def run(profile,tndiff,tstart,tend,left=None,right=None):
MRTp, MRTs, RFDs, Rept_time, single_mol_exp, pos_time_activated_ori, It,wRFD = get_fast_MRT_RFDs(
nsim, profile, tndiff, kon=kon,
fork_speed=fork_speed, dori=20*5/resolution,single_mol_exp=False,continuous=args.continuous,wholeRFD=True)
#print("check", np.sum(d3p), np.sum(np.ones_like(d3p)*np.sum(d3p)*0.1/len(d3p)))
#print("Start, end",tstart,tend,tndiff)
# Compare to exp data
"""
MRTpearson, MRTstd, MRT = compare(
MRTp[::10//resolution], "MRT", cell, res=10, ch=ch, start=start, end=end, return_exp=True)
RFDpearson, RFDstd, RFD = compare(RFDs, "OKSeq", cell, res=resolution, ch=ch,
start=start, end=end, return_exp=True, rescale=1/resolution)
"""
if left is not None:
RFDs += 1.5*left
if right is not None:
RFDs += 1.5*right
if cell != "Cerevisae":
MRTpearson, MRTstd, MRT = compare(
MRTp[::10//resolution], "MRT", cell, res=10, ch=ch, start=tstart, end=tend, return_exp=True)
RFDpearson, RFDstd, RFD = compare(RFDs, "OKSeq", cell, res=resolution, ch=ch,
start=tstart, end=tend, return_exp=True, rescale=1/resolution)
else:
MRTpearson, MRTstd, MRT = compare(
MRTp, "MRT", cell, res=1, ch=ch, start=tstart, end=tend, return_exp=True)
RFDpearson, RFDstd, RFD = compare(RFDs, "OKSeq", cell, res=1, ch=ch,
start=tstart, end=tend, return_exp=True, rescale=1,nanpolate=True,smoothf=2)
RFDs = smooth(RFDs,2)
if args.RFDo:
#print(RFDstd)
return RFDstd,MRTp,MRT,RFD,RFDs,wRFD
else:
return MRTstd+RFDstd,MRTp,MRT,RFD,RFDs,wRFD
#if len(weights)== 0
print(weights)
#print(ast.literal_eval(weights))
try:
weights = ast.literal_eval(weights)
except:
if weights not in ["submed"]:
print("Warning unrecon option %s" % weights)
pass
if type(weights) == list:
d3p = np.zeros_like(raw)
#print(weights)
for smoothv, weightv in weights:
d3p += smooth(raw, int(smoothv)) * weightv
raw = d3p
else:
#print(type(weights),np.nanmean(raw))
if weights == "submed":
raw -= np.nanmedian(raw)
d3p = raw
#print(weights, "here")
if args.nan0 and "Exp" not in signal:
d3p[np.isnan(d3p)] = 0
btosee.append([x, d3p, signal])
if sup_sig != None:
btosee.append(sup_sig)
def fun(x, alpha):
global iter
global gscore
signal = init_x0
signal[where] = x
if np.sum(x < 0) > 0:
return 2
filen = root + "/tmp.csv"
d = pd.DataFrame({
"chrom": whole_info.chrom,
"chromStart": whole_info.chromStart,
"chromEnd": whole_info.chromStart,
"signalValue": signal
})
d.to_csv(filen, index=False)
process = subprocess.Popen(command + " --signal %s --name %s" %
(filen, root + "/tmp"),
shell=True,
stdout=subprocess.PIPE)
process.wait()
scored = pd.read_csv(root + "/tmpglobal_corre.csv")
c1 = float(scored["MRTp"][0].split(",")[0][1:])
c1 = 0
c2 = float(scored["RFDp"][0].split(",")[0][1:])
print(scored)
if iter % 10 == 0:
print("every10", c1, c2)
score = 2 - c1 - c2 # + 0.01 * (np.sum(x)-1)**2
if iter == 0:
print("Initial value", gscore)
gscore = score
if score < gscore:
print("New minimum %.3f , old %.3f", score, gscore)
print(c1, c2)
d.to_csv(root + "_%i.csv" % iter, index=False)
gscore = score
iter += 1
scored = pd.read_csv(root + "/tmpglobal_profiles.csv")
def delta(s):
return np.array(s)[1:] - np.array(s)[:-1]
deltas = smooth(
delta(scored["RFDs"]) - delta(scored["RFDe"]), args.extension)
direction = deltas[where]
direction /= np.mean(np.abs(direction))
x -= alpha * direction * x
x[x < 0] = 0
return score, x
start=start,
end=end,
resolution=resolution,
raw=False)
x, CNV = replication_data(cell,
"CNV",
chromosome=ch,
start=start,
end=end,
resolution=resolution,
raw=False)
CNV[CNV == 0] = 2
DNaseI[np.isnan(DNaseI)] = 0
DNaseI /= CNV
DNaseIsm = smooth(DNaseI, 100)
DNaseIsm /= np.mean(DNaseIsm)
d3p *= DNaseIsm
d3p[np.isnan(d3p)] = 0
d3p[np.isinf(d3p)] = 0
print(np.sum(np.isnan(d3p)))
# pylab.plot(d3p)
# pylab.show()
elif args.signal == "ARSpeak":
x, d3p = replication_data(cell,
"ARS",
chromosome=ch,
start=start,
end=end,
resolution=resolution,
raw=False)
x, CNV = replication_data(cell,
"CNV",
chromosome=ch,
start=start,
end=end,
resolution=resolution,
raw=False)
CNV[CNV == 0] = 2
DNaseI[np.isnan(DNaseI)] = 0
DNaseI /= CNV
DNaseIsm = smooth(DNaseI, 100)
DNaseIsm /= np.mean(DNaseIsm)
d3p *= DNaseIsm
d3p[np.isnan(d3p)] = 0
d3p[np.isinf(d3p)] = 0
print(np.sum(np.isnan(d3p)))
# pylab.plot(d3p)
# pylab.show()
else:
x, d3p = replication_data(cell,
args.signal,
chromosome=ch,
start=start,
def load_signal(name,
marks=[
'H2az', 'H3k27ac', 'H3k79me2', 'H3k27me3', 'H3k9ac',
'H3k4me2', 'H3k4me3', 'H3k9me3', 'H3k4me1', 'H3k36me3',
"H4k20me1"
],
targets=["initiation"],
t_norm=None,
smm=None,
wig=True,
augment=None):
df = pd.read_csv(name)
#wig = True
if "signal" in df.columns:
df["initiation"] = df["signal"]
if wig:
lm = [
"DNaseI", "initiation", "Meth", "Meth450", "RFDs", "MRTs", "RFDe",
"MRTe", "AT_20"
]
marks0 = [m + "wig" for m in marks if m not in lm]
for sm in lm:
if sm in marks:
marks0 += [sm]
assert (len(marks) == len(marks0))
marks = marks0
if "notnan" in df.columns:
print("Found notnan")
notnan = df["notnan"]
else:
notnan = []
df = df[targets + marks]
print(df.describe())
yinit = [df.pop(target) for target in targets]
# print(yinit.shape,"Yinit shape")
if t_norm is not None:
transform_norm = t_norm
for col in df.columns:
# print(col)
if col not in [
"DNaseI", "initiation", "Meth", "Meth450", "RFDe", "MRTe",
"RFDs", "MRTs"
]:
df[col] = transform_norm(df[col])
elif col == "DNaseI":
df[col] = transform_DNase(df[col])
elif col in ["initiation", "Stall"]:
df[col] = df[col] / np.max(df[col])
elif "Meth" in col:
df[col] = transform_norm_meth(df[col])
elif "RFD" in col:
if "RFD" in col:
# print("Nanpo")
df[col] = nan_polate(df[col])
if smm is not None:
df[col] = smooth(df[col], smm)
df[col] = (df[col] + 1) / 2
elif "MRT" in col:
if "MRT" in col:
df[col] = nan_polate(df[col])
if augment == "test":
for asm in [10, 50, 200]:
df[col + f"_sm_{asm}"] = smooth(
nan_polate(df[col]), asm)
df[col + f"_sm_{asm}"] -= np.mean(df[col +
f"_sm_{asm}"])
df[col + f"_sm_{asm}"] /= np.std(df[col +
f"_sm_{asm}"])
pass
if np.sum(np.isnan(df[col])) != 0:
raise "NanVal"
print(np.max(yinit[0]), "max")
print(df.describe())
yinit0 = []
for y, t in zip(yinit, targets):
if t in ["initiation", "Stall"]:
trunc = y / np.max(y) #np.percentile(y,99)
#trunc[trunc>1] = 1
yinit0.append(trunc)
elif t == "DNaseI":
yinit0.append(transform_DNase(y))
elif t == "OKSeq":
yinit0.append((y + 1) / 2)
else:
raise "Undefined target"
yinit = np.array(yinit0).T
yinit[np.isnan(yinit)] = 0
# print(yinit.shape)
return df, yinit, notnan
def run(profile, tndiff, tstart, tend, actualProfile=None):
res = args.resolution
if actualProfile is not None:
tmpProfile = actualProfile.copy()
#print(tmpProfile.shape,np.array(profile).shape,start,tstart,tend,tmpProfile[tstart-start:tend-start].shape,tstart-start,tend-start)
weight_t = np.sum(tmpProfile[tstart - start:tend - start])
tmpProfile[tstart - start:tend -
start] = weight_t * np.array(profile) / np.sum(profile)
else:
tmpProfile = profile
MRTp, MRTs, RFDs, Rept_time, single_mol_exp, pos_time_activated_ori, It, wRFD = get_fast_MRT_RFDs(
nsim,
tmpProfile,
tndiff,
kon=kon,
fork_speed=fork_speed,
dori=20 * 5 / resolution,
single_mol_exp=False,
continuous=args.continuous,
wholeRFD=True)
#print("check", np.sum(d3p), np.sum(np.ones_like(d3p)*np.sum(d3p)*0.1/len(d3p)))
#print("Start, end",tstart,tend,tndiff)
# Compare to exp data
"""
MRTpearson, MRTstd, MRT = compare(
MRTp[::10//resolution], "MRT", cell, res=10, ch=ch, start=start, end=end, return_exp=True)
RFDpearson, RFDstd, RFD = compare(RFDs, "OKSeq", cell, res=resolution, ch=ch,
start=start, end=end, return_exp=True, rescale=1/resolution)
"""
if cell != "Cerevisae":
if actualProfile is not None:
MRTpearsonl, MRTstdl, MRTl = compare(
MRTp[int((tstart - start) / res):int((tend - start) /
res)][::10 // resolution],
"MRT",
cell,
res=10,
ch=ch,
start=tstart,
end=tend,
return_exp=True)
RFDpearsonl, RFDstdl, RFDl = compare(
RFDs[int((tstart - start) / res):int((tend - start) / res)],
"OKSeq",
cell,
res=resolution,
ch=ch,
start=tstart,
end=tend,
return_exp=True,
rescale=1 / resolution)
else:
RFDstdl = 0
MRTpearson, MRTstd, MRT = compare(MRTp[::10 // resolution],
"MRT",
cell,
res=10,
ch=ch,
start=start,
end=end,
return_exp=True)
RFDpearson, RFDstd, RFD = compare(RFDs,
"OKSeq",
cell,
res=resolution,
ch=ch,
start=start,
end=end,
return_exp=True,
rescale=1 / resolution)
else:
if actualProfile is not None:
MRTpearsonl, MRTstdl, MRTl = compare(
MRTp[int((tstart - start) / res):int((tend - start) / res)],
"MRT",
cell,
res=1,
ch=ch,
start=tstart,
end=tend,
return_exp=True)
RFDpearsonl, RFDstdl, RFDl = compare(
RFDs[int((tstart - start) / res):int((tend - start) / res)],
"OKSeq",
cell,
res=1,
ch=ch,
start=tstart,
end=tend,
return_exp=True,
rescale=1,
nanpolate=True,
smoothf=2)
else:
RFDstdl = 0
MRTpearson, MRTstd, MRT = compare(MRTp,
"MRT",
cell,
res=1,
ch=ch,
start=start,
end=end,
return_exp=True)
RFDpearson, RFDstd, RFD = compare(RFDs,
"OKSeq",
cell,
res=1,
ch=ch,
start=start,
end=end,
return_exp=True,
rescale=1,
nanpolate=True,
smoothf=2)
RFDs = smooth(RFDs, 2)
if args.RFDo:
print(RFDstd, RFDstdl)
return RFDstd + RFDstdl, MRTp, MRT, RFD, RFDs, wRFD
else:
return MRTstd + RFDstd, MRTp, MRT, RFD, RFDs, wRFD
start=start,
end=end,
resolution=resolution,
raw=False)
x, CNV = replication_data(cell,
"CNV",
chromosome=ch,
start=start,
end=end,
resolution=resolution,
raw=False)
CNV[CNV == 0] = 2
DNaseI[np.isnan(DNaseI)] = 0
DNaseI /= CNV
DNaseIsm = smooth(DNaseI, 100)
DNaseIsm /= np.mean(DNaseIsm)
d3p *= DNaseIsm
d3p[np.isnan(d3p)] = 0
d3p[np.isinf(d3p)] = 0
print(np.sum(np.isnan(d3p)))
# pylab.plot(d3p)
# pylab.show()
else:
x, d3p = replication_data(cell,
args.signal,
chromosome=ch,
start=start,