def sigProfilerExtractor(input_type,
output,
input_data,
reference_genome="GRCh37",
opportunity_genome="GRCh37",
context_type="default",
exome=False,
minimum_signatures=1,
maximum_signatures=25,
nmf_replicates=100,
resample=True,
batch_size=1,
cpu=-1,
gpu=False,
nmf_init="nndsvd_min",
precision="single",
matrix_normalization="gmm",
seeds="random",
min_nmf_iterations=10000,
max_nmf_iterations=1000000,
nmf_test_conv=10000,
nmf_tolerance=1e-15,
nnls_add_penalty=0.05,
nnls_remove_penalty=0.01,
de_novo_fit_penalty=0.02,
initial_remove_penalty=0.05,
refit_denovo_signatures=True,
clustering_distance="cosine",
export_probabilities=True,
make_decomposition_plots=True,
stability=0.8,
min_stability=0.2,
combined_stability=1.0,
get_all_signature_matrices=False):
memory_usage()
"""
Extracts mutational signatures from an array of samples.
Parameters
----------
INPUT DATA:-
input_type: A string. Type of input. The type of input should be one of the following:
- "vcf": used for vcf format inputs.
- "matrix": used for table format inputs using a tab seperated file.
output: A string. The name of the output folder. The output folder will be generated in the current working directory.
input_data: A string. Name of the input folder (in case of "vcf" type input) or the input file (in case of "table" type input). The project file or folder should be inside the current working directory. For the "vcf" type input,the project has to be a folder which will contain the vcf files in vcf format or text formats. The "text"type projects have to be a file.
reference_genome: A string, optional. The name of the reference genome. The default reference genome is "GRCh37". This parameter is applicable only if the input_type is "vcf".
opportunity_genome: The build or version of the reference signatures for the reference genome. The default opportunity genome is GRCh37. If the input_type is "vcf", the genome_build automatically matches the input reference genome value.
context_type: A list of strings, optional. The items in the list defines the mutational contexts to be considered to extract the signatures. The default value is "SBS96,DBS78,ID83".
exome: Boolean, optional. Defines if the exomes will be extracted. The default value is "False".
NMF RUNS:-
minimum_signature: A positive integer, optional. The minimum number of signatures to be extracted. The default value is 1
maximum_signatures: A positive integer, optional. The maximum number of signatures to be extracted. The default value is 10
nmf_replicates: A positive integer, optional. The number of iteration to be performed to extract each number signature. The default value is 100
resample: Boolean, optional. Default is True. If True, add poisson noise to samples by resampling.
seeds: Boolean. Default is "random". If random, then the seeds for resampling will be random for different analysis.
If not random, then seeds will be obtained from a given path of a .txt file that contains a list of seed.
NMF RUNS:-
matrix_normalization: A string. Method of normalizing the genome matrix before it is analyzed by NMF. Default is "log2". Other options are "gmm", "100X" or "no_normalization".
nmf_init: A String. The initialization algorithm for W and H matrix of NMF. Options are 'random', 'nndsvd', 'nndsvda', 'nndsvdar' and 'nndsvd_min'
Default is 'nndsvd_min'.
precision: A string. Values should be single or double. Default is single.
min_nmf_iterations: An integer. Value defines the minimum number of iterations to be completed before NMF converges. Default is 2000.
max_nmf_iterations: An integer. Value defines the maximum number of iterations to be completed before NMF converges. Default is 200000
nmf_test_conv: An integer. Value definer the number number of iterations to done between checking next convergence.
nmf_tolerance: A float. Value defines the tolerance to achieve to converge.
EXECUTION:-
cpu: An integer, optional. The number of processors to be used to extract the signatures. The default value is -1 which will use all available processors.
gpu:Boolean, optional. Defines if the GPU resource will used if available. Default is False. If True, the GPU resource
will be used in the computation.
batch_size: An integer. Will be effective only if the GPU is used. Defines the number of NMF replicates to be performed
by each CPU during the parallel processing. Default is 1.
SOLUTION ESTIMATION THRESH-HOLDS:-
stability: A float. Default is 0.8. The cutoff thresh-hold of the average stability. Solutions with average stabilities below this thresh-hold will not be considered.
min_stability: A float. Default is 0.2. The cutoff thresh-hold of the minimum stability. Solutions with minimum stabilities below this thresh-hold will not be considered.
combined_stability: A float. Default is 1.0. The cutoff thresh-hold of the combined stability (sum of average and minimum stability). Solutions with combined stabilities below this thresh-hold will not be considered.
DECOMPOSITION:-
de_novo_fit_penalty: Float, optional. Takes any positive float. Default is 0.02. Defines the weak (remove) thresh-hold cutoff to be assigned denovo signatures to a sample.
nnls_add_penalty: Float, optional. Takes any positive float. Default is 0.05. Defines the strong (add) thresh-hold cutoff to be assigned COSMIC signatures to a sample.
nnls_remove_penalty: Float, optional. Takes any positive float. Default is 0.01. Defines the weak (remove) thresh-hold cutoff to be assigned COSMIC signatures to a sample.
initial_remove_penalty: Float, optional. Takes any positive float. Default is 0.05. Defines the initial weak (remove) thresh-hold cutoff to be COSMIC assigned signatures to a sample.
refit_denovo_signatures: Boolean, optional. Default is False. If True, then refit the denovo signatures with nnls.
make_decomposition_plots: Boolean, optional. Defualt is True. If True, Denovo to Cosmic sigantures decompostion plots will be created as a part the results.
OTHERS:-
get_all_signature_matrices: A Boolean. If true, the Ws and Hs from all the NMF iterations are generated in the output.
export_probabilities: A Boolean. Defualt is True. If False, then doesn't create the probability matrix.
Returns
-------
To learn about the output, please visit https://osf.io/t6j7u/wiki/home/
Examples
--------
Examples
--------
>>> from SigProfilerExtractor import sigpro as sig
# to get input from vcf files
>>> path_to_example_folder_containing_vcf_files = sig.importdata("vcf")
>>> data = path_to_example_folder_containing_vcf_files # you can put the path to your folder containing the vcf samples
>>> sig.sigProfilerExtractor("vcf", "example_output", data, minimum_signatures=1, maximum_signatures=3)
Wait untill the excecution is finished. The process may a couple of hours based on the size of the data.
Check the current working directory for the "example_output" folder.
# to get input from table format (mutation catalog matrix)
>>> path_to_example_table = sig.importdata("matrix")
>>> data = path_to_example_table # you can put the path to your tab delimited file containing the mutational catalog matrix/table
>>> sig.sigProfilerExtractor("matrix", "example_output", data, opportunity_genome="GRCh38", minimum_signatures=1, maximum_signatures=3)
Wait untill the excecution is finished. The process may a couple of hours based on the size of the data.
Check the results in the "example_output" folder.
"""
#record the start time
start_time = datetime.datetime.now()
#set the output variable
out_put = output
if gpu == True:
import torch
if gpu and (torch.cuda.device_count() == 0):
raise RuntimeError("GPU not available!")
#################################### At first create the system data file ####################################
if not os.path.exists(out_put):
os.makedirs(out_put)
sysdata = open(out_put + "/JOB_METADATA.txt", "w")
sysdata.write(
"THIS FILE CONTAINS THE METADATA ABOUT SYSTEM AND RUNTIME\n\n\n")
sysdata.write("-------System Info-------\n")
sysdata.write("Operating System Name: " + platform.uname()[0] + "\n" +
"Nodename: " + platform.uname()[1] + "\n" + "Release: " +
platform.uname()[2] + "\n" + "Version: " +
platform.uname()[3] + "\n")
sysdata.write("\n-------Python and Package Versions------- \n")
sysdata.write("Python Version: " + str(platform.sys.version_info.major) +
"." + str(platform.sys.version_info.minor) + "." +
str(platform.sys.version_info.micro) + "\n")
sysdata.write("Sigproextractor Version: " + cosmic.__version__ + "\n")
sysdata.write("SigprofilerPlotting Version: " +
sigProfilerPlotting.__version__ + "\n")
sysdata.write("SigprofilerMatrixGenerator Version: " +
SigProfilerMatrixGenerator.__version__ + "\n")
sysdata.write("Pandas version: " + pd.__version__ + "\n")
sysdata.write("Numpy version: " + np.__version__ + "\n")
sysdata.write("Scipy version: " + scipy.__version__ + "\n")
sysdata.write("Scikit-learn version: " + sklearn.__version__ + "\n")
#sysdata.write("Nimfa version: "+nimfa.__version__+"\n")
#format the project_name first:
project = input_data #will use this variable as the parameter for project argument in SigprofilerMatrixGenerator
try:
if project[-1] != "/":
project_name = project.split(
"/"
)[-1] #will use this variable as the parameter for project_name argument in SigprofilerMatrixGenerator
else:
project_name = project.split("/")[-2]
except:
project_name = "Input from DataFrame"
excecution_parameters = {
"input_type": input_type,
"output": output,
"input_data": input_data,
"reference_genome": reference_genome,
"opportunity_genome": opportunity_genome,
"context_type": context_type,
"exome": exome,
"minimum_signatures": minimum_signatures,
"maximum_signatures": maximum_signatures,
"NMF_replicates": nmf_replicates,
"cpu": cpu,
"gpu": gpu,
"batch_size": batch_size,
"NMF_init": nmf_init,
"precision": precision,
"matrix_normalization": matrix_normalization,
"resample": resample,
"seeds": seeds,
"min_NMF_iterations": min_nmf_iterations,
"max_NMF_iterations": max_nmf_iterations,
"NMF_test_conv": nmf_test_conv,
"NMF_tolerance": nmf_tolerance,
"nnls_add_penalty": nnls_add_penalty,
"nnls_remove_penalty": nnls_remove_penalty,
"initial_remove_penalty": initial_remove_penalty,
"de_novo_fit_penalty": de_novo_fit_penalty,
"refit_denovo_signatures": refit_denovo_signatures,
"dist": clustering_distance,
"export_probabilities": export_probabilities,
"make_decompostion_plots": make_decomposition_plots,
"stability": stability,
"min_stability": min_stability,
"combined_stability": combined_stability,
"get_all_signature_matrices": get_all_signature_matrices
}
################################ take the inputs from the mandatory arguments ####################################
input_type = input_type
#project = input_data #the variable was already set above
################################ take the inputs from the general optional arguments ####################################
startProcess = minimum_signatures
endProcess = maximum_signatures
#totalIterations=nmf_replicates
cpu = cpu
hierarchy = False #No use
mtype = context_type
#init=nmf_init
wall = get_all_signature_matrices
add_penalty = nnls_add_penalty
remove_penalty = nnls_remove_penalty
genome_build = opportunity_genome
refgen = reference_genome
refit_denovo_signatures
#set the squence type ("genome" or "exome") for the tmb plot inside the make_final_solution function
if exome == False:
sequence = "genome"
if exome == True:
sequence = "exome"
#setting seeds
if seeds == "random":
excecution_parameters["seeds"] = seeds
replicates = list(range(1, nmf_replicates + 1))
seed = np.random.randint(0, 10000000, size=nmf_replicates)
seeds = pd.DataFrame(list(zip(replicates, seed)),
columns=["Replicates", "Seeds"])
seeds = seeds.set_index("Replicates")
seeds.to_csv(out_put + "/Seeds.txt", sep="\t")
else:
try:
excecution_parameters["seeds"] = seeds
seeds = pd.read_csv(seeds, sep="\t", index_col=0)
seeds.to_csv(out_put + "/Seeds.txt", sep="\t")
seed = np.array(seeds["Seeds"])
except:
"Please set valid seeds"
if input_type == "text" or input_type == "table" or input_type == "matrix":
################################### For text input files ######################################################
text_file = project
title = "" # set the title for plotting
if type(text_file) != str:
data = text_file
excecution_parameters["input_data"] = "Matrix[" + str(
data.shape[0]) + " rows X " + str(data.shape[1]) + " columns]"
else:
data = pd.read_csv(text_file, sep="\t").iloc[:, :]
data = data.dropna(axis=1, inplace=False)
data = data.loc[:, (data != 0).any(axis=0)]
genomes = data.iloc[:, 1:]
genomes = np.array(genomes)
allgenomes = genomes.copy(
) # save the allgenomes for the final results
#Contruct the indeces of the matrix
#setting index and columns names of processAvg and exposureAvg
index = data.iloc[:, 0]
colnames = data.columns[1:]
allcolnames = colnames.copy(
) # save the allcolnames for the final results
#creating list of mutational type to sync with the vcf type input
mtypes = [str(genomes.shape[0])]
if mtypes[0] == "78":
mtypes = ["DBS78"]
elif mtypes[0] == "83":
mtypes = ["ID83"]
else:
mtypes = ["SBS" + mtypes[0]]
###############################################################################################################
###########################################################################################################################################################################################
elif input_type == "csv":
################################# For matlab input files #######################################################
filename = project
title = "" # set the title for plotting
genomes, index, colnames, mtypes = sub.read_csv(filename)
allgenomes = genomes.copy()
allcolnames = colnames.copy()
# Define the mtypes
mtypes = [str(genomes.shape[0])]
if mtypes[0] == "78":
mtypes = ["DINUC"]
elif mtypes[0] == "83":
mtypes = ["ID"]
#################################################################################################################
###########################################################################################################################################################################################
elif input_type == "matobj":
################################# For matlab input files #######################################################
mat_file = project
title = "" # set the title for plotting
mat = scipy.io.loadmat(mat_file)
mat = sub.extract_input(mat)
genomes = mat[1]
allgenomes = genomes.copy(
) # save the allgenomes for the final results
#Contruct the indeces of the matrix
#setting index and columns names of processAvg and exposureAvg
index1 = mat[3]
index2 = mat[4]
index = []
for i, j in zip(index1, index2):
index.append(i[0] + "[" + j + "]" + i[2])
colnames = np.array(pd.Series(mat[2]))
allcolnames = colnames.copy(
) # save the allcolnames for the final results
index = np.array(pd.Series(index))
#creating list of mutational type to sync with the vcf type input
mtypes = [str(genomes.shape[0])]
if mtypes[0] == "78":
mtypes = ["DINUC"]
elif mtypes[0] == "83":
mtypes = ["ID"]
#################################################################################################################
elif input_type == "vcf":
################################# For vcf input files #######################################################
project = project
title = project # set the title for plotting
refgen = refgen
exome = exome
#project_name = project.split("/")[-1]
data = datadump.SigProfilerMatrixGeneratorFunc(project_name,
refgen,
project,
exome=exome,
bed_file=None,
chrom_based=False,
plot=False,
gs=False)
# Selecting the mutation types
if mtype == ["default"]:
if set(["96", "DINUC", "ID"]).issubset(data):
mtypes = ["SBS96", "DBS78", "ID83"]
elif set(["96", "DINUC"]).issubset(data):
mtypes = ["SBS96", "DBS78"]
elif set(["ID"]).issubset(data):
mtypes = ["ID83"]
elif mtype == "default":
if set(["96", "DINUC", "ID"]).issubset(data):
mtypes = ["SBS96", "DBS78", "ID83"]
elif set(["96", "DINUC"]).issubset(data):
mtypes = ["SBS96", "DBS78"]
elif set(["ID"]).issubset(data):
mtypes = ["ID83"]
else:
#mkeys = data.keys()
mtype = mtype.upper()
mtype = mtype.replace(" ", "")
mtypes = mtype.split(",")
# =============================================================================
# if any(x not in mkeys for x in mtypes):
# raise Exception("Please pass valid mutation types seperated by comma with no space. Carefully check (using SigProfilerMatrixGenerator)"\
# "what mutation contexts should be generated by your VCF files. Also please use the uppercase characters")
# =============================================================================
#change working directory
#set the genome_build
genome_build = refgen
else:
raise ValueError(
"Please provide a correct input_type. Check help for more details")
#recording context types
excecution_parameters["context_type"] = ",".join(mtypes)
record_parameters(sysdata, excecution_parameters, start_time)
sysdata.close()
###########################################################################################################################################################################################
for m in mtypes:
mutation_context = m
# we need to rename the m because users input could be SBS96, SBS1536, DBS78, ID83 etc
if m.startswith("SBS"):
m = m[3:] #removing "SBS"
elif m.startswith("DBS"):
m = "DINUC"
elif m.startswith("ID"):
m = "ID"
# Determine the types of mutation which will be needed for exporting and copying the files
if not (m == "DINUC" or m.startswith("DBS") or m.startswith("ID")):
if m.startswith("SBS"):
mutation_type = m
else:
mutation_type = "SBS" + m
else:
if m == "DINUC" or m.startswith("DBS"):
mutation_type = "DBS78"
elif m == "ID" or m.stratswith("ID"):
mutation_type = "ID83"
if input_type == "vcf":
try:
genomes = pd.DataFrame(data[m])
except:
raise Exception("Please pass valid mutation types seperated by comma with no space. Carefully check (using SigProfilerMatrixGenerator)"\
"what mutation contexts should be generated by your VCF files. Also please use the uppercase characters")
#check if the genome is a nonzero matrix
shape = genomes.shape
if shape == (0, 0):
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
sysdata.write(
"Sample is not a nonzero matrix for the mutation context "
+ m + "\n")
print(
"Sample is not a nozero matrix for the mutation context " +
m)
sysdata.close()
continue
genomes = genomes.loc[:, (genomes != 0).any(axis=0)]
allgenomes = genomes.copy(
) # save the allgenomes for the final results
index = genomes.index.values
colnames = genomes.columns
allcolnames = colnames.copy(
) # save the allcolnames for the final results
#check if start and end processes are bigger than the number of samples
startProcess = min(startProcess, genomes.shape[1])
endProcess = min(endProcess, genomes.shape[1])
#in the plotting funciton "ID" is used as "INDEL"
if m == "ID":
m = "INDEL" #for plotting
#create output directories to store all the results
output = out_put + "/" + mutation_type
est_genomes = np.zeros([1, 1])
H_iteration = 1
genomes = np.array(genomes)
information = []
layer_directory = output
try:
if not os.path.exists(layer_directory):
os.makedirs(layer_directory)
#os.makedirs(output+"/pickle_objects")
#os.makedirs(output+"/All solutions")
except:
print("The {} folder could not be created".format("output"))
fh = open(layer_directory + "/All_solutions_stat.csv", "w")
fh.write("Total Signatures,Stability,Matrix Frobenius%,avgStability\n")
fh.close()
# The following for loop operates to extract data from each number of signature
all_similirities_list = [
] #this list is going to store the dataframes of different similirieties as items
minimum_stabilities = []
#similarity_dataframe = pd.DataFrame({"Sample Name": list(colnames)})
# get the cutoff for normatization to handle the hypermutators
normalization_cutoff = sub.get_normalization_cutoff(genomes,
manual_cutoff=100 *
genomes.shape[0])
#print("Normalization Cutoff is :", normalization_cutoff)
excecution_parameters["normalization_cutoff"] = normalization_cutoff
#pass the seed values to inner funtions:
excecution_parameters["seeds"] = seed
if genomes.shape[1] < endProcess:
endProcess = genomes.shape[1]
#report the notmatlization criteria
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
context_start_time = datetime.datetime.now()
sysdata.write("\n##################################\n")
sysdata.write(
"\n[{}] Analysis started for {}. Matrix size [{} rows x {} columns]\n"
.format(
str(context_start_time).split(".")[0], mutation_type,
genomes.shape[0], genomes.shape[1]))
if excecution_parameters["matrix_normalization"] == "gmm":
sysdata.write("\n[{}] Normalization GMM with cutoff value set at {}\n". \
format(str(datetime.datetime.now()).split(".")[0], normalization_cutoff))
elif excecution_parameters["matrix_normalization"] == "100X":
sysdata.write("\n[{}] Normalization 100X with cutoff value set at {}\n". \
format(str(datetime.datetime.now()).split(".")[0],(genomes.shape[0]*100)))
elif excecution_parameters["matrix_normalization"] == "log2":
sysdata.write("\n[{}] Normalization Log2\n". \
format(str(datetime.datetime.now()).split(".")[0]))
elif excecution_parameters["matrix_normalization"] == "none":
sysdata.write("\n[{}] Analysis is proceeding without normalization\n". \
format(str(datetime.datetime.now()).split(".")[0]))
else:
sysdata.write("\n[{}] Normalization Custom with cutoff value set at {}\n". \
format(str(datetime.datetime.now()).split(".")[0],excecution_parameters["matrix_normalization"]))
sysdata.close()
for i in range(startProcess, endProcess + 1):
current_time_start = datetime.datetime.now()
#memory_usage()
processAvg, \
exposureAvg, \
processStd, \
exposureStd, \
avgSilhouetteCoefficients, \
clusterSilhouetteCoefficients, \
finalgenomeErrors, \
finalgenomesReconstructed, \
finalWall, \
finalHall, \
converge_information, \
reconstruction_error, \
processes = sub.decipher_signatures(excecution_parameters,
genomes= genomes,
mut_context=m,
i = i)
#denormalize the genomes and exposures
#genomes = sub.denormalize_samples(genomes, totalMutations, normalization_value=100000)
#exposureStd = sub.denormalize_samples(exposureStd, totalMutations, normalization_value=100000)
####################################################################### add sparsity in the exposureAvg #################################################################
# remove signatures only if the process stability is above a thresh-hold of 0.85
if avgSilhouetteCoefficients > -1.0:
stic = time.time()
#removing signatures:
# =============================================================================
# pool = mp.Pool()
# results = [pool.apply_async(sub.remove_all_single_signatures_pool, args=(x,processAvg,exposureAvg,genomes,)) for x in range(genomes.shape[1])]
# pooloutput = [p.get() for p in results]
#
# #print(results)
# pool.close()
#
# for i in range(len(pooloutput)):
# #print(results[i])
# exposureAvg[:,i]=pooloutput[i]
# =============================================================================
#refitting signatures:
#removing signatures:
pool = mp.Pool()
results = [
pool.apply_async(ss.fit_signatures_pool,
args=(
genomes,
processAvg,
x,
)) for x in range(genomes.shape[1])
]
pooloutput = [p.get() for p in results]
pool.close()
for i in range(len(pooloutput)):
exposureAvg[:, i] = pooloutput[i][0]
stoc = time.time()
print("Optimization time is {} seconds".format(stoc - stic))
#sysdata.write("\nAnalysis of context type {} is ended successfully\n".format(m))
#report progress to the system file:
#Get total mutationation for each signature in reverse order and order the signatures from high to low mutation barden
signature_total_mutations = np.sum(exposureAvg, axis=1).astype(int)
sorted_idx = np.argsort(-signature_total_mutations)
processAvg = np.take(processAvg, sorted_idx, axis=1)
exposureAvg = np.take(exposureAvg, sorted_idx, axis=0)
signature_total_mutations = np.sum(exposureAvg, axis=1).astype(int)
processStd = np.take(processStd, sorted_idx, axis=1)
exposureStd = np.take(exposureStd, sorted_idx, axis=0)
clusterSilhouetteCoefficients = np.take(
clusterSilhouetteCoefficients, sorted_idx, axis=0)
signature_stats = pd.DataFrame({
"Stability":
clusterSilhouetteCoefficients,
"Total Mutations":
signature_total_mutations
})
minimum_stabilities.append(
round(np.mean(clusterSilhouetteCoefficients), 2)
) #here minimum stability is the average stability !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# Compute the estimated genome from the processAvg and exposureAvg
est_genomes = np.dot(processAvg, exposureAvg)
#check the similarities between the original and estimated genome for each number of signatures
all_similarities, cosine_similarities = sub.calculate_similarities(
genomes, est_genomes, colnames)
#print(totalMutations)
##########################################################################################################################################################################
# store the resutls of the loop. Here, processStd and exposureStd are standard Errors, NOT STANDARD DEVIATIONS.
loopResults = [
genomes, processAvg, exposureAvg, processStd, exposureStd,
avgSilhouetteCoefficients, clusterSilhouetteCoefficients,
signature_total_mutations, all_similarities, signature_stats,
reconstruction_error, finalgenomeErrors,
finalgenomesReconstructed, converge_information, finalWall,
finalHall, processes
]
information.append([
processAvg, exposureAvg, processStd, exposureStd,
clusterSilhouetteCoefficients, signature_total_mutations,
signature_stats, all_similarities
]) #Will be used during hierarchycal approach
################################# Export the results ###########################################################
sub.export_information(loopResults,
m,
layer_directory,
index,
colnames,
wall=wall,
sequence=sequence)
all_similirities_list.append(all_similarities)
#
#similarity_dataframe["Total Signatures "+str(processes)] = cosine_similarities
current_time_end = datetime.datetime.now()
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
sysdata.write("\n[{}] {} de novo extraction completed for a total of {} signatures! \nExecution time:{}\n". \
format(str(datetime.datetime.now()).split(".")[0],mutation_type,processes,str(current_time_end-current_time_start).split(".")[0], current_time_end))
sysdata.close()
################################################################################################################
########################################## Plot Stabiltity vs Reconstruction Error #############################
################################################################################################################
# Print the Stabiltity vs Reconstruction Error as get the solution as well
solution, all_stats = sub.stabVsRError(
layer_directory + "/All_solutions_stat.csv",
layer_directory,
title,
all_similirities_list,
mtype=mutation_type,
stability=stability,
min_stability=min_stability,
combined_stability=combined_stability)
all_stats.insert(
1, 'Stability (Avg Silhouette)', minimum_stabilities
) #!!!!!!!!!!!!!!!!1 here minimum stability is avg stability
all_stats = all_stats.set_index(["Signatures"])
all_stats.to_csv(layer_directory + "/All_solutions_stat.csv", sep=",")
# add more information to results_stat.csv
#Set index for the the Similarity Dataframe
#similarity_dataframe = similarity_dataframe.set_index("Sample Name")
#Add the total mutations of each sample
#sample_total_mutations = list(np.sum(genomes, axis =0))
#similarity_dataframe.insert(loc=0, column = "Total Mutations", value = sample_total_mutations)
# write the name of Samples and Matrix participating in each Layer.
layer_genome = pd.DataFrame(genomes)
layer_genome = layer_genome.set_index(index)
layer_genome.columns = colnames
layer_genome = layer_genome.rename_axis("Mutation Types",
axis="columns")
# =============================================================================
# data_stat_folder = output+"/Data_Stats"
# try:
# if not os.path.exists(data_stat_folder):
# os.makedirs(data_stat_folder)
# except:
# print ("The {} folder could not be created".format("Data_Stats"))
#
# layer_genome.to_csv(data_stat_folder+"/Samples.text", sep = "\t", index_label=[layer_genome.columns.name])
# similarity_dataframe.to_csv(data_stat_folder+"/Similatiry_Data_All_Sigs.text", sep = "\t")
# del layer_genome
# for i in range(startProcess,endProcess+1):
# all_similirities_list[i-startProcess].to_csv(data_stat_folder+"/Similatiry_Data_Sig_"+str(i)+".text", sep="\t")
# =============================================================================
# record the samples
layer_genome.to_csv(output + "/Samples.txt",
sep="\t",
index_label=[layer_genome.columns.name])
#similarity_dataframe.to_csv(data_stat_folder+"/Similatiry_Data_All_Sigs"+str(H_iteration)+".text", sep = "\t")
del layer_genome
################################### Decompose the new signatures into global signatures #########################
processAvg = information[solution - startProcess][0]
exposureAvg = information[solution - startProcess][1]
processSTE = information[solution - startProcess][2]
signature_stabilities = information[solution - startProcess][4]
signature_total_mutations = information[solution - startProcess][5]
signature_stats = information[solution - startProcess][6]
all_similarities = information[solution - startProcess][7]
# create the folder for the final solution/ De Novo Solution
layer_directory1 = output + "/Suggested_Solution/" + mutation_type + "_De_Novo_Solution"
try:
if not os.path.exists(layer_directory1):
os.makedirs(layer_directory1)
except:
print("The {} folder could not be created".format("output"))
# make the texts for signature plotting
signature_stabilities = sub.signature_plotting_text(
signature_stabilities, "Stability", "float")
signature_total_mutations = sub.signature_plotting_text(
signature_total_mutations, "Total Mutations", "integer")
# make de novo solution(processAvg, allgenomes, layer_directory1)
listOfSignatures = sub.make_letter_ids(idlenth=processAvg.shape[1],
mtype=mutation_context)
allgenomes = pd.DataFrame(allgenomes)
exposureAvg = sub.make_final_solution(processAvg, allgenomes, listOfSignatures, layer_directory1, m, index, \
allcolnames, process_std_error = processSTE, signature_stabilities = signature_stabilities, \
signature_total_mutations = signature_total_mutations,denovo_exposureAvg = exposureAvg, \
signature_stats = signature_stats, add_penalty=add_penalty, remove_penalty=remove_penalty, \
initial_remove_penalty=initial_remove_penalty, refit_denovo_signatures=refit_denovo_signatures, de_novo_fit_penalty=de_novo_fit_penalty, sequence=sequence)
#try:
# create the folder for the final solution/ Decomposed Solution
layer_directory2 = output + "/Suggested_Solution/COSMIC_" + mutation_type + "_Decomposed_Solution"
try:
if not os.path.exists(layer_directory2):
os.makedirs(layer_directory2)
except:
print("The {} folder could not be created".format("output"))
originalProcessAvg = pd.DataFrame(processAvg, index=index)
if processAvg.shape[
0] == 1536: #collapse the 1596 context into 96 only for the deocmposition
processAvg = pd.DataFrame(processAvg, index=index)
processAvg = processAvg.groupby(processAvg.index.str[1:8]).sum()
genomes = pd.DataFrame(genomes, index=index)
genomes = genomes.groupby(genomes.index.str[1:8]).sum()
index = genomes.index
processAvg = np.array(processAvg)
genomes = np.array(genomes)
if processAvg.shape[
0] == 288: #collapse the 288 context into 96 only for the deocmposition
processAvg = pd.DataFrame(processAvg, index=index)
processAvg = processAvg.groupby(processAvg.index.str[2:9]).sum()
genomes = pd.DataFrame(genomes, index=index)
genomes = genomes.groupby(genomes.index.str[2:9]).sum()
index = genomes.index
processAvg = np.array(processAvg)
genomes = np.array(genomes)
originalProcessAvg.columns = listOfSignatures
final_signatures = sub.signature_decomposition(
processAvg,
m,
layer_directory2,
genome_build=genome_build,
add_penalty=add_penalty,
remove_penalty=remove_penalty,
mutation_context=mutation_context,
make_decomposition_plots=make_decomposition_plots,
originalProcessAvg=originalProcessAvg)
# extract the global signatures and new signatures from the final_signatures dictionary
globalsigs = final_signatures["globalsigs"]
globalsigs = np.array(globalsigs)
newsigs = final_signatures["newsigs"]
try:
processAvg = np.hstack([globalsigs, newsigs])
allsigids = final_signatures["globalsigids"] + final_signatures[
"newsigids"]
except:
processAvg = newsigs
allsigids = final_signatures["newsigids"]
attribution = final_signatures["dictionary"]
background_sigs = final_signatures["background_sigs"]
genomes = pd.DataFrame(genomes)
exposureAvg = sub.make_final_solution(processAvg, genomes, allsigids, layer_directory2, m, index, colnames, \
cosmic_sigs=True, attribution = attribution, denovo_exposureAvg = exposureAvg , background_sigs=background_sigs, add_penalty=add_penalty, remove_penalty=remove_penalty, initial_remove_penalty=initial_remove_penalty, genome_build=genome_build, sequence=sequence,export_probabilities=export_probabilities)
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
end_time = datetime.datetime.now()
sysdata.write("\n[{}] Analysis ended: \n".format(
str(end_time).split(".")[0]))
sysdata.write("\n-------Job Status------- \n")
sysdata.write(
"Analysis of mutational signatures completed successfully! \nTotal execution time: "
+ str(end_time - start_time).split(".")[0] +
" \nResults can be found in: " + " " + out_put + " " + " folder")
sysdata.close()
print(
"\n\n \nYour Job Is Successfully Completed! Thank You For Using SigProfiler Extractor.\n "
)
def sigProfilerExtractor(input_type,
out_put,
input_data,
refgen="GRCh37",
genome_build='GRCh37',
startProcess=1,
endProcess=10,
totalIterations=100,
init="alexandrov-lab-custom",
cpu=-1,
mtype="default",
exome=False,
penalty=0.05,
resample=True,
wall=False,
gpu=False):
memory_usage()
"""
Extracts mutational signatures from an array of samples.
Parameters
----------
input_type: A string. Type of input. The type of input should be one of the following:
- "vcf": used for vcf format inputs.
- "matrix": used for table format inputs using a tab seperated file.
out_put: A string. The name of the output folder. The output folder will be generated in the current working directory.
input_data: A string. Name of the input folder (in case of "vcf" type input) or the input file (in case of "table" type input). The project file or folder should be inside the current working directory. For the "vcf" type input,the project has to be a folder which will contain the vcf files in vcf format or text formats. The "text"type projects have to be a file.
refgen: A string, optional. The name of the reference genome. The default reference genome is "GRCh37". This parameter is applicable only if the input_type is "vcf".
startProcess: A positive integer, optional. The minimum number of signatures to be extracted. The default value is 1
endProcess: A positive integer, optional. The maximum number of signatures to be extracted. The default value is 10
totalIterations: A positive integer, optional. The number of iteration to be performed to extract each number signature. The default value is 100
init: A String. The initialization algorithm for W and H matrix of NMF
wall: A Boolean. If true, the Ws and Hs from all the NMF iterations are generated in the output.
cpu: An integer, optional. The number of processors to be used to extract the signatures. The default value is -1 which will use all available processors.
mtype: A list of strings, optional. The items in the list defines the mutational contexts to be considered to extract the signatures. The default value is ["96", "DINUC" , "ID"], where "96" is the SBS96 context, "DINUC"
is the DINULEOTIDE context and ID is INDEL context.
exome: Boolean, optional. Defines if the exomes will be extracted. The default value is "False".
penalty: Float, optional. Takes any positive float. Default is 0.05. Defines the thresh-hold cutoff to asaign signatures to a sample.
resample: Boolean, optional. Default is True. If True, add poisson noise to samples by resampling.
Returns
-------
To learn about the output, please visit https://osf.io/t6j7u/wiki/home/
Examples
--------
>>> from SigProfilerExtractor import sigpro as sig
>>> data = sig.importdata("vcf")
>>> sig.sigProfilerExtractor("vcf", "example_output", data, startProcess=1, endProcess=3)
Wait untill the excecution is finished. The process may a couple of hours based on the size of the data.
Check the results in the "example_output" folder.
"""
if gpu == True:
import torch
if gpu and (torch.cuda.device_count() == 0):
raise RuntimeError("GPU not available!")
#################################### At first create the system data file ####################################
if not os.path.exists(out_put):
os.makedirs(out_put)
sysdata = open(out_put + "/JOB_METADATA.txt", "w")
sysdata.write(
"THIS FILE CONTAINS THE METADATA ABOUT SYSTEM AND RUNTIME\n\n\n")
sysdata.write("-------System Info-------\n")
sysdata.write("Operating System Name: " + platform.uname()[0] + "\n" +
"Nodename: " + platform.uname()[1] + "\n" + "Release: " +
platform.uname()[2] + "\n" + "Version: " +
platform.uname()[3] + "\n")
sysdata.write("\n-------Python and Package Versions------- \n")
sysdata.write("Python Version: " + str(platform.sys.version_info.major) +
"." + str(platform.sys.version_info.minor) + "." +
str(platform.sys.version_info.micro) + "\n")
sysdata.write("Sigproextractor Version: " + cosmic.__version__ + "\n")
sysdata.write("SigprofilerPlotting Version: " +
sigProfilerPlotting.__version__ + "\n")
sysdata.write("SigprofilerMatrixGenerator Version: " +
SigProfilerMatrixGenerator.__version__ + "\n")
sysdata.write("Pandas version: " + pd.__version__ + "\n")
sysdata.write("Numpy version: " + np.__version__ + "\n")
sysdata.write("Scipy version: " + scipy.__version__ + "\n")
sysdata.write("Scikit-learn version: " + sklearn.__version__ + "\n")
#sysdata.write("Nimfa version: "+nimfa.__version__+"\n")
sysdata.write("\n-------Vital Parameters Used for the execution -------\n")
#format the project_name first:
project = input_data #will use this variable as the parameter for project argument in SigprofilerMatrixGenerator
try:
if project[-1] != "/":
project_name = project.split(
"/"
)[-1] #will use this variable as the parameter for project_name argument in SigprofilerMatrixGenerator
else:
project_name = project.split("/")[-2]
except:
project_name = "Input from DataFrame"
sysdata.write(
"input_type: {}\ninputdata: {}\nstartProcess: {}\nendProcess: {}\ntotalIterations: {}\ncpu: {}\nrefgen: {}\ngenome_build: {}\nmtype: {} \ninit: {}\n"
.format(input_type, project_name, startProcess, endProcess,
totalIterations, cpu, refgen, genome_build, mtype, init))
sysdata.write("\n-------Date and Time Data------- \n")
tic = datetime.datetime.now()
sysdata.write("Date and Clock time when the execution started: " +
str(tic) + "\n")
sysdata.close()
################################ take the inputs from the mandatory arguments ####################################
input_type = input_type
out_put = out_put
#project = input_data #the variable was already set above
################################ take the inputs from the general optional arguments ####################################
startProcess = startProcess
endProcess = endProcess
totalIterations = totalIterations
cpu = cpu
hierarchi = False #No use
if input_type == "text" or input_type == "table" or input_type == "matrix":
################################### For text input files ######################################################
text_file = project
title = "" # set the title for plotting
if type(text_file) != str:
data = text_file
else:
data = pd.read_csv(text_file, sep="\t").iloc[:, :]
data = data.dropna(axis=1, inplace=False)
data = data.loc[:, (data != 0).any(axis=0)]
genomes = data.iloc[:, 1:]
genomes = np.array(genomes)
allgenomes = genomes.copy(
) # save the allgenomes for the final results
#Contruct the indeces of the matrix
#setting index and columns names of processAvg and exposureAvg
index = data.iloc[:, 0]
colnames = data.columns[1:]
allcolnames = colnames.copy(
) # save the allcolnames for the final results
#creating list of mutational type to sync with the vcf type input
mtypes = [str(genomes.shape[0])]
if mtypes[0] == "78":
mtypes = ["DBS78"]
elif mtypes[0] == "83":
mtypes = ["ID83"]
else:
mtypes = ["SBS" + mtypes[0]]
###############################################################################################################
###########################################################################################################################################################################################
elif input_type == "csv":
################################# For matlab input files #######################################################
filename = project
title = "" # set the title for plotting
genomes, index, colnames, mtypes = sub.read_csv(filename)
allgenomes = genomes.copy()
allcolnames = colnames.copy()
# Define the mtypes
mtypes = [str(genomes.shape[0])]
if mtypes[0] == "78":
mtypes = ["DINUC"]
elif mtypes[0] == "83":
mtypes = ["ID"]
#################################################################################################################
###########################################################################################################################################################################################
elif input_type == "matobj":
################################# For matlab input files #######################################################
mat_file = project
title = "" # set the title for plotting
mat = scipy.io.loadmat(mat_file)
mat = sub.extract_input(mat)
genomes = mat[1]
allgenomes = genomes.copy(
) # save the allgenomes for the final results
#Contruct the indeces of the matrix
#setting index and columns names of processAvg and exposureAvg
index1 = mat[3]
index2 = mat[4]
index = []
for i, j in zip(index1, index2):
index.append(i[0] + "[" + j + "]" + i[2])
colnames = np.array(pd.Series(mat[2]))
allcolnames = colnames.copy(
) # save the allcolnames for the final results
index = np.array(pd.Series(index))
#creating list of mutational type to sync with the vcf type input
mtypes = [str(genomes.shape[0])]
if mtypes[0] == "78":
mtypes = ["DINUC"]
elif mtypes[0] == "83":
mtypes = ["ID"]
#################################################################################################################
elif input_type == "vcf":
################################# For vcf input files #######################################################
project = project
title = project # set the title for plotting
refgen = refgen
exome = exome
#project_name = project.split("/")[-1]
data = datadump.SigProfilerMatrixGeneratorFunc(project_name,
refgen,
project,
exome=exome,
bed_file=None,
chrom_based=False,
plot=False,
gs=False)
# Selecting the mutation types
if mtype == ["default"]:
if set(["96", "DINUC", "ID"]).issubset(data):
mtypes = ["SBS96", "DBS78", "ID83"]
elif set(["96", "DINUC"]).issubset(data):
mtypes = ["SBS96", "DBS78"]
elif set(["ID"]).issubset(data):
mtypes = ["ID83"]
elif mtype == "default":
if set(["96", "DINUC", "ID"]).issubset(data):
mtypes = ["SBS96", "DBS78", "ID83"]
elif set(["96", "DINUC"]).issubset(data):
mtypes = ["SBS96", "DBS78"]
elif set(["ID"]).issubset(data):
mtypes = ["ID83"]
else:
#mkeys = data.keys()
mtype = mtype.upper()
mtype = mtype.replace(" ", "")
mtypes = mtype.split(",")
# =============================================================================
# if any(x not in mkeys for x in mtypes):
# raise Exception("Please pass valid mutation types seperated by comma with no space. Carefully check (using SigProfilerMatrixGenerator)"\
# "what mutation contexts should be generated by your VCF files. Also please use the uppercase characters")
# =============================================================================
#change working directory
#set the genome_build
genome_build = refgen
else:
raise ValueError(
"Please provide a correct input_type. Check help for more details")
###########################################################################################################################################################################################
for m in mtypes:
mutation_context = m
# we need to rename the m because users input could be SBS96, SBS1536, DBS78, ID83 etc
if m.startswith("SBS"):
m = m[3:] #removing "SBS"
elif m.startswith("DBS"):
m = "DINUC"
elif m.startswith("ID"):
m = "ID"
# Determine the types of mutation which will be needed for exporting and copying the files
if not (m == "DINUC" or m.startswith("DBS") or m.startswith("ID")):
if m.startswith("SBS"):
mutation_type = m
else:
mutation_type = "SBS" + m
else:
if m == "DINUC" or m.startswith("DBS"):
mutation_type = "DBS78"
elif m == "ID" or m.stratswith("ID"):
mutation_type = "ID83"
if input_type == "vcf":
try:
genomes = pd.DataFrame(data[m])
except:
raise Exception("Please pass valid mutation types seperated by comma with no space. Carefully check (using SigProfilerMatrixGenerator)"\
"what mutation contexts should be generated by your VCF files. Also please use the uppercase characters")
#check if the genome is a nonzero matrix
shape = genomes.shape
if shape == (0, 0):
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
sysdata.write(
"Sample is not a nonzero matrix for the mutation context "
+ m + "\n")
print(
"Sample is not a nozero matrix for the mutation context " +
m)
sysdata.close()
continue
genomes = genomes.loc[:, (genomes != 0).any(axis=0)]
allgenomes = genomes.copy(
) # save the allgenomes for the final results
index = genomes.index.values
colnames = genomes.columns
allcolnames = colnames.copy(
) # save the allcolnames for the final results
#check if start and end processes are bigger than the number of samples
startProcess = min(startProcess, genomes.shape[1])
endProcess = min(endProcess, genomes.shape[1])
#in the plotting funciton "ID" is used as "INDEL"
if m == "ID":
m = "INDEL" #for plotting
#create output directories to store all the results
output = out_put + "/" + mutation_type
est_genomes = np.zeros([1, 1])
H_iteration = 1
genomes = np.array(genomes)
information = []
layer_directory = output
try:
if not os.path.exists(layer_directory):
os.makedirs(layer_directory)
#os.makedirs(output+"/pickle_objects")
#os.makedirs(output+"/All solutions")
except:
print("The {} folder could not be created".format("output"))
fh = open(layer_directory + "/All_solutions_stat.csv", "w")
fh.write("Total Signatures,Stability,Matrix Frobenius%,avgStability\n")
fh.close()
# The following for loop operates to extract data from each number of signature
all_similirities_list = [
] #this list is going to store the dataframes of different similirieties as items
minimum_stabilities = []
#similarity_dataframe = pd.DataFrame({"Sample Name": list(colnames)})
# set up the seeds generation same matrices for different number of signatures
seeds = np.random.randint(
0, 10000000, size=totalIterations
) # set the seeds ranging from 0 to 10000000 for resampling and same seeds are used in different number of signatures
# get the cutoff for normatization to handle the hypermutators
normalization_cutoff = sub.get_normalization_cutoff(genomes)
#print("Normalization Cutoff is :", normalization_cutoff)
#genomes = sub.normalize_samples(genomes, normalize=False, all_samples=False, number=30000)
for i in range(startProcess, endProcess + 1):
current_time_start = datetime.datetime.now()
#memory_usage()
processAvg, \
exposureAvg, \
processStd, \
exposureStd, \
avgSilhouetteCoefficients, \
clusterSilhouetteCoefficients, \
finalgenomeErrors, \
finalgenomesReconstructed, \
finalWall, \
finalHall, \
converge_information, \
reconstruction_error, \
processes = sub.decipher_signatures(genomes= genomes, \
i = i, \
totalIterations=totalIterations, \
cpu=cpu, \
mut_context=m, \
resample = resample,
seeds=seeds,
init = init,
normalization_cutoff=normalization_cutoff,
gpu=gpu,)
#denormalize the genomes and exposures
#genomes = sub.denormalize_samples(genomes, totalMutations, normalization_value=100000)
#exposureStd = sub.denormalize_samples(exposureStd, totalMutations, normalization_value=100000)
####################################################################### add sparsity in the exposureAvg #################################################################
# remove signatures only if the process stability is above a thresh-hold of 0.85
if avgSilhouetteCoefficients > -1.0:
stic = time.time()
#removing signatures:
# =============================================================================
# pool = mp.Pool()
# results = [pool.apply_async(sub.remove_all_single_signatures_pool, args=(x,processAvg,exposureAvg,genomes,)) for x in range(genomes.shape[1])]
# pooloutput = [p.get() for p in results]
#
# #print(results)
# pool.close()
#
# for i in range(len(pooloutput)):
# #print(results[i])
# exposureAvg[:,i]=pooloutput[i]
# =============================================================================
#refitting signatures:
#removing signatures:
pool = mp.Pool()
results = [
pool.apply_async(ss.fit_signatures_pool,
args=(
genomes,
processAvg,
x,
)) for x in range(genomes.shape[1])
]
pooloutput = [p.get() for p in results]
pool.close()
for i in range(len(pooloutput)):
exposureAvg[:, i] = pooloutput[i][0]
stoc = time.time()
print("Optimization time is {} seconds".format(stoc - stic))
#report progress to the system file:
current_time_end = datetime.datetime.now()
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
if hierarchi is True:
sysdata.write(
"\nSignature extraction for {} completed for layer {} {} signatures for {}! TimeStamp: {}\n"
.format(mutation_type, H_iteration, processes,
current_time_end - current_time_start,
current_time_end))
else:
sysdata.write(
"\nSignature extraction for {} completed for {} signatures for {}! TimeStamp: {}\n"
.format(mutation_type, processes,
current_time_end - current_time_start,
current_time_end))
#Get total mutationation for each signature in reverse order and order the signatures from high to low mutation barden
signature_total_mutations = np.sum(exposureAvg, axis=1).astype(int)
sorted_idx = np.argsort(-signature_total_mutations)
processAvg = np.take(processAvg, sorted_idx, axis=1)
exposureAvg = np.take(exposureAvg, sorted_idx, axis=0)
signature_total_mutations = np.sum(exposureAvg, axis=1).astype(int)
signature_stats = pd.DataFrame({
"Stability":
clusterSilhouetteCoefficients,
"Total Mutations":
signature_total_mutations
})
minimum_stabilities.append(
round(np.mean(clusterSilhouetteCoefficients), 2)
) #here minimum stability is the average stability !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
# Compute the estimated genome from the processAvg and exposureAvg
est_genomes = np.dot(processAvg, exposureAvg)
#check the similarities between the original and estimated genome for each number of signatures
all_similarities, cosine_similarities = sub.calculate_similarities(
genomes, est_genomes, colnames)
#print(totalMutations)
##########################################################################################################################################################################
# store the resutls of the loop. Here, processStd and exposureStd are standard Errors, NOT STANDARD DEVIATIONS.
loopResults = [
genomes, processAvg, exposureAvg, processStd, exposureStd,
avgSilhouetteCoefficients, clusterSilhouetteCoefficients,
signature_total_mutations, all_similarities, signature_stats,
reconstruction_error, finalgenomeErrors,
finalgenomesReconstructed, converge_information, finalWall,
finalHall, processes
]
information.append([
processAvg, exposureAvg, processStd, exposureStd,
clusterSilhouetteCoefficients, signature_total_mutations,
signature_stats, all_similarities
]) #Will be used during hierarchical approach
################################# Export the results ###########################################################
sub.export_information(loopResults,
m,
layer_directory,
index,
colnames,
wall=wall)
all_similirities_list.append(all_similarities)
#
#similarity_dataframe["Total Signatures "+str(processes)] = cosine_similarities
################################################################################################################
########################################## Plot Stabiltity vs Reconstruction Error #############################
################################################################################################################
# Print the Stabiltity vs Reconstruction Error as get the solution as well
solution, all_stats = sub.stabVsRError(
layer_directory + "/All_solutions_stat.csv", layer_directory,
title, all_similirities_list, mutation_type)
all_stats.insert(
0, 'Stability (Avg Silhouette)', minimum_stabilities
) #!!!!!!!!!!!!!!!!1 here minimum stability is avg stability
all_stats.to_csv(layer_directory + "/All_solutions_stat.csv", sep=",")
# add more information to results_stat.csv
#Set index for the the Similarity Dataframe
#similarity_dataframe = similarity_dataframe.set_index("Sample Name")
#Add the total mutations of each sample
#sample_total_mutations = list(np.sum(genomes, axis =0))
#similarity_dataframe.insert(loc=0, column = "Total Mutations", value = sample_total_mutations)
# write the name of Samples and Matrix participating in each Layer.
layer_genome = pd.DataFrame(genomes)
layer_genome = layer_genome.set_index(index)
layer_genome.columns = colnames
layer_genome = layer_genome.rename_axis("Mutation Types",
axis="columns")
# =============================================================================
# data_stat_folder = output+"/Data_Stats"
# try:
# if not os.path.exists(data_stat_folder):
# os.makedirs(data_stat_folder)
# except:
# print ("The {} folder could not be created".format("Data_Stats"))
#
# layer_genome.to_csv(data_stat_folder+"/Samples.text", sep = "\t", index_label=[layer_genome.columns.name])
# similarity_dataframe.to_csv(data_stat_folder+"/Similatiry_Data_All_Sigs.text", sep = "\t")
# del layer_genome
# for i in range(startProcess,endProcess+1):
# all_similirities_list[i-startProcess].to_csv(data_stat_folder+"/Similatiry_Data_Sig_"+str(i)+".text", sep="\t")
# =============================================================================
# record the samples
layer_genome.to_csv(output + "/Samples.txt",
sep="\t",
index_label=[layer_genome.columns.name])
#similarity_dataframe.to_csv(data_stat_folder+"/Similatiry_Data_All_Sigs"+str(H_iteration)+".text", sep = "\t")
del layer_genome
################################### Decompose the new signatures into global signatures #########################
processAvg = information[solution - startProcess][0]
processSTE = information[solution - startProcess][2]
signature_stabilities = information[solution - startProcess][4]
signature_total_mutations = information[solution - startProcess][5]
signature_stats = information[solution - startProcess][6]
all_similarities = information[solution - startProcess][7]
# create the folder for the final solution/ De Novo Solution
layer_directory1 = output + "/Suggested_Solution/De_Novo_Solution"
try:
if not os.path.exists(layer_directory1):
os.makedirs(layer_directory1)
except:
print("The {} folder could not be created".format("output"))
# make the texts for signature plotting
signature_stabilities = sub.signature_plotting_text(
signature_stabilities, "Stability", "float")
signature_total_mutations = sub.signature_plotting_text(
signature_total_mutations, "Total Mutations", "integer")
# make de novo solution(processAvg, allgenomes, layer_directory1)
listOfSignatures = sub.make_letter_ids(idlenth=processAvg.shape[1],
mtype=mutation_context)
allgenomes = pd.DataFrame(allgenomes)
exposureAvg = sub.make_final_solution(processAvg, allgenomes, listOfSignatures, layer_directory1, m, index, \
allcolnames, process_std_error = processSTE, signature_stabilities = signature_stabilities, \
signature_total_mutations = signature_total_mutations, signature_stats = signature_stats, penalty=penalty)
try:
# create the folder for the final solution/ Decomposed Solution
layer_directory2 = output + "/Suggested_Solution/Decomposed_Solution"
try:
if not os.path.exists(layer_directory2):
os.makedirs(layer_directory2)
except:
print("The {} folder could not be created".format("output"))
if processAvg.shape[
0] == 1536: #collapse the 1596 context into 96 only for the deocmposition
processAvg = pd.DataFrame(processAvg, index=index)
processAvg = processAvg.groupby(
processAvg.index.str[1:8]).sum()
genomes = pd.DataFrame(genomes, index=index)
genomes = genomes.groupby(genomes.index.str[1:8]).sum()
index = genomes.index
processAvg = np.array(processAvg)
genomes = np.array(genomes)
final_signatures = sub.signature_decomposition(
processAvg,
m,
layer_directory2,
genome_build=genome_build,
mutation_context=mutation_context)
# extract the global signatures and new signatures from the final_signatures dictionary
globalsigs = final_signatures["globalsigs"]
globalsigs = np.array(globalsigs)
newsigs = final_signatures["newsigs"]
processAvg = np.hstack([globalsigs, newsigs])
allsigids = final_signatures["globalsigids"] + final_signatures[
"newsigids"]
attribution = final_signatures["dictionary"]
background_sigs = final_signatures["background_sigs"]
genomes = pd.DataFrame(genomes)
#print(exposureAvg)
exposureAvg = sub.make_final_solution(processAvg, genomes, allsigids, layer_directory2, m, index, colnames, \
remove_sigs=True, attribution = attribution, denovo_exposureAvg = exposureAvg , background_sigs=background_sigs, penalty=penalty, genome_build=genome_build)
except:
print(
"\nWARNING!!! We apolozize we don't have a global signature database for the mutational context you provided. We have a database only for SBS96, DINUC and INDELS.\nTherefore no result for signature Decomposition is generated."
)
shutil.rmtree(layer_directory2)
sysdata = open(out_put + "/JOB_METADATA.txt", "a")
toc = datetime.datetime.now()
sysdata.write("\nDate and Clock time when the execution ended: " +
str(toc) + "\n")
sysdata.write("-------Job Status------- \n")
sysdata.write(
"Analysis of mutational signatures completed successfully! Total execution time: "
+ str(toc - tic) + ". Results can be found in: [" + out_put +
"] folder")
sysdata.close()
print(
"\n\n \nYour Job Is Successfully Completed! Thank You For Using SigProfiler Extractor.\n "
)
def remove_all_single_signatures(W,
H,
genomes,
metric="l2",
solver="nnls",
cutoff=0.05,
background_sigs=[],
verbose=False):
# make the empty list of the successfull combinations
signature_ids = sub.make_letter_ids(idlenth=W.shape[1], mtype="Signature ")
current_signatures = signature_ids
#print(current_signatures)
successList = [0, [], 0]
background_sig = copy.deepcopy(background_sigs)
#setting the base_similarity
base_H_index = list(np.nonzero(H)[0])
reg = nnls(W[:, base_H_index], genomes)
base_weights = reg[0]
if metric == "cosine":
# get the cos_similarity with sample for the oringinal W and H[:,i]
originalSimilarity = 1 - cos_sim(
genomes, np.dot(W[:, base_H_index], base_weights))
if verbose == True:
print("originalSimilarity", 1 - originalSimilarity)
elif metric == "l2":
originalSimilarity = np.linalg.norm(
genomes - np.dot(W[:, base_H_index], base_weights),
ord=2) / np.linalg.norm(genomes, ord=2)
if verbose == True:
print("originalSimilarity", originalSimilarity)
# make the original exposures of specific sample round
oldExposures = np.round(H)
# set the flag for the while loop
if len(oldExposures[np.nonzero(oldExposures)]) > 1:
Flag = True
else:
Flag = False
if metric == "cosine":
return oldExposures, 1 - originalSimilarity, cos_sim(
genomes, np.dot(W, H)
) #first value is the exprosure, second value is the similarity (based on the similarity matic), third value is the cosine similarity
else:
return oldExposures, originalSimilarity, cos_sim(
genomes, np.dot(W, H)
) #first value is the exprosure, second value is the similarity (based on the similarity matic), third value is the cosine similarity
# The while loop starts here
current_signatures = sub.get_items_from_index(signature_ids,
np.nonzero(oldExposures)[0])
while Flag:
# get the list of the indices those already have zero values
if len(successList[1]) == 0:
initialZerosIdx = list(np.where(oldExposures == 0)[0])
#get the indices to be selected
selectableIdx = list(np.where(oldExposures > 0)[0])
elif len(successList[1]) > 1:
initialZerosIdx = list(np.where(successList[1] == 0)[0])
#get the indices to be selected
selectableIdx = list(np.where(successList[1] > 0)[0])
else:
print("iteration is completed")
#break
# get the total mutations for the given sample
maxmutation = round(np.sum(genomes))
# new signature matrix omiting the column for zero
#Winit = np.delete(W, initialZerosIdx, 1)
Winit = W[:, selectableIdx]
# set the initial cos_similarity or other similarity distance
record = [np.inf, [], 1] #
# get the number of current nonzeros
l = Winit.shape[1]
background_sig = get_changed_background_sig_idx(
list(oldExposures), background_sig)
for i in range(l):
if i in background_sig:
continue
loopSelection = list(range(l))
del loopSelection[i]
W1 = Winit[:, loopSelection]
#initialize the guess
x0 = np.random.rand(l - 1, 1) * maxmutation
x0 = x0 / np.sum(x0) * maxmutation
#set the bounds and constraints
bnds = create_bounds([], genomes, W1.shape[1])
cons1 = {'type': 'eq', 'fun': constraints1, 'args': [genomes]}
#the optimization step
if solver == "slsqp":
sol = minimize(parameterized_objective2_custom,
x0,
args=(W1, genomes),
bounds=bnds,
constraints=cons1,
tol=1e-15)
#print (sol.success)
#print (sol.x)
#convert the newExposure vector into list type structure
newExposure = list(sol.x)
if solver == "nnls":
### using NNLS algorithm
reg = nnls(W1, genomes)
weights = reg[0]
newSample = np.dot(W1, weights)
normalised_weights = weights / sum(weights)
solution = normalised_weights * sum(genomes)
newExposure = list(solution)
#insert the loopZeros in its actual position
newExposure.insert(i, 0)
#insert zeros in the required position the newExposure matrix
initialZerosIdx.sort()
for zeros in initialZerosIdx:
newExposure.insert(zeros, 0)
# get the maximum value the new Exposure
#maxcoef = max(newExposure)
#idxmaxcoef = newExposure.index(maxcoef)
#newExposure = np.round(newExposure)
#if np.sum(newExposure)!=maxmutation:
#newExposure[idxmaxcoef] = round(newExposure[idxmaxcoef])+maxmutation-sum(newExposure)
newExposure = np.array(newExposure)
if verbose == True:
#print(newExposure)
print("\nRemoving {}".format(current_signatures[i]))
if metric == "cosine":
newSimilarity = 1 - cos_sim(genomes, newSample)
if verbose == True:
print("newSimilarity", 1 - newSimilarity)
elif metric == "l2":
newSimilarity = np.linalg.norm(genomes - newSample,
ord=2) / np.linalg.norm(genomes,
ord=2)
if verbose == True:
print("newSimilarity", newSimilarity)
difference = newSimilarity - originalSimilarity
if difference < 0:
difference = cutoff + 1e-100
if verbose == True:
print("difference", difference)
if difference < record[0]:
record = [difference, newExposure, newSimilarity]
#print("difference", difference)
if verbose == True:
print("--------------------------------------")
if verbose == True:
print(
"\n############################\n############################")
#print("Selected Exposure")
#print(record[1])
selected_sigs = sub.get_items_from_index(signature_ids,
np.nonzero(record[1])[0])
dropped_sig = list(set(current_signatures) - set(selected_sigs))
print("Dropped Signature: {}".format(dropped_sig))
current_signatures = selected_sigs
print("New Similarity", record[2])
print("Similarity Difference: {}".format(record[2] -
originalSimilarity))
print("Current Signatures: {}".format(selected_sigs))
print(
"****************************\n****************************\n\n"
)
#print ("This loop's selection is {}".format(record))
if record[0] > cutoff:
Flag = False
elif len(record[1][np.nonzero(record[1])]) == 1:
successList = record
Flag = False
else:
successList = record
background_sig = get_changed_background_sig_idx(
list(record[1]), background_sig)
originalSimilarity = record[2]
#print("The loop selection is {}".format(successList))
#print (Flag)
#print ("\n\n")
#print ("The final selection is {}".format(successList))
if len(successList[1]) == 0:
successList = [0.0, oldExposures, originalSimilarity]
if verbose == True:
print("\n")
print("Final Exposure")
print(successList[1])
if metric == "cosine":
print("Final Similarity: ", 1 - successList[2])
if metric == "l2":
print("Final Similarity: ", successList[2])
H = successList[1]
return H, successList[2], cos_sim(
genomes, np.dot(W, H)
) #first value is the exprosure, second value is the similarity (based on the similarity matic), third value is the cosine similarity
def decompose(signatures,
activities,
samples,
output,
genome_build="GRCh37",
verbose=False):
"""
Decomposes the De Novo Signatures into COSMIC Signatures and assigns COSMIC signatures into samples.
Parameters:
signatures: A string. Path to a tab delimited file that contains the signaure table where the rows are mutation types and colunms are signature IDs.
activities: A string. Path to a tab delimilted file that contains the activity table where the rows are sample IDs and colunms are signature IDs.
samples: A string. Path to a tab delimilted file that contains the activity table where the rows are mutation types and colunms are sample IDs.
output: A string. Path to the output folder.
genome_build = A string. The genome type. Example: "GRCh37", "GRCh38", "mm9", "mm10". The default value is "GRCh37"
verbose = Boolean. Prints statements. Default value is False.
Values:
The files below will be generated in the output folder.
Cluster_of_Samples.txt
comparison_with_global_ID_signatures.csv
Decomposed_Solution_Activities.txt
Decomposed_Solution_Samples_stats.txt
Decomposed_Solution_Signatures.txt
decomposition_logfile.txt
dendogram.pdf
Mutation_Probabilities.txt
Signature_assaignment_logfile.txt
Signature_plot[MutatutionContext]_plots_Decomposed_Solution.pdf
Example:
>>>from SigProfilerExtractor import decomposition as decomp
>>>signatures = "path/to/dDe_Novo_Solution_Signatures.txt"
>>>activities="path/to/De_Novo_Solution_Activities.txt"
>>>samples="path/to/Samples.txt"
>>>output="name or path/to/output.txt"
decomp.decompose(signatures, activities, samples, output, genome_build="GRCh37", verbose=False)
"""
processAvg = pd.read_csv(signatures, sep="\t", index_col=0)
exposureAvg = pd.read_csv(activities, sep="\t", index_col=0)
processAvg = np.array(processAvg)
genomes = pd.read_csv(samples, sep="\t", index_col=0)
mutation_type = str(genomes.shape[0])
#creating list of mutational type to sync with the vcf type input
if mutation_type == "78":
mutation_context = "DBS78"
elif mutation_type == "83":
mutation_context = "ID83"
else:
mutation_context = "SBS" + mutation_type
signature_names = sub.make_letter_ids(idlenth=processAvg.shape[1],
mtype=mutation_context)
exposureAvg.columns = signature_names
index = genomes.index
m = mutation_type
layer_directory2 = output
if not os.path.exists(layer_directory2):
os.makedirs(layer_directory2)
if processAvg.shape[
0] == 1536: #collapse the 1596 context into 96 only for the deocmposition
processAvg = pd.DataFrame(processAvg, index=index)
processAvg = processAvg.groupby(processAvg.index.str[1:8]).sum()
genomes = genomes.groupby(genomes.index.str[1:8]).sum()
index = genomes.index
processAvg = np.array(processAvg)
final_signatures = sub.signature_decomposition(
processAvg,
m,
layer_directory2,
genome_build=genome_build,
mutation_context=mutation_context)
#final_signatures = sub.signature_decomposition(processAvg, m, layer_directory2, genome_build=genome_build)
# extract the global signatures and new signatures from the final_signatures dictionary
globalsigs = final_signatures["globalsigs"]
globalsigs = np.array(globalsigs)
newsigs = final_signatures["newsigs"]
processAvg = np.hstack([globalsigs, newsigs])
allsigids = final_signatures["globalsigids"] + final_signatures["newsigids"]
attribution = final_signatures["dictionary"]
background_sigs = final_signatures["background_sigs"]
index = genomes.index
colnames = genomes.columns
result = sub.make_final_solution(processAvg, genomes, allsigids, layer_directory2, m, index, colnames, \
remove_sigs=True, attribution = attribution, denovo_exposureAvg = exposureAvg , penalty=0.01, background_sigs=background_sigs, verbose=verbose, genome_build=genome_build)
return result
def decompose(signatures,
activities,
samples,
output,
nnls_add_penalty=0.05,
nnls_remove_penalty=0.01,
initial_remove_penalty=0.05,
genome_build="GRCh37",
refit_denovo_signatures=True,
make_decomposition_plots=True,
connected_sigs=True,
verbose=False):
"""
Decomposes the De Novo Signatures into COSMIC Signatures and assigns COSMIC signatures into samples.
Parameters:
signatures: A string. Path to a tab delimited file that contains the signaure table where the rows are mutation types and colunms are signature IDs.
activities: A string. Path to a tab delimilted file that contains the activity table where the rows are sample IDs and colunms are signature IDs.
samples: A string. Path to a tab delimilted file that contains the activity table where the rows are mutation types and colunms are sample IDs.
output: A string. Path to the output folder.
genome_build = A string. The genome type. Example: "GRCh37", "GRCh38", "mm9", "mm10". The default value is "GRCh37"
verbose = Boolean. Prints statements. Default value is False.
Values:
The files below will be generated in the output folder.
Cluster_of_Samples.txt
comparison_with_global_ID_signatures.csv
Decomposed_Solution_Activities.txt
Decomposed_Solution_Samples_stats.txt
Decomposed_Solution_Signatures.txt
decomposition_logfile.txt
dendogram.pdf
Mutation_Probabilities.txt
Signature_assaignment_logfile.txt
Signature_plot[MutatutionContext]_plots_Decomposed_Solution.pdf
Example:
>>>from SigProfilerExtractor import decomposition as decomp
>>>signatures = "path/to/dDe_Novo_Solution_Signatures.txt"
>>>activities="path/to/De_Novo_Solution_Activities.txt"
>>>samples="path/to/Samples.txt"
>>>output="name or path/to/output.txt"
decomp.decompose(signatures, activities, samples, output, genome_build="GRCh37", verbose=False)
"""
processAvg = pd.read_csv(signatures, sep="\t", index_col=0)
exposureAvg = pd.read_csv(activities, sep="\t", index_col=0)
genomes = pd.read_csv(samples, sep="\t", index_col=0)
mutation_type = str(genomes.shape[0])
m = mutation_type
index = genomes.index
colnames = genomes.columns
listOfSignatures = processAvg.columns
#creating list of mutational type to sync with the vcf type input
if mutation_type == "78":
mutation_context = "DBS78"
elif mutation_type == "83":
mutation_context = "ID83"
elif mutation_type == "48":
mutation_context = "CNV48"
print("Mutation Type is: CNV")
paths = cosmic.__path__[0]
sigDatabase = pd.read_csv(paths + "/data/CNV_signatures.txt",
sep="\t",
index_col=0)
genomes = genomes.loc[sigDatabase.index]
processAvg = processAvg.loc[sigDatabase.index]
else:
mutation_context = "SBS" + mutation_type
processAvg = np.array(processAvg)
signature_names = sub.make_letter_ids(idlenth=processAvg.shape[1],
mtype=mutation_context)
exposureAvg.columns = signature_names
# create the folder for the final solution/ De Novo Solution
try:
if not os.path.exists(output):
os.makedirs(output)
except:
print("The {} folder could not be created".format("output"))
# make the texts for signature plotting
layer_directory1 = output + "/De_Novo_Solution"
try:
if not os.path.exists(layer_directory1):
os.makedirs(layer_directory1)
except:
print("The {} folder could not be created".format("De_Novo_Solution"))
listOfSignatures = sub.make_letter_ids(idlenth=processAvg.shape[1],
mtype=mutation_context)
genomes = pd.DataFrame(genomes)
denovo_exposureAvg = np.array(exposureAvg.T)
exposureAvg = sub.make_final_solution(processAvg, genomes, listOfSignatures, layer_directory1, m, index,\
colnames,denovo_exposureAvg = denovo_exposureAvg, add_penalty=nnls_add_penalty, remove_penalty=nnls_remove_penalty, initial_remove_penalty=initial_remove_penalty, connected_sigs=connected_sigs, refit_denovo_signatures=refit_denovo_signatures)
layer_directory2 = output + "/Decompose_Solution"
try:
if not os.path.exists(layer_directory2):
os.makedirs(layer_directory2)
except:
print(
"The {} folder could not be created".format("Decomposed_Solution"))
if processAvg.shape[
0] == 1536: #collapse the 1596 context into 96 only for the deocmposition
processAvg = pd.DataFrame(processAvg, index=index)
processAvg = processAvg.groupby(processAvg.index.str[1:8]).sum()
genomes = genomes.groupby(genomes.index.str[1:8]).sum()
index = genomes.index
processAvg = np.array(processAvg)
if processAvg.shape[
0] == 288: #collapse the 288 context into 96 only for the deocmposition
processAvg = pd.DataFrame(processAvg, index=index)
processAvg = processAvg.groupby(processAvg.index.str[2:9]).sum()
genomes = pd.DataFrame(genomes, index=index)
genomes = genomes.groupby(genomes.index.str[2:9]).sum()
index = genomes.index
processAvg = np.array(processAvg)
final_signatures = sub.signature_decomposition(
processAvg,
m,
layer_directory2,
genome_build=genome_build,
mutation_context=mutation_context,
add_penalty=nnls_add_penalty,
connected_sigs=connected_sigs,
remove_penalty=nnls_remove_penalty,
make_decomposition_plots=make_decomposition_plots)
#final_signatures = sub.signature_decomposition(processAvg, m, layer_directory2, genome_build=genome_build)
# extract the global signatures and new signatures from the final_signatures dictionary
globalsigs = final_signatures["globalsigs"]
globalsigs = np.array(globalsigs)
newsigs = final_signatures["newsigs"]
processAvg = np.hstack([globalsigs, newsigs])
allsigids = final_signatures["globalsigids"] + final_signatures["newsigids"]
attribution = final_signatures["dictionary"]
background_sigs = final_signatures["background_sigs"]
index = genomes.index
colnames = genomes.columns
result = sub.make_final_solution(processAvg, genomes, allsigids, layer_directory2, m, index, colnames, \
cosmic_sigs=True, attribution = attribution, denovo_exposureAvg = exposureAvg , background_sigs=background_sigs, verbose=verbose, genome_build=genome_build, add_penalty=nnls_add_penalty, remove_penalty=nnls_remove_penalty, initial_remove_penalty=initial_remove_penalty,connected_sigs=connected_sigs,refit_denovo_signatures=False)
return result
