cohort.py

from .utils import *

import os
import numpy as np
import pandas as pd
import scipy.stats as sc
from sklearn.preprocessing import StandardScaler
import itertools as it

class Cohort(object):
	"""
	Patient cohort object for patient proteomics data.

	Dataframes, variables, and functions facilitating the processing, analysis, and integration of the cohort data.

	Parameters
	----------
	cohort: str
		name of the patient cohort

	file_dir: str
		directory where replicate dataframe and 
		sample group membership dataframe file names are located

	replicates_file: str
		name of the replicates dataframe file

		A proteins x B replicates
		comma (*.csv) or tab (*.tsv) delimited

		replicate = "SampleName" + "_Rep[0-9]"

	sample_groups_file: str
		name of the sample group file
		comma (*.csv) or tab (*.tsv) delimited

		N groups x M samples

		sample = "SampleName"

	data_dir: str
		directory where extra data files are located
		
	uniprot_file: str
		name of the uniprot database flat file located in data_dir

	Examples
	--------
	>>>c = cohorts.Cohort(cohort='cohort_name',
		file_dir="path/to/files/"
		replicates_file="file_name",
		sample_groups_file="sample_groups_file_name",
		data_dir="path/to/data/dir/",
		uniprot_file="uniprot_flat_file"
		)
	>>>c.set_replicates_hq()
	>>>c.set_trans_replicates_hq()
	>>>c.set_samples_hq()
	>>>c.set_trans_samples_hq()

	"""

	tests = [ 
					( "t-test",sc.ttest_ind ),
					("Wilcoxon_RankSum_test",sc.ranksums)
				]


	def __init__(self,cohort='cumc',
		replicates_file=None,sample_groups_file=None,
		uniprot_file=None,
		file_dir="",data_dir="../../data/"):

		self.cwd = os.getcwd()
		self.data_dir = data_dir
		self.file_dir = file_dir
		self.cohort = cohort
		self.replicates_file = replicates_file
		self.sample_groups_file = sample_groups_file
		self.uniprot_file = uniprot_file
		self.raw_samples = None
		self.raw_replicates = None
		self.replicates_hq = None
		self.trans_replicates_hq = None
		self.samples_hq = None
		self.trans_samples_hq = None
		self.sample_replicate_dictionary = None
		self.samples = None
		self.replicates = None
		self.proteins = None
		self.replicate_groups = None
		self.sample_groups = None
		self.tidy_replicate_groups = None
		self.tidy_sample_groups = None
		self.groups = None
		self.ref = None
		self.treat = None
		self.set_replicates_file()
		self.set_sample_groups_file()
		self.set_raw_replicates()
		self.set_replicates()
		self.set_sample_replicate_dictionary()
		self.set_samples()
		self.set_sample_groups()
		self.set_replicate_groups()
		self.set_raw_samples()
		self.set_tidy_sample_groups()
		self.set_tidy_replicate_groups()
		self.set_proteins()
		self.set_groups()
		self.set_ref()
		self.set_treat()

		#extra parameters/data objects that can be 
		#attributed to an object instance
		self.data = {}
		self.params = {}

		
	#SET FUNCTIONS
	
	def set_replicates_file(self):
		"""
		Setting the replicates file string.
		Combination of the file directory string and the replicates file string

		A csv file is needed!
		Parameters
		----------
		None

		"""

		self.replicates_file = str(self.file_dir) + str(self.replicates_file)

	def set_sample_groups_file(self):
		"""
		Setting the replicate_groups string
		Combination of the path string and the replicates file string

		A csv file is needed!
		
		Parameters
		----------
		None

		"""
		self.sample_groups_file = str(self.file_dir) + str(self.sample_groups_file)

	def set_uniprot_file(self):
		"""
		Setting the path to the uniprot flat file

		Parameters
		----------
		None

		"""
		self.uniprot_file = str(self.data_dir) + str(self.uniprot_file)

	def set_raw_replicates(self):
		"""
		Setting raw replicate dataframe from files given to object

		Parameters
		----------
		None

		"""		

		#checking if tab separated (tsv) or comma separated (csv) format
		splt = self.replicates_file.split(".")
		format = splt[len(splt)-1][0]
		if format is 'c':
			self.raw_replicates = pd.read_csv(self.replicates_file,delimiter=",",index_col=0)
		else:
			self.raw_replicates = pd.read_csv(self.replicates_file,delimiter="\t",index_col=0)

	def set_replicates(self):
		"""
		Setting replicate names 

		Depends on : raw_replicates

		Parameters
		----------
		None

		"""

		self.replicates = self.raw_replicates.columns.tolist()

	def set_sample_replicate_dictionary(self):
		"""
		Setting dictionary of samples (keys) to replicates (values)

		Depends on : replicates

		Parameters
		----------
		None

		"""

		#get replicates
		replicates = self.replicates

		#strip replicate indicator to get array of sample names for replicates
		duplicate_samples = [y[0] for y in [x for x in np.chararray.split(replicates,"_")]]

		#set previous to series for easy unique sample name retrieval
		samples = pd.Series(duplicate_samples).unique()

		#initialize dictionary
		dictionary = {}

		#loop through unique samples
		for key in samples:
			#get indice of replicates for sample
			replicate_inds = pd.Series(duplicate_samples).isin([key])

			#get replicates for sample
			values = np.array(replicates)[replicate_inds.tolist()]

			#set key:value pair (sample : replicates) in dictionary
			dictionary[key] = values.tolist()

		self.sample_replicate_dictionary = dictionary

	def set_raw_samples(self,agg='mean'):
		"""
		Setting raw sample dataframe 

		Take statistic of replicates with particular sample membership

		Depends on : raw_replicates and make_df_samples()

		Parameters
		----------

		agg: {'mean','median', 'variance'}
			statistic to apply to replicate values for sample membership

		"""

		df = self.raw_replicates.fillna(0)
		
		self.raw_samples = self.make_df_samples(df,agg=agg)

	def set_replicates_hq(self,
		uniprot_annot=False,
		annot_status='reviewed',
		quant_least_reps_per_samps=False,
		n_reps=1,
		all_reps_quant=False,
		threshold=88,
		intrasamp_var=False,
		higher=False):
		"""
		Setting proteins in raw dataframe to be reviewed by Uniprot

		Depends on : raw_replicates, proteins_in_n_replicates_per_samples, proteins_quant_all_reps, proteins_with_uniprot_annot, proteins_by_intrasample_variability

		Parameters
		----------
		
		uniprot_annot {True,False}
			Subset proteins by annot=ation status in uniprot database
		annot_status {'reviewed','unreviewed'}
			Annotation status in uniprot
		quant_least_reps_per_samps {True,False}
			Subset proteins by the number of replicates quantified per sample
		n_reps
			Number of replicates quantified per sample. See above
		all_reps_quant {True,False}
			Subset proteins by only those quantified in every replicate per sample
		intrasamp_var {True,False}
			Subset proteins by the amount of replicate variance per sample
		threshold
			Quantile to subset variance from. See above
		higher {True,False}
			Indicates whether to subset by proteins with a high annotation score from Uniprot

		"""

		#get raw data
		df = self.raw_replicates.fillna(0)

		#subset to have proteins found in atleast sufficient_reps 
		#per sample for all samples
		if quant_least_reps_per_samps:
			prots = self.proteins_in_n_replicates_per_samples(df,n_reps=n_reps)
			df = df.loc[prots]

		#subset to have proteins found in atleast n_reps 
		#per sample for all samples
		if all_reps_quant:
			prots = self.proteins_quant_all_reps(df)
			df = df.loc[prots]

		#subset to have proteins with high annotation score
		if uniprot_annot:
			#set uniprot file if asked for, for protein filtering
			self.set_uniprot_file()
			prots = self.proteins_with_uniprot_annot(df,status=annot_status)
			df = df.loc[prots]

		# subset to have proteins with certain intrasample variability
		if intrasamp_var:

			prots = self.proteins_by_intrasample_variability(df,
							higher=higher,
							threshold=threshold)
			df = df.loc[prots]

		self.replicates_hq = df

	def set_trans_replicates_hq(self,
		trans='None',
		add_small=False,
		stat_thresh=False,
		threshold=95,
		higher=False,
		statistic='variance'):
		"""
		Transforming hq values with different functions, by default log1p.
		
		Depends on : replicates_hq, proteins_by_statistic_threshold

		Parameters
		----------

		trans: numerical transformation. Default: scikitlearn's StandardScaler
			Indicates how to transform the raw protein values
		add_small
			Add small value if any dataframe value is 0 - when log transforming
		stat_threshold
			Subsetting dataframe by a statistic threshold
		threshold [0,100]
			Quantile for statistic threshold
		higher
			Subset above or below statistic threshold
		statistic {'mean','median','variance'}
			Statistic to be applied

		"""

		#set hq dataframe; make sure there's no NaN values
		df = self.replicates_hq

		#instantiate/declare/set up Standard scaler model from
		#scikitlearn on data
		scaler = StandardScaler()
		scaler.fit(df)

		#list of name-function pairs
		func = [('log1p', np.log1p),
				('log2', np.log2),
				('sklearn', scaler.transform),
				('rank_normalize', rank_normalize),
				('quantile_normalize', quantileNormalize),
				('None', pd.DataFrame.copy)]

		#get appropriate list index for function from trans string
		m = [x for x in range(len(func)) if trans in func[x][0]][0]

		#add small epsilon value to all protein values
		if add_small:
			eps = 0.0001
			df = df.applymap(lambda x : x + eps)

		#apply function from appropriate function list index
		data = pd.DataFrame(func[m][1](df),
								index=df.index,
								columns=df.columns
							)

		#subset to have proteins that meet a certain statistical threshold
		if stat_thresh:
			prots = self.proteins_by_statistic_threshold(data,
							statistic=statistic,
							threshold=threshold,
							higher=higher)
			data = data.loc[prots]

		self.trans_replicates_hq = data

	def set_samples_hq(self,
		agg='mean',
		uniprot_annot=False,
		annot_status='reviewed'):
		"""
		Setting sample dataframe from processed replicate dataframe

		Depends on : trans_replicates_hq, make_df_samples
		Parameters
		----------
		
		agg: {'mean','median', 'variance'}
			statistic to apply to replicate values for sample membership
		uniprot_annot {True,False}
			Subset proteins by annot=ation status in uniprot database
		annot_status {'reviewed','unreviewed'}
			Annotation status in uniprot
		"""

		#get processed replicates
		df = self.trans_replicates_hq

		#subset to have proteins with high annotation score
		if uniprot_annot:
			#set uniprot file if asked for, for protein filtering
			self.set_uniprot_file()
			prots = self.proteins_with_uniprot_annot(df,status=annot_status)
			df = df.loc[prots]

		self.samples_hq = self.make_df_samples( df , agg = agg )

	def set_trans_samples_hq(self,
		trans='None',
		add_small=False,
		stat_thresh=False,
		threshold=95,
		higher=False,
		statistic='variance'):
		"""
		Transforming hq values with different functions, by default log1p.
		
		Depends on : samples_hq, proteins_by_statistic_threshold

		Parameters
		----------

		trans: numerical transformation. Default: scikitlearn's StandardScaler
			Indicates how to transform the raw protein values
		add_small
			Add small value if any dataframe value is 0 - when log transforming
		stat_threshold
			Subsetting dataframe by a statistic threshold
		threshold [0,100]
			Quantile for statistic threshold
		higher
			Subset above or below statistic threshold
		statistic {'mean','median','variance'}
			Statistic to be applied

		"""

		#set hq dataframe; make sure there's no NaN values
		data = self.samples_hq

		#instantiate/declare/set up Standard scaler model from
		#scikitlearn on data
		scaler = StandardScaler()
		scaler.fit(data)

		#only do log (other than log1p) transformations if values are
		#all nonzero
		if len(np.where(data.as_matrix().ravel()==0)[0])>0:
			add_small = True
			print('There are zero values in the sample dataframe, a small number epsilon must be added to the protein values when there are zero values. Adding small epsilon...')

		#list of name-function pairs
		func = [('log1p', np.log1p),
				('log2', np.log2),
				('sklearn', scaler.transform),
				('rank_normalize', rank_normalize),
				('quantile_normalize', quantileNormalize),
				('None', pd.DataFrame.copy)]

		#get appropriate list index for function from trans string
		m = [x for x in range(len(func)) if trans in func[x][0]][0]

		#add small epsilon value to all protein values
		if add_small:
			eps = 0.0001
			data = data.applymap(lambda x : x + eps)

		#apply function from appropriate function list index
		data = pd.DataFrame(func[m][1](data),
								index=data.index,
								columns=data.columns
							)

		#subset to have proteins that meet a certain statistical threshold
		if stat_thresh:
			prots = self.proteins_by_statistic_threshold(data,
							statistic=statistic,
							threshold=threshold,
							higher=higher)
			data = data.loc[prots]

		self.trans_samples_hq = data

	def set_samples(self):
		"""
		Setting patient names from raw dataframe

		Depends on : sample_replicate_dictionary

		Parameters
		----------
		None

		"""
		self.samples = [x for x in self.sample_replicate_dictionary.keys()]

	def set_proteins(self):
		"""
		Setting protein ids from raw dataframe

		Depends on raw_replicates

		Parameters
		----------
		None

		"""
		self.proteins = self.raw_replicates.index.values

	def set_sample_groups(self,file=None):
		"""
		Setting group membership of samples dataframe where 1 indicates membership and 0 no membership. 

		Parameters
		----------
		file
			file to read sample groups data. Defaults to parameters from cohorts instantiation
		"""		

		#checking if tab separated (tsv) or comma separated (csv) format
		splt = self.sample_groups_file.split(".")
		format = splt[len(splt)-1][0]
		if format is 'c':
			self.sample_groups = pd.read_csv(self.sample_groups_file,delimiter=",",index_col=0)
		else:
			self.sample_groups = pd.read_csv(self.sample_groups_file,delimiter="\t",index_col=0)
			
	def set_df_replicate_groups(self):
		"""
		Aggregating df_replicate_groups to derive sample group 
		membership

		Parameters
		----------
		None

		"""

		mats = {}
		for samp in self.samples:
			single = self.sample_groups.loc[:,samp]
			reps = self.sample_replicate_dictionary[samp]
			for r in reps:
				mats[r] = single

		return pd.DataFrame.from_dict(mats)

	def set_replicate_groups(self,file=None):
		"""
		Setting group membership of replicate dataframe where 1 indicates membership and 0 no membership. 
		
		Parameters
		----------
		file
			file to read replicate groups data. Defaults to parameters from cohorts instantiation

		"""
			
		self.replicate_groups = self.set_df_replicate_groups()

	def set_groups(self):
		"""
		Setting groups from sample groups dataframe

		Parameters
		----------
		None
		
		"""

		self.groups = self.sample_groups.index.values
 
	def set_ref(self,ref='NL'):
		"""
		Setting reference group (hard coded to be normal patients or 'NL', but can change this after declaration)

		Parameters
		----------

		ref: string
			Reference group name


		"""

		#get ref index
		ind = np.where(self.groups==ref)[0]

		#make sure there's only one trt name in all groups
		if len(ind)==1:

			self.ref = self.groups[ind]

		else:

			self.ref = np.asarray(self.groups[0],dtype='object')

	def set_treat(self,trt='PGD'):
		"""
		Setting treatment group

		Parameters
		----------

		trt: string
			Treatment group name
		"""

		#get trt index
		ind = np.where(self.groups==trt)[0]

		#make sure there's only one trt name in all groups
		if len(ind)==1:

			self.treat = self.groups[ind]
		else:

			self.treat = np.asarray(self.groups[1],dtype='object')
			
	def set_tidy_replicate_groups(self):
		"""
		Setting tidy group membership of replicates, where each replicate is the observation and group membership is an attribute (column)

		Parameters
		----------
		None
		
		"""

		#copy sample groups dataframe
		df = self.replicate_groups.T.copy()

		#set Replicates as column in dataframe
		df.loc[:,'Replicates'] = df.index

		#melt to tidy dataframe
		melted = pd.melt(df, 
							id_vars=['Replicates'], 
							value_vars=df.columns.tolist()[0:len(df.columns)-1],
							var_name='Groups'
						)

		#filter for group membership to samples
		melted_filtered = melted.query('value != 0')

		#delete value {0,1} column-unnecessary since it now redundantly indicates membership 
		del melted_filtered['value']

		#set Samples dtype as string
		melted_filtered.loc[:,'Replicates'] = melted_filtered['Replicates'].astype(str)

		self.tidy_replicate_groups = melted_filtered

	def set_tidy_sample_groups(self):
		"""
		Setting tidy group membership of samples, where each sample is the observation and group membership is an attribute (column)

		Parameters
		----------
		None
		
		"""

		#copy sample groups dataframe
		df = self.sample_groups.copy()

		#set Samples as column in dataframe
		df.loc[:,'Samples'] = df.index

		#melt to tidy dataframe
		melted = pd.melt(df, 
							id_vars=['Samples'], 
							value_vars=df.columns.tolist()[0:len(df.columns)-1],
							var_name='Groups'
						)

		#filter for group membership to samples
		melted_filtered = melted.query('value != 0')

		#delete value {0,1} column-unnecessary since it now redundantly indicates membership 
		del melted_filtered['value']

		#set Samples dtype as string
		melted_filtered.loc[:,'Samples'] = melted_filtered['Samples'].astype(str)

		self.tidy_sample_groups = melted_filtered

	def get_protein_annotations(self,string='hq'):

		'''
		Get Uniprot protein ontology data

		Parameters
		----------
		string : string
			Indicate ontology data for either uniprot expert-curated or 'reviewed' proteins or all proteins from raw

		Output
		-------
		tab : Ontology dataframe
		'''

		#get uniprot ontology
		tab = get_uniprot_table()

		#subset ontology with proteins of interest
		if string=='hq':

			inds = tab.index.isin(self.proteins)

			return tab[inds]

		if string=='all':

			return tab

	def get_uniprot_table(self):
		"""
		Upload uniprot database flat file

		Parameters
		----------
		None

		"""
	
		tab = pd.read_csv(self.data_dir+self.uniprot_file,delimiter="\t",index_col=0)

		return tab

	#ANALYSIS FUNCTIONS

	def manual_feature_extraction(self,df):
		"""
		Query protein quantification between reference and other groups

		Parameters
		----------
		df
			dataframe to manually extract features
		"""
		
		#load raw and sample group dataframes
		df_samples = df
		df_sample_groups = self.sample_groups

		#make rownames-presence/absence/mixed conditions of proteins amongst samples
		val_grps = ('allq', 'allnotq', 'mixed')
		tmp = list(it.product(val_grps,val_grps))
		rownames = []
		for i in tmp:
			rownames.append("_".join(i))

		#set column names-reference/indicator scenarios
		x = self.groups != self.ref[0]
		t = [ self.ref[0] + '/' + x for x in self.groups[x] ]
		colnames = tuple(t)
		

		#populate length and protein array dataframes
		df_len = pd.DataFrame(index=rownames,columns=colnames)
		df_arr = pd.DataFrame(index=rownames,columns=colnames)

		#populate each element with an empty array or list
		for i, j in it.product(range(df_len.shape[0]),range(df_len.shape[1])):
			df_len.iloc[i][j] = ()
		for i, j in it.product(range(df_arr.shape[0]),range(df_arr.shape[1])):
			df_arr.iloc[i][j] = []

		#make list of functions for assessing presence/absence/mixed protein status
		singlefuncs = [ ( "allq", allq ) , 
						("allnotq", allnotq ) ,
						("mixed", mixed ) 
					]

		#set reference group
		ref_grp = self.ref[0]

		#loop through indicators to assess presense/absence/mixed proteins for each reference/indicator scenario and populate length/protein array dataframes
		for i in self.groups[x]:

			#set reference/indicator scenario
			gr1 = ref_grp
			gr2 = i
			colname = gr1+"/"+gr2
			print(colname)
			
			#make arrays of proteins for reference samples for each assessment condition 
			gr1_arr = []
			tmp = df_samples.T
			arr1 = df_sample_groups.loc[gr1] == 1
			X = tmp[arr1.values].transpose()
			for l in range(len(val_grps)):
				tmp = singlefuncs[l][1](X)#slow
				gr1_arr.append(tmp)
		
			#make arrays of proteins for reference samples for each assessment condition 
			gr2_arr = []
			tmp = df_samples.T
			arr2 = df_sample_groups.loc[gr2] == 1
			X = tmp[arr2.values].transpose()
			for k in range(len(val_grps)):
				tmp = singlefuncs[k][1](X)
				gr2_arr.append(tmp)
			
			#for each combination of gr1_arr and gr2_arr, do intersection and append to corresponding row in column of data frame. This gives proteins in each condition for each reference/indicator scenario
			for m,n in it.product( range( len(gr1_arr) ) , range( len(gr2_arr) ) ):
				rowname = singlefuncs[m][0]+"_"+singlefuncs[n][0]
				df_len.loc[ rowname , colname ] = len ( np.intersect1d( gr1_arr[m], gr2_arr[n] ) ) 
				df_arr.loc[ rowname , colname ] = np.intersect1d( gr1_arr[m], gr2_arr[n] ) 

		#dictionary of length and protein array dataframes
		dictionary = { 'df_len' : df_len, 'df_arr' : df_arr} 

		helper_dictionary = self.make_protein_substraction(dictionary)

		self.data['mfe'] = { 'main' : dictionary, 'helper' : helper_dictionary } 

	def make_protein_substraction(self,dictionary=None):
		'''
		Helper function only for manual_feature_extraction method to do set operations on protein results

		Right now supports difference of proteins

		Parameters
		----------
		dictionary
			dictionary to fill in-fed from manual_feature_extraction

		Output
		------
		Dictionary of protein array length band array from difference ofv manually extracted features between ref and treat groups
		'''
		#declare dictionary from manual feature extraction
		mfe = dictionary

		#set comparisons made in mfe
		comps_grps = mfe['df_len'].columns.tolist()

		#set protein feature comparisons from manual feature extraction
		comps_prots = mfe['df_len'].index.tolist() 

		#make combination of comparisons-resulting dataframe rows and columns
		combs = list(
			it.product(
				comps_grps,
				comps_grps
			)
		)

		#make dictionary of presence,absence, mixed. Values will be NxN dataframe labeled with combs above
		protein_dict = dict.fromkeys(
			comps_prots
		)

		#set dataframe to extract proteins from
		df = mfe['df_arr']

		#loop through dictionary keys which are the protein feature comparisons
		for i in protein_dict.keys():

			#make empty NxN dataframe to store protein lengths and arrays
			empty_len = pd.DataFrame(
					index = comps_grps,
			 		columns = comps_grps
					)

			#make copied empty dataframe to store protein arrays
			empty_diff = empty_len.copy()

			#loopp through dataframe index, columns and store length of difference and proteins from difference of sets
			for j in comps_grps:
				for k in comps_grps:
					diff_arr = set(df.loc[i,j]) - set(df.loc[i,k])
					length = len(diff_arr)
					empty_len.at[j,k] = length
					empty_diff.at[j,k] = diff_arr

			#store dictionary of dataframes as value in dictionary
			protein_dict[i] = { 'df_len' : empty_len, 'df_arr' : empty_diff }

		return protein_dict

	def hypothesis_testing(self,df,df_groups):
		"""
		Hypothesis testing of reference sample proteins versus treatment sample proteins.
		
		Parameters
		----------
		df
			dataframe of values to test
		df_groups
			group membership of samples/replicates in df to test

		Outputs
		-------
		df
			hypothesis test results
		"""

		#get subset dataframes by reference and treatment groups
		df1 = self.get_sub_df(df,df_groups,self.ref[0])
		df2 = self.get_sub_df(df,df_groups,self.treat[0])

		#List of hypothesis test names and functions
		tests = self.tests

		#set dataframe to fill
		df = pd.DataFrame(columns=['Protein','Test','Pvalue','Statistic'])

		#for each hypothesis test name and function
		for i, hyp in enumerate(tests):

			#set test name
			test = hyp[0]

			#for each protein
			for j in range(0,df1.shape[0]):

				#get protein location in dataframes
				firstloc = df1.iloc[j]
				secondloc = df2.iloc[j]

				#make sure the arrays are filled with floats
				a = firstloc.values.astype(float)
				b = secondloc.values.astype(float)

				#make sure the name of the protein is the same in each dataframe
				if firstloc.name == secondloc.name:

					#set protein name
					protein = firstloc.name
				else:
					break
				#set statistic from test
				stat = hyp[1](a,b)[0]

				#set pvalue from test
				pval = hyp[1](a,b)[1]

				#put data in pandas series
				row = pd.Series([protein,test,pval,stat],index=df.columns)

				#add row to dataframe
				df = df.append(row,ignore_index=True)

		#set correct object membership of dataframe columns
		df = df.astype(dtype= {
								"Protein":"str",
								"Test":"str",
								"Pvalue":"float64",
								"Statistic":'float64'
							}
						)

		return df

	def get_sub_df(self,df,df_groups,grp):
		'''
		Helper function in class to get patient-subset dataframe.

		Subset larger dataframe by reference and treatment groups

		Used in hypothesis_testing()

		Parameters
		----------
		df
			proteomics data dataframe given to hypothesis_testing
		df_groups
			groups dataframe given to hypothesis testing
		grp str
			group name for subsetting

		Output
		------
		subsetted dataset
		'''

		#get indices of replicates/samples in groups
		inds = np.where(df_groups.loc[grp] == 1)
        
        #get 
        
		return df.T.iloc[inds].T

	#PROTEIN SUBSET FUNCTIONS

	def proteins_with_uniprot_annot(self,df,status='reviewed'):
		"""
		Get proteins that have high annotation score by uniprot

		Parameters
		----------
		df
			dataframe to subset
		status {'reviewed','unreviewed'}
			protein annotation score from uniprot curation
		
		Output
		------
		subsetted uniprot database
		"""

		return self.uniprot_annot(df,status=status).index.ravel()

	def proteins_in_n_replicates_per_samples(self,df,n_reps=1):
		"""
		Subsetting dataset by proteins that are quantified in more atleast n replicates per sample for all samples

		Parameters
		----------
		df
			dataframe to subset
		n_reps
			Least number of replicates quantified per sample

		Output
		------
		subsetted proteins
		"""
		samp_rep_dict = self.sample_replicate_dictionary
		
		atleast1_reps = []
		for samp in samp_rep_dict.keys():
			reps = samp_rep_dict[samp]
			sub = df.T.loc[reps]
			mask = sub.apply(lambda x : x > 0,axis=1)
			#proteins quantified in more than 1 replicates of a sample
			prot_bool = mask.sum(axis=0) > n_reps
			[atleast1_reps.append(x) for x in prot_bool.index[prot_bool].tolist()]
		
		atleast1_rep_per_samp_prots = np.unique(atleast1_reps)

		return atleast1_rep_per_samp_prots

	def proteins_quant_all_reps(self,df):
		"""
		Helper function returning all proteins that 
		are quantified in every replicate.

		Parameters
		----------
		df
			dataframe to subset

		Output
		------
		subsetted proteins
		"""

		#determine how many reps show quantification of the protein
		num_reps_quant = df.apply(lambda x : sum(x > 0),axis=1)

		#is the num reps all of them?
		all_reps_quant = num_reps_quant.apply(
			lambda x : x == df.shape[1])

		#get protein names
		allquant_prots = df.index[all_reps_quant]

		return allquant_prots

	def proteins_by_statistic_threshold(self,
		df,
		statistic='mean',
		higher=False,
		threshold=80):
		"""
		Proteins that meet a certain statistical threshold
		e.g. proteins that have low variance across the dataset
		
		outputs proteins above/below that threshold

		e.g. threshold=25 gives proteins that are below the 25th quantile
		of the statistical function. Threshold=80 gives proteins that are 
		below the 80th quantile of the statistical function
		
		Parameters
		----------
		df
			dataframe to subset
		statistic: {'mean','median', 'variance'}
			statistic to apply across values
		higher {True,False}
			Indicates whether to subset by proteins with a high annotation score from Uniprot
		threshold [0,100]
			Quantile for statistic threshold

		Output
		------
		subsetted proteins
		
		"""
		#possible statistical functions to threshold by
		funcs = {
			'mean' : np.mean,
			'median' : np.median,
			'variance' : np.var,
			'std' : np.std
		}
		
		#collect proteins in dataset
		prots = df.index
		
		#create series by applying statistic
		series = df.apply(funcs[statistic],axis=1)
		
		#store dataset value of threshold
		thresh = (series.describe(
							percentiles=np.arange(0,1,.01)
						)
						.loc['0%':]
						.iloc[threshold]
				)
		
		#determine direction of threshold
		#if greater than 50, greater than or equal to,
		#otherwise, less than
		if higher:
			return prots[series.where(series > thresh).notnull()]
		else:
			return prots[series.where(series <= thresh).notnull()]
		
	def proteins_by_intrasample_variability(self,
		df,higher=False,threshold=80):
		"""
		Output proteins below a certain variance threshold for 
		replicates within all samples

		Parameters
		----------
		df
			dataframe to subset
		higher {True,False}
			Indicates whether to subset by proteins with a high annotation score from Uniprot
		threshold [0,100]
			Quantile for statistic threshold

		Output
		------
		subsetted proteins

		"""

		#transpose
		df = df.T

		samp_rep_dict = self.sample_replicate_dictionary

		prots = pd.Index([])

		for samp in samp_rep_dict.keys():
			reps = samp_rep_dict.get(samp)
			sub = df.loc[np.asarray(reps)]
			prots = prots.append( \
				self.proteins_by_statistic_threshold(df=sub.T,
					higher=False,
					statistic='variance',
					threshold=threshold))

		return prots.unique()

	#HELPER FUNCTIONS
	def uniprot_annot(self, df,status='reviewed'):
		"""
		Subset dataset by proteins that have high annotation score by uniprot

		Parameters
		----------
		df
			dataframe to subset
		annot_status {'reviewed','unreviewed'}
			Annotation status in uniprot

		Output
		------
		subsetted dataset
		"""

		#get uniprot data
		annot = self.get_protein_annotations('hq')

		#get reviewed uniprot data
		annot_reviewed = annot[annot['Status'].str.match(status)]

		#getting all uniprot reviewed proteins
		cprots = annot_reviewed.iloc[:,0].index

		#filter raw for reviewed proteins
		raw_reviewed = df[df.index.isin(cprots)]

		#set new protein column
		tmp = raw_reviewed.assign(Proteins=raw_reviewed.index.values)

		#group proteins because there might be duplicates
		grouped = tmp.groupby('Proteins')

		#get one of the duplicated proteins
		reviewed = grouped.max()

		return reviewed

	def make_df_samples(self,df,agg='mean'):
		"""
		From df_replicates, make df_samples by applying to the replicates for a sample the estimator indicated in the helper function

		Parameters
		----------
		df
			replicates dataframe to aggregate
		agg: {'mean','median', 'variance'}
			statistic to apply to replicate values for sample membership

		Output
		------
		subsetted dataset
		"""
		df_replicates = df
		acc = df.index.tolist()
		samples = self.samples
		dictionary = self.sample_replicate_dictionary

		# initializing empty, desired dataframe of samples by proteins
		df_samples = pd.DataFrame(columns=samples)

		#for each sample, apply the helper function to df_replicates sub dataframe to make df_samples
		for sample in dictionary.keys():

			#dataframe of replicates for a given sample
			holder = df_replicates[dictionary[sample]]

			#determine sample protein value from replicate values
			df = holder.agg(func=agg,axis=1)

			#put in df to larger df_samples
			df_samples[sample] = df

		#add index-protein accessions
		df_samples.index = acc

		return df_samples

class IntegratedCohort(object):
	"""
	Integrated patient cohorts object for shared patient clinical and proteomics data

	Dataframes, variables, and functions facilitating the processing, analysis, and integration of the integrated cohort data.
	
	Parameters
	----------
	cohort_objs
		dictionary of individual cohort objects to integrate through shared clinical and proteomic data

	Examples
	--------

	"""

	def __init__(self,cohort_objs={}):

		self.cohort_objs = cohort_objs
		self.df = None
		self.df_groups = None

	def sayHello(self):

		return "Hello!"

	def integrate_cohorts(self,treat_name="PGD",ref_name="NL",dataset_type='samples_hq',groups_type='sample_groups'):
		'''
		For created dataset integrating cohorts data
		
		Uses treat and ref attributes of each object to integrate the sample groups. Uses treat_name and ref_name to rename the treat and ref groups (in case in cohorts the treat and ref groups are named differently but should be integrated)
		
		Can set the object attribute type to use for joining e.g. the samples or replicates dataset
		
		Returns:
		--------
		
		integrated_samples_dataset:
			common proteins x all cohort samples
		integrated_sample_groups
			groups x all cohort samples
		response
			response_group x samples
		'''
		dfs = {}
		df_groups = {}
		for cohort,obj in self.cohort_objs.items():
			dfs[cohort] = getattr(obj,dataset_type)
			df = getattr(obj,groups_type)
			df_groups[cohort] = df

		self.df = multi_df_join(dfs).dropna()
		self.df_groups = multi_df_join(df_groups).dropna()

def make_cohorts_dict(names,
					file_dirs,
					treats,refs,
					replicates_files,
					sample_groups_files,
					data_dir="../../data/",
					uniprot_file="uniprot-all_20171124.tab.gz"):
	"""
	Create cohort objects via dictionaries of attributes where the key of each dfictionary is the name of the cohort object
	
	Parameters:
	----------


	Returns:
	------

	objs
		Dictionary where keys are each cohort object name and values are the instantiaated cohort object.
	"""
	if len(names)==0:
		return {}
	else:
		objs = {}
		for cohort in names:
			obj = Cohort(cohort=cohort,
								data_dir=data_dir,file_dir=file_dirs[cohort],
								replicates_file=replicates_files[cohort],
								sample_groups_file=sample_groups_files[cohort],
								uniprot_file=uniprot_file)
			obj.set_ref(refs[cohort])
			obj.set_treat(treats[cohort])
			objs[cohort] = obj
		return objs

if __name__=="__main__":
	"""
	Execute
	"""
	name =  'cumc'
	c = cohorts.Cohort(name)