-
Notifications
You must be signed in to change notification settings - Fork 0
/
EAST2.py
277 lines (248 loc) · 11.8 KB
/
EAST2.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
import numpy as np
from timeit import time
from numba import guvectorize,float64,int64
import os
from scipy import sparse
from detect_peaks import detect_peaks
import importlib
pnd = importlib.find_loader('pandas')
PANDAS_INSTALLED = pnd is not None
if PANDAS_INSTALLED:
import pandas as pd
RESOLUTION = 5000
if RESOLUTION < 500000:
W = 20 # insulation window size
else:
W = 10
maxL = 2*int(np.round(3200000/RESOLUTION)) + 1 # maximum length of TAD allowed
Nfactor = 0.35 # normalization factor: larger values lead to smaller TADs
class cellType: # add your own data
K526 = 'K526'
hES = 'hES'
mES = 'mES'
class dataType:
Dixon = 'Dixon'
Rao = 'Rao'
CELLTYPE = cellType.K526
DataType = dataType.Rao
if CELLTYPE=='hIMR90' or CELLTYPE=='hES' or CELLTYPE=='K526':
NUM_OF_CHRMS = 22
elif CELLTYPE == 'mES' or CELLTYPE == 'mCO':
NUM_OF_CHRMS = 20
for CHRM in range(1,NUM_OF_CHRMS+1):
if DataType == dataType.Rao:
print('Loading Chromosome '+ str(CHRM))
st = time.time()
if PANDAS_INSTALLED:
if RESOLUTION < 1000000:
name = str(int(RESOLUTION/1000))+'kb'
else:
name = str(int(RESOLUTION/1000000))+'mb'
chr1Data = pd.read_csv(os.path.abspath(os.sep)+'Dataset/'+CELLTYPE+'/'+name+'_resolution_intrachromosomal/chr'+str(CHRM)+'/MAPQGE30/chr'+str(CHRM)+'_'+name +'.RAWobserved',sep='\t',header=None)
chr1Data = chr1Data.values
chr1Data[:,0:2] = np.floor(chr1Data[:,0:2]/RESOLUTION)
knorm = pd.read_csv(os.path.abspath(os.sep)+'Dataset/'+CELLTYPE+'/'+name+'_resolution_intrachromosomal/chr'+str(CHRM)+'/MAPQGE30/chr'+str(CHRM)+'_'+ name+'.KRnorm',sep='\t',header=None)
else:
chr1Data = np.genfromtxt(os.path.abspath(os.sep)+'Dataset/'+CELLTYPE+'/'+str(int(RESOLUTION/1000))+'kb_resolution_intrachromosomal/chr'+str(CHRM)+'/MAPQGE30/chr'+str(CHRM)+'_'+name+'.RAWobserved')
chr1Data[:,0:2] = np.round(chr1Data[:,0:2]/RESOLUTION)
knorm = np.genfromtxt(os.path.abspath(os.sep)+'Dataset/'+CELLTYPE+'/'+str(int(RESOLUTION/1000))+'kb_resolution_intrachromosomal/chr'+str(CHRM)+'/MAPQGE30/chr'+str(CHRM)+'_'+name+'.KRnorm')
# Normalizing the data
#knorm = knorm.values.T.tolist()
#knorm = np.array(knorm[0])
#chr1Data[:,2] = np.divide(chr1Data[:,2],np.multiply(knorm[np.array(chr1Data[:,1],dtype=np.int)],knorm[np.array(chr1Data[:,0],dtype=np.int)]))
n = np.max([np.max(chr1Data[:,0]),np.max(chr1Data[:,1])])+1
chr1 = sparse.csr_matrix((chr1Data[:,2],(chr1Data[:,0],chr1Data[:,1])),shape = (n,n))
# comment this line if you don't have enough memory to store the dense matrix for higher resolution
chr1 = chr1.todense()
print('time to read the chromosome',CHRM,time.time()-st)
elif DataType == dataType.Dixon: # Dixon data Type
print('Loading Chromosome '+ str(CHRM))
st = time.time()
if PANDAS_INSTALLED:
chr1 = pd.read_csv(os.path.abspath(os.sep)+'Dataset/'+CELLTYPE+'/nij/nij.chr'+str(CHRM),sep='\t',header=None)
chr1 = chr1.values
else:
chr1 = np.genfromtxt(os.path.abspath(os.sep)+'Dataset/'+CELLTYPE+'/nij/nij.chr'+str(CHRM))
print('time to read the chromosome ',CHRM,':',time.time()-st)
N = chr1.shape[0]
# compute the integral image (We ignore the values on the diagonal)
st = time.time()
intgMat = np.zeros([2*maxL,N],dtype=np.float64)
I = np.arange(N-1) # delta = 1
intgMat[1,I+1] = chr1[I,I+1] + intgMat[0,I+1] + intgMat[0,I]
for delta in range(2,2*maxL):
I = np.arange(N-delta)
intgMat[delta,I+delta] = chr1[I,I+delta] + intgMat[delta-1,I+delta] + intgMat[delta-1,I+delta-1] - intgMat[delta-2,I+delta-1]
print('time to compute the integral image:',time.time()-st)
#print('********************** TAD DETECTION *****************************')
# TAD Detection
@guvectorize([(float64[:,:], int64[:], int64[:], float64[:])], '(m,p),(),()->()',target='parallel')
def score(intgMAT,i,l,res):
i_indent = maxL
j_indent = maxL
wScore = 0
pixel = i[0]
if(l[0]<=maxL or l[0]<5):
if l[0] % 2 == 0:
w = (l[0])/2
pixel = pixel + w
w2 = np.math.ceil(w/5)
A = [int(pixel-w-l[0]+i_indent),int(pixel-w+j_indent)]
B = [int(pixel-w-l[0]+i_indent),int(pixel+w+j_indent)]
D = [int(pixel-w+i_indent),int(pixel+w+j_indent)]
E = [int(pixel-w+i_indent),int(pixel+w+l[0]+j_indent)]
F = [int(pixel+w+i_indent),int(pixel+w+l[0]+j_indent)]
else:
w = (l[0]-1)/2
pixel = pixel + w
w2 = np.math.ceil(w/5)
A = [int(pixel-w-l[0]-1+i_indent),int(pixel-w-1+j_indent)]
B = [int(pixel-w-l[0]-1+i_indent),int(pixel+w+j_indent)]
D = [int(pixel-w-1+i_indent),int(pixel+w+j_indent)]
E = [int(pixel-w-1+i_indent),int(pixel+w+l[0]+j_indent)]
F = [int(pixel+w+i_indent),int(pixel+w+l[0]+j_indent)]
wScore = intgMAT[D[1]-D[0]+i_indent, D[1]]
res[0] = wScore/np.power(l[0],Nfactor)
else:
res[0]=0
@guvectorize([(float64[:,:], int64[:], int64[:], float64[:])], '(m,p),(),(o)->(o)',target='parallel')
def det_score(intgMAT,i,w,res):
i_indent = maxL
j_indent = maxL
B = [int(i[0]-w[0]+i_indent),int(i[0]+j_indent)]
C = [int(i[0]-w[0]+i_indent),int(i[0]+w[0]+j_indent)]
F = [int(i[0]+i_indent),int(i[0]+w[0]+j_indent)]
a = intgMAT[B[1]-B[0]+i_indent, B[1]] #left
f = intgMAT[F[1]-F[0]+i_indent, F[1]] #right
res[0] = intgMAT[C[1]-C[0] +i_indent, C[1]] - a - f #center
res[1] = (a - res[0])
res[2] = (f - res[0])
@guvectorize([(float64[:,:],int64[:],int64[:],float64[:], int64[:], int64[:],int64[:],int64[:],int64[:], float64[:])], '(u,v),(s),(s),(r),(p),(q),(p),(),(n)->(n)',target='parallel')
def parDP(scores,i2I,j2J,T,startPeaks,endPeaks,prev_end,j,k, res):
for counter in range(k.shape[0]):
res[counter] = T[prev_end[k[counter]]] + scores[i2I[startPeaks[k[counter]]], j2J[endPeaks[j[0]]]]#scores[startPeaks[k[counter]], endPeaks[j[0]]]
def findTADs(intgMAT):
st = time.time()
intgMAT2 = np.lib.pad(intgMAT, ((maxL, maxL), (maxL, maxL)), 'constant', constant_values=0)
# find potential starts/ends of TADs
det_scores = np.zeros([N,3],dtype=np.float64)
det_scores = det_score(intgMAT2,np.arange(N),[W,W,W])
S = np.sum(det_scores[:,0])/N
det_scores[:,1] = det_scores[:,1]/S
det_scores[:,2] = det_scores[:,2]/S
start_peaks = np.asarray(detect_peaks(det_scores[:,2], mpd=2),dtype=np.int64) #np.median(det_scores[:,1])
end_peaks = np.asarray(detect_peaks(det_scores[:,1], mpd=2),dtype=np.int64) #np.median(det_scores[:,2])
# we need to know the preceding end for each start
prev_end = np.zeros(len(start_peaks),dtype=np.int64)
k_old = 0
for i in range(len(start_peaks)):
k = k_old
for j in range(k,len(end_peaks)):
if end_peaks[k] <= start_peaks[i]:
k = k + 1
else:
break
if k==0:
prev_end[i] = -1
k_old = 0
elif end_peaks[k-1] < start_peaks[0]:
prev_end[i] = -1
k_old = k-1
else:
prev_end[i] = k-1
k_old = k - 1
# we need to know the preceding start for each end
prev_start = np.zeros(len(end_peaks),dtype=np.int64)
k_old = 0
for i in range(len(end_peaks)):
k = k_old
for j in range(k,len(start_peaks)):
if start_peaks[k] <= end_peaks[i]:
k = k + 1
else:
break
if k==0:
prev_start[i] = -1
k_old = 0
else:
prev_start[i] = k-1
k_old = k - 1
# pre-compute scores for start/end combinations
M1 = len(start_peaks)
M2 = len(end_peaks)
rep = []
indx = []
for i in range(M1):
ind = prev_end[i]
indx.append(min(ind+1,len(end_peaks)))
if ind == -1:
rep.append(M2)
elif ind == M2:
rep.append(0)
else:
rep.append(M2 - ind - 1)
long_i = np.repeat(start_peaks,rep)
long_j = np.zeros(np.sum(rep),dtype=np.int64)
for i in range(M1):
start = sum(rep[0:i])
long_j[start:start+rep[i]] = end_peaks[indx[i]:len(end_peaks)]
long_l = long_j - long_i + 1
temp_s = score(intgMAT2,long_i,long_l) #long_i is the center of a domain; long_l is the length of that domain
# build a dense representation of the 'scores' but much smaller than N*N (for higher resolutions)
i2I = np.zeros(N,dtype=np.int64)
for i in range(len(start_peaks)):
i2I[start_peaks[i]] = i
j2J = np.zeros(N,dtype=np.int64)
for j in range(len(end_peaks)):
j2J[end_peaks[j]] = j
scores = np.zeros([len(start_peaks),len(end_peaks)],dtype=np.float64)
scores[i2I[long_i],j2J[long_j]] = temp_s
# Dynamic Programming applies on potential start/end TAD boundaries we have found above
T = np.zeros(len(end_peaks),dtype=np.float64)
backT = np.zeros(len(end_peaks),dtype=np.int64) # shows the index of the start point (out of all start points) for each end point
k_old = 0
for j in range(len(end_peaks)):
if prev_start[j] == -1:
backT[j] = -1
continue
k = k_old
kk = np.arange(int(prev_start[j])+1)
vals = parDP(scores,i2I,j2J,T,start_peaks,end_peaks,prev_end,j,kk)
T[j] = np.max([np.max(vals),T[j-1]])
backT[j] = np.argmax(vals)
k_old = k
# Extracting TADs based on backT
counter = len(end_peaks) - 1
tadCount = 0
TADx1 = []
TADx2 = []
while True:# backT[counter] != 0 and backT[counter] != -1:
if backT[counter] == -1:
break
elif backT[counter] == 0:
TADx2.append(end_peaks[counter])
TADx1.append(start_peaks[0])
tadCount = tadCount + 1
break
elif counter > 0:
TADx2.append(end_peaks[counter])
TADx1.append(start_peaks[backT[counter]])
tadCount = tadCount + 1
counter = prev_end[backT[counter]]
if prev_end[backT[counter]] == counter:
counter = counter - 1
if counter == -1:
break
elif counter == 0:
TADx2.append(end_peaks[counter])
TADx1.append(start_peaks[backT[counter]])
tadCount = tadCount + 1
break
print('Number of TADs:',tadCount)
TADx1.sort()
TADx2.sort()
TADx1 = (np.asarray(TADx1)).reshape([len(TADx1),1])
TADx2 = (np.asarray(TADx2)).reshape([len(TADx2),1])
TAD = np.concatenate((TADx1,TADx2),axis=1)
np.savetxt(CELLTYPE+'_nij_chr'+str(CHRM)+'_'+str(N),TAD,delimiter=' ',fmt='%d')
findTADs(intgMat)