'
, '\n===============================================================')
dfm2_mis['mutationinformation'] = dfm2_mis['wild_type'] + dfm2_mis['position'] + dfm2_mis['mutant_type']
ncols_add+=1
print('No. of cols added:', ncols_add)
if len(dfm2_mis.columns) == df_ncols + ncols_add:
print('PASS: mcsm style muts added to df'
, '\n===============================================================')
else:
print('FAIL: No. of cols mismatch'
,'\nOriginal length:', df_ncols
, '\nExpected no. of cols:', df_ncols + ncols_add
, '\nGot:', len(dfm2_mis.columns))
sys.exit()
del(df_ncols, ncols_add)
#%% now adding mutation style = <gene>_p.abc1cde
dfm2_mis['mutation'] = gene.lower() + '_' + dfm2_mis['mut_info_f2'].astype(str)
# convert to lowercase
dfm2_mis['mutation'] = dfm2_mis['mutation'].str.lower()
# quick sanity check
check = dfm2_mis['mutation'].value_counts().value_counts() == dfm2_mis['mut_info_f2'].value_counts().value_counts()
if check.all():
print('PASS: added column "mutation" containing mutation format: <gene>_p.abc1cde')
else:
print('FAIL: could not add "mutation" column!')
sys.exit()
#%% Calculating OR from beta coeff
print('Calculating OR...')
df_ncols = dfm2_mis.shape[1]
print('No. of cols pre-formatting data:', df_ncols
, '\n===================================================================')
#1) Add column: OR for kinship calculated from beta coef
ncols_add = 0
if not 'or_kin' in dfm2_mis.columns:
#dfm2_mis['or_kin'] = np.exp(dfm2_mis['beta']) # gives copy warning
dfm2_mis.loc[:,'or_kin'] = np.exp(dfm2_mis.loc[:,'beta'])
print(dfm2_mis['or_kin'].head())
ncols_add+=1
print('Calculating OR from beta coeff by exponent function and adding column:'
, '\nNo. of cols added:', ncols_add
, '\n', dfm2_mis['beta'].head()
, '\nNo. of cols after adding OR_kin:', len(dfm2_mis.columns)
, '\n===================================================================')
if dfm2_mis.shape[1] == df_ncols + ncols_add:
print('PASS: Dimension of df match'
, '\nDim of df:', dfm2_mis.shape
, '\n================================================================')
else:
print('FAIL: Dim mismatch'
, '\nOriginal no. of cols:', df_ncols
, '\nExpected no. of cols:', df_ncols + ncols_add
, '\nGot:', dfm2_mis.shape[1])
sys.exit()
#2) rename columns to reflect that it is coming from kinship matrix adjustment
dfm2_mis.rename(columns = {'af': 'af_kin'
, 'beta': 'beta_kin'
, 'p_wald': 'pwald_kin'
, 'se': 'se_kin', 'logl_H1': 'logl_H1_kin'
, 'l_remle': 'l_remle_kin'
, 'ref_allele1': 'reference_allele'}, inplace = True)
del(df_ncols, ncols_add)
#%%==============================!!!!!!!=======================================
# FIXME: should be at source
# checking tot_diff column
#print((dfm2_mis['tot_diff']==1).all()) and remove these cols
#%%==============================!!!!!!!=======================================
#3) drop some not required cols (including duplicate if you want)
#3a) drop duplicate columns
dfm2_mis2 = dfm2_mis.T.drop_duplicates().T #changes dtypes in cols, only used for sanity check
dup_cols = set(dfm2_mis.columns).difference(dfm2_mis2.columns)
print('Total no of duplicate columns:', len(dup_cols)
, '\nDuplicate columns identified:', dup_cols
, '\n===================================================================')
#dup_cols = {'alt_allele0', 'ps'} # didn't want to remove tot_diff
#print('removing duplicate columns: kept one of the dup_cols i.e tot_diff')
df_ncols = dfm2_mis.shape[1]
print('Removing', len(dup_cols), 'duplicate columns:', dup_cols
, '\nOriginal dim:', dfm2_mis.shape)
dfm2_mis.drop(list(dup_cols), axis = 1, inplace = True)
if dfm2_mis.shape[1] == df_ncols - len(dup_cols):
print('PASS: Dimensions match'
, '\nDim:', dfm2_mis.shape
, '\nRemoved', len(dup_cols), 'columns from' , df_ncols
, '\n===============================================================')
else:
print('FAIL: Dimensions mismatch'
, '\nOriginal no. of cols:', df_ncols
, '\nNo. of cols to drop:', len(dup_cols)
, '\nExpected:', df_ncols - len(dup_cols)
, '\nGot:', dfm2_mis.shape[1])
sys.exit()
del(df_ncols)
#3b) other not useful columns
print('Dropping other redundant or unnecessary columns...')
cols_to_drop = ['chromosome_text', 'n_diff', 'chr', '_merge' , 'mut_region' , 'reference_allele', 'alternate_allele']
df_ncols = dfm2_mis.shape[1]
dfm2_mis.drop(cols_to_drop, axis = 1, inplace = True)
#dfm2_mis.rename(columns = {'ref_allele1': 'reference_allele'}, inplace = True)
if dfm2_mis.shape[1] == df_ncols - len(cols_to_drop):
print('PASS:', len(cols_to_drop), 'columns successfully dropped'
, '\nDim:', dfm2_mis.shape
, '\nRemoved', len(cols_to_drop), 'columns from', df_ncols
, '\nDim after dropping', len(cols_to_drop), 'columns:', dfm2_mis.shape
, '\n===========================================')
else:
print('FAIL: Dimensions mismatch'
, '\nOriginal no. of cols:', df_ncols
, '\nExpected:', df_ncols - len(cols_to_drop)
, '\nGot:', dfm2_mis.shape[1])
sys.exit()
del(df_ncols)
#%%=====================================================================
#4) reorder columnn
print('Reordering', dfm2_mis.shape[1], 'columns'
, '\n===============================================')
#dfm2_mis.columns
column_order = ['mutation',
'mutationinformation',
'wild_type',
'position',
'mutant_type',
#'chr_num_allele',
'ref_allele',
'alt_allele',
'mut_info_f1',
'mut_info_f2',
'mut_type',
'gene_id',
'gene_name',
'chromosome_number',
#'afs
'af_kin',
'or_kin',
# 'ors_logistic',
# 'ors_chi_cus',
# 'ors_fisher',
'pwald_kin',
# 'pvals_logistic',
# 'pvals_fisher',
# 'ci_lb_fisher',
# 'ci_ub_fisher' ,
'beta_kin',
'se_kin',
'logl_H1_kin',
'l_remle_kin',
# 'stat_chi',
# 'pvals_chi',
# 'n_diff',
# 'tot_diff',
'n_miss',
#'wt_3let',
#'mt_3let'
]
if (len(column_order) == dfm2_mis.shape[1] and pd.DataFrame(column_order).isin(dfm2_mis.columns).all()).all():
print('PASS: Column order generated for', len(column_order), 'columns'
, '\nColumn names match to perform reordering'
, '\nApplying column order to df...' )
orkin_linked = dfm2_mis[column_order]
else:
print('FAIL: Mismatch in no. of cols to reorder'
, '\nNo. of cols in df to reorder:', dfm2_mis.shape[1]
, '\nOrder generated for:', len(column_order), 'columns'
, '\n', dfm2_mis.shape[1], 'should match', len(column_order))
sys.exit()
# sanity check after reassigning columns
if orkin_linked.shape == dfm2_mis.shape and set(orkin_linked.columns) == set(dfm2_mis.columns):
print('PASS: Successfully formatted df with rearranged columns')
else:
sys.exit('FAIL: something went wrong when rearranging columns!')
# converting position and chromosome number to numeric
orkin_linked.dtypes
#orkin_linked['chromosome_number'] = pd.to_numeric(orkin_linked['chromosome_number'])
orkin_linked[['chromosome_number', 'position']] = orkin_linked[['chromosome_number', 'position']].apply(pd.to_numeric)
orkin_linked.dtypes
# write frequency of position counts
orkin_pos = pd.DataFrame(orkin_linked['position'])
z = orkin_pos['position'].value_counts()
z1 = z.to_dict()
orkin_pos['or_pos_count'] = orkin_pos['position'].map(z1)
orkin_pos['or_pos_count'].value_counts()
orkin_linked['position']
foo = orkin_linked['position'].value_counts()
# order df by position
orkin_linked_o = orkin_linked.sort_values(by = ['position'])
bar = orkin_linked_o['position'].value_counts()
if (foo == bar).all():
print('PASS: df reorderd by position for output'
, '\nWriting output file')
else:
print('FAIL: could not reorder by position')
sys.exit()
#%% write file
print('\n====================================================================='
, '\nWriting output file:\n', outfile_or_kin
, '\nNo. of rows:', len(dfm2_mis)
, '\nNo. of cols:', len(dfm2_mis.columns))
orkin_linked_o.to_csv(outfile_or_kin, index = False)
#%% diff b/w allele0 and 1: or_df
#https://stackoverflow.com/questions/40348541/pandas-diff-with-string
#df = or_df.iloc[[5, 15, 17, 19, 34]]
#df[['alt_allele0','ref_allele1']].ne(df[['alt_allele0','ref_allele1']].shift()).any(axis=1).astype(int)
#df[['alt_allele0','ref_allele1']].ne(df[['alt_allele0','ref_allele1']].shift()).any(axis=1).astype(int)