-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathutils.py
359 lines (322 loc) · 13.3 KB
/
utils.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
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
import os
import sys
import Bio
import scipy as sp
import numpy as np
import pandas as pd
import pickle as pkl
import networkx as nx
import scipy.stats as stats
import scipy.sparse as sparse
import subprocess
from Bio import SeqIO
from Bio.Seq import Seq
from Bio.SeqRecord import SeqRecord
import torch
from torch import nn
from torch.nn import functional as F
def make_diamond_db(cpu: int):
diamond_db_bp = "database/database.dmnd"
aa_fp = "database/protein.fasta"
make_diamond_cmd = ['diamond', 'makedb', '--threads', str(cpu), '--in', aa_fp, '-d', diamond_db_bp]
print("Creating Diamond database...")
res = subprocess.run(make_diamond_cmd, check=True, stdout=subprocess.PIPE)
if res.returncode != 0:
print('Error creating Diamond database')
exit(1)
diamond_db_fp = diamond_db_bp + '.dmnd'
return diamond_db_fp
def run_diamond(aa_fp, db_fp, cpu: int, diamond_out_fnn):
# More sensitive as an option?
diamond_cmd = ['diamond', 'blastp', '--threads', str(cpu), '--sensitive', '-d', db_fp, '-q', aa_fp,
'-o', diamond_out_fnn]
print("Running Diamond...")
res = subprocess.run(diamond_cmd, check=True, stdout=subprocess.PIPE)
if res.returncode != 0:
print('Error running Diamond')
exit(1)
return diamond_out_fnn
def make_protein_clusters_mcl(abc_fp, out_p, inflation=2):
print("Running MCL...")
abc_fn = "merged"
mci_fn = '{}.mci'.format(abc_fn)
mci_fp = os.path.join(out_p, mci_fn)
mcxload_fn = '{}_mcxload.tab'.format(abc_fn)
mcxload_fp = os.path.join(out_p, mcxload_fn)
subprocess.check_call("mcxload -abc {0} --stream-mirror --stream-neg-log10 -stream-tf 'ceil(200)' -o {1}"
" -write-tab {2}".format(abc_fp, mci_fp, mcxload_fp), shell=True)
mcl_clstr_fn = "{0}_mcl{1}.clusters".format(abc_fn, int(inflation*10))
mcl_clstr_fp = os.path.join(out_p, mcl_clstr_fn)
subprocess.check_call("mcl {0} -I {1} -use-tab {2} -o {3}".format(
mci_fp, inflation, mcxload_fp, mcl_clstr_fp), shell=True)
return mcl_clstr_fp
def load_mcl_clusters(fi):
with open(fi) as f:
c = [line.rstrip("\n").split("\t") for line in f]
c = [x for x in c if len(c) > 1]
nb_clusters = len(c)
formatter = "PC_{{:>0{}}}".format(int(round(np.log10(nb_clusters))+1))
name = [formatter.format(str(i)) for i in range(nb_clusters)]
size = [len(i) for i in c]
clusters_df = pd.DataFrame({"size": size, "pc_id": name}).set_index("pc_id")
return clusters_df, name, c
def build_clusters(fp, gene2genome):
# Read MCL
clusters_df, name, c = load_mcl_clusters(fp)
print("Using MCL to generate PCs.")
# Assign each prot to its cluster
gene2genome.set_index("protein_id", inplace=True) # id, contig, keywords, cluster
for prots, clust in zip(c, name):
try:
gene2genome.loc[prots, "cluster"] = clust
except KeyError:
prots_in = [p for p in prots if p in gene2genome.index]
not_in = frozenset(prots) - frozenset(prots_in)
print("{} protein(s) without contig: {}".format(len(not_in), not_in))
gene2genome.loc[prots_in, "cluster"] = clust
# Keys
for clust, prots in gene2genome.groupby("cluster"):
clusters_df.loc[clust, "annotated"] = prots.keywords.count()
if prots.keywords.count():
keys = ";".join(prots.keywords.dropna().values).split(";")
key_count = {}
for k in keys:
k = k.strip()
try:
key_count[k] += 1
except KeyError:
key_count[k] = 1
clusters_df.loc[clust, "keys"] = "; ".join(["{} ({})".format(x, y) for x, y in key_count.items()])
gene2genome.reset_index(inplace=True)
clusters_df.reset_index(inplace=True)
profiles_df = gene2genome.loc[:, ["contig_id", "cluster"]].drop_duplicates()
profiles_df.columns = ["contig_id", "pc_id"]
contigs_df = pd.DataFrame(gene2genome.fillna(0).groupby("contig_id").count().protein_id)
contigs_df.index.name = "contig_id"
contigs_df.columns = ["proteins"]
contigs_df.reset_index(inplace=True)
return gene2genome, clusters_df, profiles_df, contigs_df
# Create P-edges
def build_pc_matrices(profiles, contigs, pcs):
pc_by_cont = profiles.groupby("contig_id").count().pc_id
pc_by_cont = pd.merge(contigs.sort_values("pos").loc[:, ["pos", "contig_id", "proteins"]], pc_by_cont.to_frame(), how="left",
left_on="contig_id", right_on="contig_id").fillna(0)
singletons = (pc_by_cont.proteins - pc_by_cont.pc_id).values
singletons = sparse.lil_matrix(singletons).transpose()
# Matrix
profiles.index.name = "pos"
profiles.reset_index(inplace=True)
profiles = pd.merge(profiles, pcs.loc[:, ["pc_id", "pos"]], left_on="pc_id", right_on="pc_id", how="inner",
suffixes=["", "_pc"]) # pos, contig_id, pc_id, id (pc), pos_pc
profiles = pd.merge(profiles, contigs.loc[:, ["contig_id", "pos"]], left_on="contig_id", right_on="contig_id", how="inner",
suffixes=["", "_contig"])
profiles = profiles.loc[:, ["pos_contig", "pos_pc"]]
matrix = sparse.coo_matrix(([1]*len(profiles), (zip(*profiles.values))), shape=(len(contigs), len(pcs)),
dtype="bool")
return matrix.tocsr(), singletons.tocsr()
def to_clusterer(matrix, fi, contigs=None,names=None):
names = contigs if names is None else names
names = names.set_index("pos").contig_id
with open(fi, "wt") as f:
matrix = sparse.dok_matrix(matrix)
for r, c in zip(*matrix.nonzero()):
f.write(" ".join([str(x) for x in (names[r], names[c], matrix[r, c])]))
f.write("\n")
print("Saving network in file {0} ({1} lines).".format(fi, matrix.getnnz()))
return fi
def create_network(matrix, singletons, thres=1, max_sig=1000):
contigs, pcs = matrix.shape
pcs += singletons.sum()
# Number of comparisons
T = 0.5 * contigs * (contigs - 1)
logT = np.log10(T)
# Number of protein clusters in each contig
# = # shared pcs + #singletons
number_of_pc = matrix.sum(1) + singletons
number_of_pc = number_of_pc.A1 # Transform into a flat array
# Number of common protein clusters between two contigs, tuple + commons
commons_pc = matrix.dot(sparse.csr_matrix(matrix.transpose(), dtype=int))
S = sparse.lil_matrix((contigs, contigs))
total_c = float(commons_pc.getnnz())
i = 0 # Display
for A, B in zip(*commons_pc.nonzero()): # For A & B sharing contigs
if A != B:
a, b = sorted([number_of_pc[A], number_of_pc[B]])
pval = stats.hypergeom.sf(commons_pc[A, B] - 1, pcs, a, b)
sig = min(max_sig, np.nan_to_num(-np.log10(pval) - logT))
if sig > thres:
S[min(A, B), max(A, B)] = sig
# Display
i += 1
if i % 1000 == 0:
sys.stdout.write(".")
if i % 10000 == 0:
sys.stdout.write("{:6.2%} {}/{}\n".format(i / total_c, i, total_c))
S += S.T
S = S.tocsr()
if len(S.data) != 0:
print("Hypergeometric contig-similarity network:\n {0:10} contigs,\n {1:10} edges (min:{2:.2}"
"max: {3:.2}, threshold was {4})".format(contigs, S.getnnz(), S.data.min(), S.data.max(), thres))
else:
raise ValueError("No edge in the similarity network !")
return S
def masked_ECE(out, train_label, train_mask):
device = torch.device('cuda')
m = nn.Softmax(dim=1)
prob_box = dict.fromkeys([0, 1, 2, 3, 4])
pred_box = dict.fromkeys([0, 1, 2, 3, 4])
label_box = dict.fromkeys([0, 1, 2, 3, 4])
# calculate pred
prob = torch.max(m(out), 1)[0]
pred = torch.argmax(out, 1)
# calculate box
for i in range(len(prob)):
if test_mask[i] == True:
if prob[i] > 0.9:
try:
prob_box[9].append(prob[i])
pred_box[9].append(pred[i])
label_box[9].append(test_label[i])
except:
prob_box[9] = [prob[i]]
pred_box[9] = [pred[i]]
label_box[9]= [test_label[i]]
elif prob[i] > 0.8:
try:
prob_box[8].append(prob[i])
pred_box[8].append(pred[i])
label_box[8].append(test_label[i])
except:
prob_box[8] = [prob[i]]
pred_box[8] = [pred[i]]
label_box[8]= [test_label[i]]
elif prob[i] > 0.7:
try:
prob_box[7].append(prob[i])
pred_box[7].append(pred[i])
label_box[7].append(test_label[i])
except:
prob_box[7] = [prob[i]]
pred_box[7] = [pred[i]]
label_box[7]= [test_label[i]]
elif prob[i] > 0.6:
try:
prob_box[6].append(prob[i])
pred_box[6].append(pred[i])
label_box[6].append(test_label[i])
except:
prob_box[6] = [prob[i]]
pred_box[6] = [pred[i]]
label_box[6]= [test_label[i]]
elif prob[i] > 0.5:
try:
prob_box[5].append(prob[i])
pred_box[5].append(pred[i])
label_box[5].append(test_label[i])
except:
prob_box[5] = [prob[i]]
pred_box[5] = [pred[i]]
label_box[5]= [test_label[i]]
elif prob[i] > 0.4:
try:
prob_box[4].append(prob[i])
pred_box[4].append(pred[i])
label_box[4].append(test_label[i])
except:
prob_box[4] = [prob[i]]
pred_box[4] = [pred[i]]
label_box[4]= [test_label[i]]
elif prob[i] > 0.3:
try:
prob_box[3].append(prob[i])
pred_box[3].append(pred[i])
label_box[3].append(test_label[i])
except:
prob_box[3] = [prob[i]]
pred_box[3] = [pred[i]]
label_box[3]= [test_label[i]]
elif prob[i] > 0.2:
try:
prob_box[2].append(prob[i])
pred_box[2].append(pred[i])
label_box[2].append(test_label[i])
except:
prob_box[2] = [prob[i]]
pred_box[2] = [pred[i]]
label_box[2]= [test_label[i]]
elif prob[i] > 0.1:
try:
prob_box[1].append(prob[i])
pred_box[1].append(pred[i])
label_box[1].append(test_label[i])
except:
prob_box[1] = [prob[i]]
pred_box[1] = [pred[i]]
label_box[1]= [test_label[i]]
else:
try:
prob_box[0].append(prob[i])
pred_box[0].append(pred[i])
label_box[0].append(test_label[i])
except:
prob_box[0] = [prob[i]]
pred_box[0] = [pred[i]]
label_box[0]= [test_label[i]]
# calculate ECE
ECE = 0
for key in label_box.keys():
if label_box[key] == None:
continue
accuarcy = torch.sum(torch.FloatTensor(label_box[key]) == torch.FloatTensor(pred_box[key]))/len(pred_box[key])
confidence = torch.mean(torch.FloatTensor(prob_box[key]))
gap = len(pred_box[key])*torch.abs(accuarcy-confidence)
ECE+=gap
return ECE.to(device)/len(out)
def masked_loss(out, label, mask):
#if torch.cuda.is_available():
# w = torch.Tensor([3.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 1.0]).cuda()
#else:
# w = torch.Tensor([3.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 1.0])
#loss = F.cross_entropy(out, label, w, reduction='none')
loss = F.cross_entropy(out, label, reduction='none')
#all phage
#w = torch.Tensor([3.0, 2.0, 2.0, 2.0, 2.0, 2.0, 2.0, 1.0, 3.0, 2.0, 3.0]).cuda()
#loss = F.cross_entropy(out, label, w, reduction='none')
mask = mask.float()
mask = mask / mask.mean()
loss *= mask
loss = loss.mean()
return loss
def masked_acc(out, label, mask):
# [node, f]
pred = out.argmax(dim=1)
correct = torch.eq(pred, label).float()
mask = mask.float()
mask = mask / mask.mean()
correct *= mask
acc = correct.mean()
return acc
def sparse_dropout(x, rate, noise_shape):
"""
:param x:
:param rate:
:param noise_shape: int scalar
:return:
"""
random_tensor = 1 - rate
random_tensor += torch.rand(noise_shape).to(x.device)
dropout_mask = torch.floor(random_tensor).byte()
i = x._indices() # [2, 49216]
v = x._values() # [49216]
# [2, 4926] => [49216, 2] => [remained node, 2] => [2, remained node]
i = i[:, dropout_mask]
v = v[dropout_mask]
out = torch.sparse.FloatTensor(i, v, x.shape).to(x.device)
out = out * (1./ (1-rate))
return out
def dot(x, y, sparse=False):
if sparse:
res = torch.sparse.mm(x, y)
else:
res = torch.mm(x, y)
return res