LSHTM_analysis/mcsm/mcsm.py

494 lines
22 KiB
Python

#%% load packages
import os,sys
import subprocess
import argparse
import requests
import re
import time
from bs4 import BeautifulSoup
import pandas as pd
from pandas.api.types import is_string_dtype
from pandas.api.types import is_numeric_dtype
import numpy as np
#from csv import reader
from mcsm import *
#==============================
#%% global variables for defs
#==============================
#%%
def format_data(data_file):
"""
Read file containing SNPs for mcsm analysis and remove duplicates
@param data_file csv file containing nsSNPs for given drug and gene.
csv file format:
single column with no headers with nsSNP format as below:
A1B
B2C
@type data_file: string
@return unique SNPs
@type list
"""
data = pd.read_csv(data_file, header = None, index_col = False)
data = data.drop_duplicates()
mutation_list = data[0].tolist()
# print(data.head())
return mutation_list
# FIXME: documentation
def request_calculation(pdb_file, mutation, chain, ligand_id, wt_affinity, prediction_url, output_dir, gene_name, host):
"""
Makes a POST request for a ligand affinity prediction.
@param pdb_file: valid path to pdb structure
@type string
@param mutation: single mutation of the format: {WT}<POS>{Mut}
@type string
@param chain: single-letter(caps)
@type chr
@param lig_id: 3-letter code (should match pdb file)
@type string
@param wt affinity: in nM
@type number
@param prediction_url: mcsm url for prediction
@type string
@return response object
@type object
"""
with open(pdb_file, "rb") as pdb_file:
files = {"wild": pdb_file}
body = {
"mutation": mutation,
"chain": chain,
"lig_id": ligand_id,
"affin_wt": wt_affinity
}
response = requests.post(prediction_url, files = files, data = body)
#print(response.status_code)
#result_status = response.raise_for_status()
if response.history:
# if result_status is not None: # doesn't work!
print('PASS: valid mutation submitted. Fetching result url')
#return response
url_match = re.search('/mcsm_lig/results_prediction/.+(?=")', response.text)
url = host + url_match.group()
#===============
# writing file: result urls
#===============
out_url_file = output_dir + '/' + gene_name.lower() + '_result_urls.txt'
myfile = open(out_url_file, 'a')
myfile.write(url + '\n')
myfile.close()
else:
print('ERROR: invalid mutation! Wild-type residue doesn\'t match pdb file.'
, '\nSkipping to the next mutation in file...')
#===============
# writing file: invalid mutations
#===============
out_error_file = output_dir + '/' + gene_name.lower() + '_errors.txt'
failed_muts = open(out_error_file, 'a')
failed_muts.write(mutation + '\n')
failed_muts.close()
#=======================================================================
def scrape_results(result_url):
"""
Extract results data using the result url
@params result_url: txt file containing result url
one per line for each mutation
@type string
returns: mcsm prediction results (raw)
@type chr
"""
result_response = requests.get(result_url)
# if results_response is not None:
# page = results_page.text
if result_response.status_code == 200:
print('Fetching results')
# extract results using the html parser
soup = BeautifulSoup(result_response.text, features = 'html.parser')
# print(soup)
web_result_raw = soup.find(class_ = 'span4').get_text()
#metatags = soup.find_all('meta')
metatags = soup.find_all('meta', attrs={'http-equiv':'refresh'})
#print('meta tags:', metatags)
if metatags:
print('WARNING: Submission not ready for URL:', result_url)
# TODO: Add logging
#if debug:
# debug.warning('submission not ready for URL:', result_url)
else:
return web_result_raw
else:
sys.exit('FAIL: Could not fetch results'
, '\nCheck if url is valid')
def build_result_dict(web_result_raw):
"""
Build dict of mcsm output for a single mutation
Format web results which is preformatted to enable building result dict
# preformatted string object: Problematic!
# make format consistent
@params web_result_raw: directly from html parser extraction
@type string
@returns result dict
@type {}
"""
# remove blank lines from web_result_raw
mytext = os.linesep.join([s for s in web_result_raw.splitlines() if s])
# affinity change and DUET stability change cols are are split over
# multiple lines and Mutation information is empty!
mytext = mytext.replace('ange:\n', 'ange: ')
#print(mytext)
# initiliase result_dict
result_dict = {}
for line in mytext.split('\n'):
fields = line.split(':')
#print(fields)
if len(fields) > 1: # since Mutaton information is empty
dict_entry = dict([(x, y) for x, y in zip(fields[::2], fields[1::2])])
result_dict.update(dict_entry)
print(result_dict)
return result_dict
#%%
#=======================================================================
def format_mcsm_output(mcsm_outputcsv):
"""
@param mcsm_outputcsv: file containing mcsm results for all muts
which is the result of build_result_dict() being called for each
mutation and then converting to a pandas df and output as csv.
@type string
@return formatted mcsm output
@type pandas df
"""
#############
# Read file
#############
mcsm_data_raw = pd.read_csv(mcsm_outputcsv, sep = ',')
# strip white space from both ends in all columns
mcsm_data = mcsm_data_raw.apply(lambda x: x.str.strip() if x.dtype == 'object' else x)
dforig_shape = mcsm_data.shape
print('dimensions of input file:', dforig_shape)
#############
# rename cols
#############
# format colnames: all lowercase, remove spaces and use '_' to join
print('Assigning meaningful colnames i.e without spaces and hyphen and reflecting units'
, '\n=======================================================')
my_colnames_dict = {'Predicted Affinity Change': 'PredAffLog' # relevant info from this col will be extracted and the column discarded
, 'Mutation information': 'mutationinformation' # {wild_type}<position>{mutant_type}
, 'Wild-type': 'wild_type' # one letter amino acid code
, 'Position': 'position' # number
, 'Mutant-type': 'mutant_type' # one letter amino acid code
, 'Chain': 'chain' # single letter (caps)
, 'Ligand ID': 'ligand_id' # 3-letter code
, 'Distance to ligand': 'ligand_distance' # angstroms
, 'DUET stability change': 'duet_stability_change'} # in kcal/mol
mcsm_data.rename(columns = my_colnames_dict, inplace = True)
#%%=====================================================================
#################################
# populate mutationinformation
# col which is currently blank
#################################
# populate mutationinformation column:mcsm style muts {WT}<POS>{MUT}
print('Populating column : mutationinformation which is currently empty\n', mcsm_data['mutationinformation'])
mcsm_data['mutationinformation'] = mcsm_data['wild_type'] + mcsm_data['position'].astype(str) + mcsm_data['mutant_type']
print('checking after populating:\n', mcsm_data['mutationinformation']
, '\n=======================================================')
# Remove spaces b/w pasted columns
print('removing white space within column: \mutationinformation')
mcsm_data['mutationinformation'] = mcsm_data['mutationinformation'].str.replace(' ', '')
print('Correctly formatted column: mutationinformation\n', mcsm_data['mutationinformation']
, '\n=======================================================')
#%%=====================================================================
#############
# sanity check: drop dupliate muts
#############
# shouldn't exist as this should be eliminated at the time of running mcsm
print('Sanity check:'
, '\nChecking duplicate mutations')
if mcsm_data['mutationinformation'].duplicated().sum() == 0:
print('PASS: No duplicate mutations detected (as expected)'
, '\nDim of data:', mcsm_data.shape
, '\n===================================================')
else:
print('WARNING: Duplicate mutations detected'
, '\nDim of df with duplicates:', mcsm_data.shape
, 'Removing duplicate entries')
mcsm_data = mcsm_data.drop_duplicates(['mutationinformation'])
print('Dim of data after removing duplicate muts:', mcsm_data.shape
, '\n===========================================================')
#%%=====================================================================
#############
# Create col: duet_outcome
#############
# classification based on DUET stability values
print('Assigning col: duet_outcome based on DUET stability values')
print('Sanity check:')
# count positive values in the DUET column
c = mcsm_data[mcsm_data['duet_stability_change']>=0].count()
DUET_pos = c.get(key = 'duet_stability_change')
# Assign category based on sign (+ve : Stabilising, -ve: Destabilising, Mind the spelling (British spelling))
mcsm_data['duet_outcome'] = np.where(mcsm_data['duet_stability_change']>=0, 'Stabilising', 'Destabilising')
print('DUET Outcome:', mcsm_data['duet_outcome'].value_counts())
#if DUET_pos == mcsm_data['duet_outcome'].value_counts()['Stabilising']:
# print('PASS: DUET outcome assigned correctly')
#else:
# print('FAIL: DUET outcome assigned incorrectly'
# , '\nExpected no. of stabilising mutations:', DUET_pos
# , '\nGot no. of stabilising mutations', mcsm_data['duet_outcome'].value_counts()['Stabilising']
# , '\n======================================================')
#%%=====================================================================
#############
# Extract numeric
# part of ligand_distance col
#############
# Extract only the numeric part from col: ligand_distance
# number: '-?\d+\.?\d*'
mcsm_data['ligand_distance']
print('extracting numeric part of col: ligand_distance')
mcsm_data['ligand_distance'] = mcsm_data['ligand_distance'].str.extract('(\d+\.?\d*)')
print('Ligand Distance:',mcsm_data['ligand_distance'])
#%%=====================================================================
#############
# Create 2 columns:
# ligand_affinity_change and ligand_outcome
#############
# the numerical and categorical parts need to be extracted from column: PredAffLog
# regex used
# numerical part: '-?\d+\.?\d*'
# categorocal part: '\b(\w+ing)\b'
print('Extracting numerical and categorical parts from the col: PredAffLog')
print('to create two columns: ligand_affinity_change and ligand_outcome'
, '\n=======================================================')
# 1) Extracting the predicted affinity change (numerical part)
mcsm_data['ligand_affinity_change'] = mcsm_data['PredAffLog'].str.extract('(-?\d+\.?\d*)', expand = True)
print(mcsm_data['ligand_affinity_change'])
# 2) Extracting the categorical part (Destabillizing and Stabilizing) using word boundary ('ing')
#aff_regex = re.compile(r'\b(\w+ing)\b')
mcsm_data['ligand_outcome']= mcsm_data['PredAffLog'].str.extract(r'(\b\w+ing\b)', expand = True)
print(mcsm_data['ligand_outcome'])
print(mcsm_data['ligand_outcome'].value_counts())
#############
# changing spelling: British
#############
# ensuring spellings are consistent
american_spl = mcsm_data['ligand_outcome'].value_counts()
print('Changing to Bristish spellings for col: ligand_outcome')
mcsm_data['ligand_outcome'].replace({'Destabilizing': 'Destabilising', 'Stabilizing': 'Stabilising'}, inplace = True)
print(mcsm_data['ligand_outcome'].value_counts())
british_spl = mcsm_data['ligand_outcome'].value_counts()
# compare series values since index will differ from spelling change
check = american_spl.values == british_spl.values
if check.all():
print('PASS: spelling change successfull'
, '\nNo. of predicted affinity changes:\n', british_spl
, '\n===================================================')
else:
sys.exit('FAIL: spelling change unsucessfull'
, '\nExpected:\n', american_spl
, '\nGot:\n', british_spl
, '\n===================================================')
#%%=====================================================================
#############
# ensuring corrrect dtype for numeric columns
#############
# check dtype in cols
print('Checking dtypes in all columns:\n', mcsm_data.dtypes
, '\n=======================================================')
print('Converting the following cols to numeric:'
, '\nligand_distance'
, '\nduet_stability_change'
, '\nligand_affinity_change'
, '\n=======================================================')
# using apply method to change stabilty and affinity values to numeric
numeric_cols = ['duet_stability_change', 'ligand_affinity_change', 'ligand_distance']
mcsm_data[numeric_cols] = mcsm_data[numeric_cols].apply(pd.to_numeric)
# check dtype in cols
print('checking dtype after conversion')
cols_check = mcsm_data.select_dtypes(include='float64').columns.isin(numeric_cols)
if cols_check.all():
print('PASS: dtypes for selected cols:', numeric_cols
, '\nchanged to numeric'
, '\n===================================================')
else:
sys.exit('FAIL:dtype change to numeric for selected cols unsuccessful'
, '\n===================================================')
print(mcsm_data.dtypes)
#%%=====================================================================
#############
# scale duet values
#############
# Rescale values in DUET_change col b/w -1 and 1 so negative numbers
# stay neg and pos numbers stay positive
duet_min = mcsm_data['duet_stability_change'].min()
duet_max = mcsm_data['duet_stability_change'].max()
duet_scale = lambda x : x/abs(duet_min) if x < 0 else (x/duet_max if x >= 0 else 'failed')
mcsm_data['duet_scaled'] = mcsm_data['duet_stability_change'].apply(duet_scale)
print('Raw duet scores:\n', mcsm_data['duet_stability_change']
, '\n---------------------------------------------------------------'
, '\nScaled duet scores:\n', mcsm_data['duet_scaled'])
#!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# additional check added
c2 = mcsm_data[mcsm_data['duet_scaled']>=0].count()
DUET_pos2 = c2.get(key = 'duet_scaled')
if DUET_pos == DUET_pos2:
print('\nPASS: DUET values scaled correctly')
else:
print('\nFAIL: DUET values scaled numbers MISmatch'
, '\nExpected number:', DUET_pos
, '\nGot:', DUET_pos2
, '\n======================================================')
#!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
#%%=====================================================================
#############
# scale affinity values
#############
# rescale values in affinity change col b/w -1 and 1 so negative numbers
# stay neg and pos numbers stay positive
aff_min = mcsm_data['ligand_affinity_change'].min()
aff_max = mcsm_data['ligand_affinity_change'].max()
aff_scale = lambda x : x/abs(aff_min) if x < 0 else (x/aff_max if x >= 0 else 'failed')
mcsm_data['affinity_scaled'] = mcsm_data['ligand_affinity_change'].apply(aff_scale)
print('Raw affinity scores:\n', mcsm_data['ligand_affinity_change']
, '\n---------------------------------------------------------------'
, '\nScaled affinity scores:\n', mcsm_data['affinity_scaled'])
#!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# additional check added
c_lig = mcsm_data[mcsm_data['ligand_affinity_change']>=0].count()
Lig_pos = c_lig.get(key = 'ligand_affinity_change')
c_lig2 = mcsm_data[mcsm_data['affinity_scaled']>=0].count()
Lig_pos2 = c_lig2.get(key = 'affinity_scaled')
if Lig_pos == Lig_pos2:
print('\nPASS: Ligand affintiy values scaled correctly')
else:
print('\nFAIL: Ligand affinity values scaled numbers MISmatch'
, '\nExpected number:', Lig_pos
, '\nGot:', Lig_pos2
, '\n======================================================')
#!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
#%%=====================================================================
#############
# adding column: wild_pos
# useful for plots and db
#############
print('Creating column: wild_pos')
mcsm_data['wild_pos'] = mcsm_data['wild_type'] + mcsm_data['position'].astype(str)
print(mcsm_data['wild_pos'].head())
# Remove spaces b/w pasted columns
print('removing white space within created column: wild_pos')
mcsm_data['wild_pos'] = mcsm_data['wild_pos'].str.replace(' ', '')
print('Correctly formatted column: wild_pos\n', mcsm_data['wild_pos'].head()
, '\n=========================================================')
#%%=====================================================================
#############
# adding column: wild_chain_pos
# useful for plots and db and its explicit
#############
print('Creating column: wild_chain_pos')
mcsm_data['wild_chain_pos'] = mcsm_data['wild_type'] + mcsm_data['chain'] + mcsm_data['position'].astype(str)
print(mcsm_data['wild_chain_pos'].head())
# Remove spaces b/w pasted columns
print('removing white space within created column: wild_chain_pos')
mcsm_data['wild_chain_pos'] = mcsm_data['wild_chain_pos'].str.replace(' ', '')
print('Correctly formatted column: wild_chain_pos\n', mcsm_data['wild_chain_pos'].head()
, '\n=========================================================')
#%%=====================================================================
#############
# ensuring corrrect dtype in non-numeric cols
#############
#) char cols
char_cols = ['PredAffLog', 'mutationinformation', 'wild_type', 'mutant_type', 'chain', 'ligand_id', 'duet_outcome', 'ligand_outcome', 'wild_pos', 'wild_chain_pos']
#mcsm_data[char_cols] = mcsm_data[char_cols].astype(str)
cols_check_char = mcsm_data.select_dtypes(include = 'object').columns.isin(char_cols)
if cols_check_char.all():
print('PASS: dtypes for char cols:', char_cols, 'are indeed string'
, '\n===================================================')
else:
sys.exit('FAIL:dtype change to numeric for selected cols unsuccessful'
, '\n===================================================')
#mcsm_data['ligand_distance', 'ligand_affinity_change'].apply(is_numeric_dtype(mcsm_data['ligand_distance', 'ligand_affinity_change']))
print(mcsm_data.dtypes)
#%%=====================================================================
# Removing PredAff log column as it is not needed?
print('Removing col: PredAffLog since relevant info has been extracted from it')
mcsm_data_f = mcsm_data.drop(columns = ['PredAffLog'])
#%%=====================================================================
# sort df by position for convenience
print('Sorting df by position')
mcsm_data_fs = mcsm_data_f.sort_values(by = ['position'])
print('sorted df:\n', mcsm_data_fs.head())
# Ensuring column names are lowercase before output
mcsm_data_fs.columns = mcsm_data_fs.columns.str.lower()
#%%=====================================================================
#############
# sanity check before writing file
#############
expected_ncols_toadd = 6 # beware hardcoding!
dforig_len = dforig_shape[1]
expected_cols = dforig_len + expected_ncols_toadd
if len(mcsm_data_fs.columns) == expected_cols:
print('PASS: formatting successful'
, '\nformatted df has expected no. of cols:', expected_cols
, '\n---------------------------------------------------'
, '\ncolnames:', mcsm_data_fs.columns
, '\n---------------------------------------------------'
, '\ndtypes in cols:', mcsm_data_fs.dtypes
, '\n---------------------------------------------------'
, '\norig data shape:', dforig_shape
, '\nformatted df shape:', mcsm_data_fs.shape
, '\n===================================================')
else:
print('FAIL: something went wrong in formatting df'
, '\nLen of orig df:', dforig_len
, '\nExpected number of cols to add:', expected_ncols_toadd
, '\nExpected no. of cols:', expected_cols, '(', dforig_len, '+', expected_ncols_toadd, ')'
, '\nGot no. of cols:', len(mcsm_data_fs.columns)
, '\nCheck formatting:'
, '\ncheck hardcoded value:', expected_ncols_toadd
, '\nis', expected_ncols_toadd, 'the no. of expected cols to add?'
, '\n===================================================')
sys.exit()
return mcsm_data_fs