import os
import pandas as pd
from BLRun.runner import Runner
[docs]class SINGERunner(Runner):
"""Concrete runner for the SINGE GRN inference algorithm."""
[docs] def run(self):
'''
Function to run SINGE algorithm
'''
# if the parameters aren't specified, then use default parameters
# TODO allow passing in multiple sets of hyperparameters
# these must be in the right order!
params_order = [
'lambda', 'dT', 'num_lags', 'kernel_width',
'prob_zero_removal', 'prob_remove_samples',
'family'
]
default_params = {
'lambda': '0.01',
'dT': '10',
'num_lags': '5',
'kernel_width': '4',
'prob_zero_removal': '0',
'prob_remove_samples': '0.2',
'family': 'gaussian',
'num_replicates': '2',
}
params = self.params
for param, val in default_params.items():
if param not in params:
params[param] = val
num_replicates = int(params['num_replicates'])
replicates = []
for replicate in range(num_replicates):
replicates.append(' '.join('--' + p.replace('_', '-') + ' ' + str(params[p]) for p in params_order) + ' '.join(['', '--replicate', str(replicate), '--ID', str(replicate)]))
params_str = '\n'.join(replicates)
PTData = pd.read_csv(self.input_dir / self.pseudoTimeData,
header = 0, index_col = 0)
colNames = PTData.columns
for idx in range(len(colNames)):
os.makedirs(str(self.working_dir / str(idx)), exist_ok = True)
outFileSymlink = "out" + str(idx)
inputFile = "/usr/working_dir/ExpressionData"+str(idx)+".csv"
inputMat = "/usr/working_dir/ExpressionData"+str(idx)+".mat"
geneListMat = "/usr/working_dir/GeneList"+str(idx)+".mat"
paramsFile = "/usr/working_dir/hyperparameters.txt"
'''
This is a workaround for https://github.com/gitter-lab/SINGE/blob/master/code/parseParams.m#L39
not allowing '/' characters in the outDir parameter.
'''
symlink_out_file = ' '.join(['ln -s', "/usr/working_dir/" + str(idx) + "/", outFileSymlink])
'''
See https://github.com/gitter-lab/SINGE/blob/master/README.md. SINGE expects a data matfile with variables "X" and "ptime",
and a gene_list matfile with the variable "gene_list".
Saving fullKp is a very hacky workaround for https://github.com/gitter-lab/SINGE/blob/master/code/iLasso_for_SINGE.m#L56,
that assumes this input was saved in matfile v7.3 which octave does not support.
'''
convert_input_to_matfile = 'octave -q --eval \\"CSV = csvread(\'' + inputFile + '\'); ' + \
'X = sparse(CSV(2:end,1:end-1).\'); ptime = CSV(2:end,end).\'; ' + \
'Kp2.Kp = single(ptime); Kp2.sumKp = single(ptime*X.\'); fullKp(1, ' + \
str(int(params['dT'])*int(params['num_lags'])) + ') = Kp2; ' + \
'save(\'-v7\',\'' + inputMat + '\', \'X\', \'ptime\', \'fullKp\'); ' + \
'f = fopen(\'' + inputFile + '\'); gene_list = strsplit(fgetl(f), \',\')(1:end-1).\'; fclose(f); ' + \
'save(\'-v7\',\'' + geneListMat + '\', \'gene_list\')\\"'
cmdToRun = ' '.join(['docker run --rm --entrypoint /bin/sh',
f"-v {self.working_dir}:/usr/working_dir",
f'{self.image} -c \"echo \\"',
params_str, '\\" >', paramsFile, '&&', symlink_out_file, '&&', convert_input_to_matfile,
'&& time -v -o', "/usr/working_dir/time" + str(idx) + ".txt",
'/usr/local/SINGE/SINGE.sh /usr/local/MATLAB/MATLAB_Runtime/v94 standalone',
inputMat, geneListMat, outFileSymlink, paramsFile, '\"'])
self._run_docker(cmdToRun, append=(idx > 0))
[docs] def parseOutput(self):
'''
Function to parse outputs from SINGE.
'''
workDir = self.working_dir
PTData = pd.read_csv(self.input_dir / self.pseudoTimeData,
header = 0, index_col = 0)
colNames = PTData.columns
OutSubDF = [0]*len(colNames)
for idx in range(len(colNames)):
# Quit if output directory does not exist
if not (workDir / str(idx) / 'SINGE_Ranked_Edge_List.txt').exists():
print(str(workDir / str(idx) / 'SINGE_Ranked_Edge_List.txt') + ' does not exist, skipping...')
return
# Read output
OutSubDF[idx] = pd.read_csv(workDir / str(idx) / 'SINGE_Ranked_Edge_List.txt',
sep = '\t', header = 0)
# megre the dataframe by taking the maximum value from each DF
# Code from here:
# https://stackoverflow.com/questions/20383647/pandas-selecting-by-label-sometimes-return-series-sometimes-returns-dataframe
outDF = pd.concat(OutSubDF)
outDF.columns= ['Gene1','Gene2','EdgeWeight']
# Group by rows code is from here:
# https://stackoverflow.com/questions/53114609/pandas-how-to-remove-duplicate-rows-but-keep-all-rows-with-max-value
res = outDF[outDF['EdgeWeight'] == outDF.groupby(['Gene1','Gene2'])['EdgeWeight'].transform('max')]
# Sort values in the dataframe
finalDF = res.sort_values('EdgeWeight', ascending=False)
self._write_ranked_edges(finalDF[['Gene1', 'Gene2', 'EdgeWeight']])