def prepare_indelphi(seq, cut, celltype):
print(celltype)
inDelphi.init_model(celltype=celltype)
pred_df, stats = inDelphi.predict(seq, cut)
pred_df = inDelphi.add_mhless_genotypes(pred_df, stats)
pred_df = inDelphi.add_genotype_column(pred_df, stats)
pred_df = inDelphi.add_name_column(pred_df, stats)
freq = pred_df.loc[:, 'Predicted frequency']
pred_df.loc[:, 'Predicted frequency'] = freq / freq.sum()
pred_df = pred_df.sort_values(by=['Predicted frequency'], ascending=False)
return (pred_df)
ref = ref.upper()
m = re.search(grna, ref)
if m:
cut = m.end() - 3 #?
orientation = 1
else:
ref = str(Seq(ref).reverse_complement())
m = re.search(grna, ref)
orientation = -1
cut = m.end() - 3
return ref, cut, orientation
if __name__ == '__main__':
f = sys.argv[1] #input
inDelphi.init_model(celltype='U2OS')
lines = open(f).readlines()
h = lines[0].rstrip(
) + "\twt_cut\twt_orientation\tmut_cut\tmut_orientation"
Fr = open(f.split(".")[0] + "_ori.xls", "w")
Fr.write(h + "\n")
for x in lines[1:]:
x = x.rstrip()
l = x.split("\t")
wt_grna = l[2].upper()
mut_grna = l[3].upper()
wt_ref = l[6].upper()
mut_ref = l[7].upper()
wt_ref, wt_cut, wt_orientation = my_cut(wt_grna, wt_ref)
mut_ref, mut_cut, mut_orientation = my_cut(mut_grna, mut_ref)
#print wt_cut,wt_orientation
import inDelphi
from scipy.stats import linregress
import numpy as np
import pandas as pd
sys.path.append("/cluster/bh0085")
from mybio import util
from _config import REDUCED_LIB, OUT_PLACE
import imp
if not "__file__" in vars(): __file__ = "f_test"
NAME = util.get_fn(__file__)
OUT_DIR = os.path.join(OUT_PLACE, NAME)
util.ensure_dir_exists(OUT_DIR)
all_predictions = pd.DataFrame()
for model in ["mESC", "U2OS"]:
imp.reload(inDelphi)
inDelphi.init_model(celltype=model)
for k, row in REDUCED_LIB.iterrows():
target_seq = row[
"Designed sequence (61-bp, cutsite at position 34 by 0-index)"]
CUTSITE = 34
pred_df, stats = inDelphi.predict(target_seq, CUTSITE)
pred_df = inDelphi.add_mhless_genotypes(pred_df, stats)
pred_df = pred_df.assign(**{"celltype": model, "libid": k})
all_predictions = all_predictions.append(pred_df, ignore_index=True)
all_predictions.to_csv(os.path.join(OUT_DIR, "indelphi_genotypes.csv"),
index=False)
sim_info = pd.DataFrame()
bar = Bar('Simulating sequences:', max=len(target_seqs_data))
for gRNA_id, target_seq in target_seqs_data.items():
# Calculate activity score using doench for that guide
## 30mer: 4bp 5', 20bp guide, 3bp PAM, 3bp 5'
seq_for_doench = target_seq[cutsite - 21:cutsite + 9]
doench_score = calc_doench_score(seq_for_doench)
## doench score is from 0 to 100, scale to get numb of edited reads
n_edit_seqs = round(doench_score * sim_reads /
100) #round to have an integer number of reads
# Calculate editing outcomes
inDelphi.init_model(celltype='mESC')
pred_df, stats = inDelphi.predict(target_seq, cutsite)
pred_df = inDelphi.add_mhless_genotypes(pred_df, stats)
#pred_df = inDelphi.add_genotype_column(pred_df,stats)
## adds gaps in the deletions, use add_genotype_column to avoid gaps, but sequences could be confused
pred_df = inDelphi.add_genotype_column_wgaps(pred_df, stats)
pred_frequency = np.array(pred_df["Predicted frequency"])
# normalize probabilities to sum exactly 1
pred_frequency /= pred_frequency.sum()
# Simulate data
## first, create the edited reads
edit_seqs = np.random.choice(pred_df["Genotype"],
p=pred_frequency,
size=(n_edit_seqs))
## add non edited reads up to the sim_reads objective
def runDelphi(dataframe, truth_reference, max_oligos = 1, cell_type = 'mESC', file_prefix = "inDelphi", pathout = ""):
# init model
inDelphi.init_model(celltype = cell_type)
print("init runDelphi.."); time.sleep(1)
# Open files so that data can be written
outfile1 = open(pathout + file_prefix + "_" + cell_type + "_statistics.txt", 'w')
outfile2 = open(pathout + file_prefix + "_" + cell_type + "_indels_frequency_predicted.txt", 'w')
outfile3 = open(pathout + file_prefix + "_" + cell_type + "_indels_frequency_actual.txt", 'w')
outfile4 = open(pathout + file_prefix + "_" + cell_type + "_modelstats.txt", 'w')
# prepare dictionaries for storing relevant data
mutations_reference_actual = compile_mutations()
mutations_reference_pred = compile_mutations()
# Save some info to the outfiles
outfile2.write("oligo_id" + "\t" + "\t".join([v for v in mutations_reference_pred.keys()]) + "\n")
outfile3.write("oligo_id" + "\t" + "\t".join([v for v in mutations_reference_pred.keys()]) + "\n")
errs = list()
klavg = 0
# Variables for loop
elapsed_time = 0
iteration = 0; n_oligos = 0
while n_oligos < max_oligos:
start = time.time()
oligo = dataframe.iloc[[iteration]].values.tolist()[0]
oligo_id = oligo[0]; seq = oligo[2]; pam_idx = int(oligo[6])
sys.stdout.write("Query [{0}/{1}]: {2}\n".format(n_oligos, max_oligos, oligo_id))
try:
# Get a dataframe from the ground truth and
# the the prediction of in-delphi. Merge afterwards.
#print("pam=",pam_idx)
actual = format_ground_truth(truth_reference, oligo_id)
prediction, stats = format_predction(seq, pam_idx) # cutsite is modified in func
df = actual.merge(prediction,how='outer', on='type' ).fillna(0)
## Run statistical analysis
loss = L(list(df['pred']),list(df['actual']))
stat01 = str(1-stats['Frameshift frequency']/100) #stats_frameshift(df, 'pred')
stat02 = stats_frameshift(df, 'actual')
acc01 = accuracy_type_agreement(x,y, top = 1)
acc02 = accuracy_type_agreement(x, y, top = 2)
acc03 = accuracy_type_agreement(x, y, top = 3)
acc04 = None #accuracy_type_disrupt_reading_frame(df)
acc05 = None #accuracy_type_disrupt_reading_frame(df)
acc06 = None #accuracy_type_disrupt_reading_frame(df)
stat_freq_pred = stats_frequency_mutations(df, 'pred', mutations_reference_pred.copy())
stat_freq_actual = stats_frequency_mutations(actual, 'actual', mutations_reference_actual.copy())
#time.sleep(5)
outfile2.write(oligo_id + "\t" + "\t".join([str(v) for v in x]) +"\n") #predicted
outfile3.write(oligo_id + "\t" + "\t".join([str(v) for v in y]) +"\n")
outfile1.write("{0}\t{1}\t{2}\t{3}\t{4}\t{5}\t{6}\t{7}\t{8}\t{9}\n".format(oligo_id, loss, acc01, acc02, acc03, acc04, acc05, acc06, stat01, stat02))
outfile4.write("\t".join([str(v) for k, v in stats.items()]))
# Gather statistics about the code, i.e. timing
klavg += loss
n_oligos += 1
end = time.time(); elapsed_time += (end-start)
if n_oligos % 20 == 0:
print("KL=",np.round(klavg/n_oligos,5))
time.sleep(1)
sys.stdout.write("SElapsed Time: " + str(round(elapsed_time,0)) + "s\n")
except Exception as e:
sys.stdout.write("\n>>>ERROR: " + str(e) + "\n")
errs.append(oligo_id)
print("CUR ERRORS: " + str(errs))
time.sleep(3)
pass
iteration += 1
# close all files
outfile1.close(); outfile2.close()
outfile3.close(); outfile4.close()