-
Notifications
You must be signed in to change notification settings - Fork 17
/
Copy pathutils.py
1780 lines (1621 loc) · 60.3 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
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
from anndata import read_h5ad
import sys
from time import time
from scipy import stats, sparse
import numpy as np
import collections
import pickle
from sklearn.preprocessing import normalize
import os
from collections import Counter
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score,accuracy_score,precision_recall_fscore_support, cohen_kappa_score, auc, average_precision_score,f1_score,precision_recall_curve
import time
import umap
import copy
from sklearn import preprocessing
from fbpca import pca
from sklearn.metrics import roc_auc_score, roc_curve
from sklearn.metrics.pairwise import cosine_similarity
from scanorama import VERBOSE, KNN, ALPHA, APPROX, SIGMA
#from libs import *
from scanorama import find_alignments,merge_datasets,process_data,transform,vstack
from scipy.sparse.csgraph import shortest_path as graph_shortest_path
from scipy.sparse.linalg import svds, eigs
nn_nhidden = [1000]
rsts = [0.5,0.6,0.7,0.8]
dfs_depth = 1
co_dim = 5
keep_prob = 1.0
use_diagonal = True
max_iter = 20
niter = 5
def translate_paramter(ps):
s = []
for p in ps:
if isinstance(p, list):
p = [str(i) for i in p]
p = '.'.join(p)
s.append(p)
else:
s.append(str(p))
s = '_'.join(s)
return s
pname = translate_paramter([max_iter])
def make_folder(folder):
if not os.path.exists(folder):
os.makedirs(folder)
return folder
def create_propagate_networks(dname, l2i, onto_net, cls2cls, ontology_nlp_file, rsts = [0.5,0.6,0.7,0.8], diss=[2,3], thress=[1,0.8]):
ncls = np.shape(cls2cls)[0]
if dname != 'allen':
onto_net_nlp, onto_net_bin, stack_net_nlp, stack_net_bin, onto_net_nlp_all_pairs = create_nlp_networks(l2i, onto_net, cls2cls, ontology_nlp_file)
#network = create_consensus_networks(rsts, stack_net_nlp, onto_net_nlp_all_pairs, cls2cls)
network = create_consensus_networks(rsts, stack_net_nlp, onto_net_nlp_all_pairs, cls2cls, diss = diss, thress = thress)
else:
stack_net_bin = np.zeros((ncls,ncls))
for n1 in onto_net:
for n2 in onto_net[n1]:
if n1==n2:
continue
stack_net_bin[n1,n2] = 1
stack_net_bin[n2,n1] = 1
network = [RandomWalkRestart(stack_net_bin, rst) for rst in rsts]
return network
def fine_nearest_co_using_nlp(sentences,co2emb,obo_file,nlp_mapping_cutoff=0.8):
co2name, name2co = get_ontology_name(obo_file = obo_file)
from sentence_transformers import SentenceTransformer
model = SentenceTransformer('bert-base-nli-mean-tokens')
sentences = np.array([sentence.lower() for sentence in sentences])
sentence_embeddings = model.encode(sentences)
co_embeddings = []
cos = []
for co in co2emb:
co_embeddings.append(co2emb[co])
cos.append(co)
co_embeddings = np.array(co_embeddings)
sent2co = {}
for sentence, embedding, ind in zip(sentences, sentence_embeddings, range(len(sentences))):
scs = cosine_similarity(co_embeddings, embedding.reshape(1,-1))
co_id = np.argmax(scs)
sc = scs[co_id]
if sc>nlp_mapping_cutoff:
sent2co[sentence.lower()] = cos[co_id]
names = set()
for name in name2co:
if name2co[name].upper() == cos[co_id]:
names.add(name)
#print (sentence, cos[co_id], sc, co2name[cos[co_id]],names)
return sent2co
def ImputeUnseenCls(y_vec, y_raw, cls2cls, nseen, knn=1):
nclass = np.shape(cls2cls)[0]
seen2unseen_sim = cls2cls[:nseen, nseen:]
nngh = np.argsort(seen2unseen_sim*-1, axis = 0)[0,:]
ncell = len(y_vec)
y_mat = np.zeros((ncell, nclass))
y_mat[:,:nseen] = y_raw[:, :nseen]
for i in range(ncell):
if y_vec[i] == -1:
#kngh = np.argsort(y_raw[i,:nseen]*-1)[0:knn]
#if len(kngh) == 0:
# continue
y_mat[i,nseen:] = y_mat[i,nngh]
y_mat[i,:nseen] -= 1000000
return y_mat
def ImputeUnseenCls_Backup(y_vec, y_raw, cls2cls, nseen, knn=1):
nclass = np.shape(cls2cls)[0]
seen2unseen_sim = cls2cls[:nseen, nseen:]
ncell = len(y_vec)
y_mat = np.zeros((ncell, nclass))
y_mat[:,:nseen] = y_raw[:, :nseen]
for i in range(ncell):
if y_vec[i] == -1:
kngh = np.argsort(y_raw[i,:nseen]*-1)[0:knn]
if len(kngh) == 0:
continue
y_mat[i,:nseen] -= 1000000
y_mat[i,nseen:] = np.dot(y_raw[i,kngh], seen2unseen_sim[kngh,:])
return y_mat
def find_gene_ind(genes, common_genes):
gid = []
for g in common_genes:
gid.append(np.where(genes == g)[0][0])
gid = np.array(gid)
return gid
def RandomWalkOntology(onto_net, l2i, ontology_nlp_file, ontology_nlp_emb_file, rst = 0.7):
ncls = len(l2i)
onto_net_nlp, _, onto_nlp_emb = read_cell_ontology_nlp(l2i, ontology_nlp_file, ontology_nlp_emb_file)
onto_net_nlp = (cosine_similarity(onto_nlp_emb) + 1 ) /2#1 - spatial.distance.cosine(onto_nlp_emb, onto_nlp_emb)
onto_net_mat = np.zeros((ncls, ncls))
for n1 in onto_net:
for n2 in onto_net[n1]:
if n1==n2:
continue
onto_net_mat[n1,n2] = onto_net_nlp[n1, n2]
onto_net_mat[n2,n1] = onto_net_nlp[n2, n1]
onto_net_rwr = RandomWalkRestart(onto_net_mat, rst)
return onto_net_rwr
def process_expression(c2g_list):
#this data process function is motivated by ACTINN, please check ACTINN for more information.
c2g = np.vstack(c2g_list)
c2g = c2g.T
#print ('onclass d0',np.shape(c2g))
c2g = c2g[np.sum(c2g, axis=1)>0, :]
#print (c2g)
#print ('onclass d1',np.shape(c2g))
c2g = np.divide(c2g, np.sum(c2g, axis=0, keepdims=True)) * 10000
c2g = np.log2(c2g+1)
expr = np.sum(c2g, axis=1)
#total_set = total_set[np.logical_and(expr >= np.percentile(expr, 1), expr <= np.percentile(expr, 99)),]
c2g = c2g[np.logical_and(expr >= np.percentile(expr, 1), expr <= np.percentile(expr, 99)),]
#print (c2g)
#print ('onclass d2',np.shape(c2g))
cv = np.std(c2g, axis=1) / np.mean(c2g, axis=1)
c2g = c2g[np.logical_and(cv >= np.percentile(cv, 1), cv <= np.percentile(cv, 99)),]
#print (c2g)
#print ('onclass d3',np.shape(c2g))
c2g = c2g.T
#print (c2g)
#print ('onclass d4',np.shape(c2g))
c2g_list_new = []
index = 0
for c in c2g_list:
ncell = np.shape(c)[0]
c2g_list_new.append(c2g[index:index+ncell,:])
index = ncell
return c2g_list_new
def read_ontology_file(dname, data_folder):
if 'allen' in dname:
cell_type_network_file = data_folder + 'allen.ontology'
cell_type_nlp_emb_file = None
cl_obo_file = None
if not os.path.isfile(cell_type_network_file):
sys.error(cell_type_network_file + ' not found!')
else:
cell_type_network_file = data_folder + 'cl.ontology'
cell_type_nlp_emb_file = data_folder + 'cl.ontology.nlp.emb'
cl_obo_file = data_folder + 'cl.obo'
if not os.path.isfile(cell_type_nlp_emb_file):
sys.exit(cell_type_nlp_emb_file + ' not found!')
if not os.path.isfile(cell_type_network_file):
sys.exit(cell_type_network_file + ' not found!')
if not os.path.isfile(cl_obo_file):
sys.exit(cl_obo_file + ' not found!')
return cell_type_nlp_emb_file, cell_type_network_file, cl_obo_file
def read_data_file(dname, data_dir):
if 'microcebus' in dname:
tech = '10x'
feature_file = data_dir + 'Lemur/' + dname +'.h5ad'
filter_key={'method':tech }
label_file = None
gene_file = ''
label_key = 'cell_ontology_class'
elif 'muris' in dname:
tech = dname.split('_')[1]
feature_file = data_dir + 'Tabula_Muris_Senis/' + 'tabula-muris-senis-'+tech+'-official-raw-obj.h5ad'
filter_key = {}
label_file = None
gene_file = ''
batch_key = ''
label_key = 'cell_ontology_class'
elif 'sapiens' in dname:
feature_file = data_dir + 'sapiens/' + 'Pilot1_Pilot2_decontX_Oct2020.h5ad'
filter_key = {}
label_file = None
gene_file = ''
batch_key = ''
label_key = 'cell_ontology_type'
elif 'allen' in dname:
feature_file = data_dir + '/Allen_Brain/features.pkl'
label_file = data_dir + '/Allen_Brain/labels.pkl'
gene_file = data_dir + '/Allen_Brain/genes.pkl'
label_key = ''
filter_key = {}
elif 'krasnow' in dname:
tech = dname.split('_')[1]
feature_file = data_dir + '/HLCA/'+tech+'_features.pkl'
label_file = data_dir + '/HLCA/'+tech+'_labels.pkl'
gene_file = data_dir + '/HLCA/'+tech+'_genes.pkl'
label_key = ''
filter_key = {}
else:
sys.exit('wrong dname '+dname)
if feature_file.endswith('.pkl'):
return feature_file, filter_key, label_key, label_file, gene_file
elif feature_file.endswith('.h5ad'):
return feature_file, filter_key, label_key, label_file, gene_file
sys.exit('wrong file suffix')
def read_singlecell_data(dname, data_dir, ontology_dir, nsample = 500000000, read_tissue = False, exclude_non_leaf_ontology = True):
if 'microcebus' in dname:
tech = '10x'
#file = data_dir + 'TMS_official_060520/' + 'tabula-microcebus_smartseq2-10x_combined_annotated_filtered_gene-labels-correct.h5ad'
file = data_dir + 'TMS_official_060520/' + dname +'.h5ad'
filter_key={'method':tech }
batch_key = ''#original_channel
ontology_nlp_file = ontology_dir + '/cell_ontology/cl.ontology.nlp'
ontology_file = ontology_dir + '/cell_ontology/cl.ontology'
cl_obo_file = ontology_dir + '/cell_ontology/cl.obo'
if not read_tissue:
feature, label, genes = parse_h5ad(file, nsample = nsample, read_tissue = read_tissue, label_key='cell_ontology_class', batch_key = batch_key, filter_key = filter_key, cell_ontology_file = ontology_file, exclude_non_leaf_ontology = exclude_non_leaf_ontology, exclude_non_ontology = True, cl_obo_file = cl_obo_file)
else:
feature, label, genes, tissues = parse_h5ad(file, nsample = nsample, read_tissue = read_tissue, label_key='cell_ontology_class', batch_key = batch_key, filter_key = filter_key, cell_ontology_file = ontology_file, exclude_non_leaf_ontology = exclude_non_leaf_ontology, exclude_non_ontology = True, cl_obo_file = cl_obo_file)
elif 'muris' in dname:
tech = dname.split('_')[1]
file = data_dir + 'TMS_official_060520/' + 'tabula-muris-senis-'+tech+'-official-raw-obj.h5ad'
filter_key = {}
batch_key = ''
ontology_nlp_file = ontology_dir + '/cell_ontology/cl.ontology.nlp'
ontology_file = ontology_dir + '/cell_ontology/cl.ontology'
cl_obo_file = ontology_dir + '/cell_ontology/cl.obo'
if not read_tissue:
feature, label, genes = parse_h5ad(file, nsample = nsample, read_tissue = read_tissue, label_key='cell_ontology_class', batch_key = batch_key, cell_ontology_file = ontology_file, filter_key=filter_key, exclude_non_leaf_ontology = exclude_non_leaf_ontology, exclude_non_ontology = True, cl_obo_file = cl_obo_file)
else:
feature, label, genes, tissues = parse_h5ad(file, nsample = nsample, read_tissue = read_tissue, label_key='cell_ontology_class', batch_key = batch_key, cell_ontology_file = ontology_file, filter_key=filter_key, exclude_non_leaf_ontology = exclude_non_leaf_ontology, exclude_non_ontology = True, cl_obo_file = cl_obo_file)
elif 'allen_part' in dname:
feature_file = data_dir + 'Allen/matrix_part.csv'
label_file = data_dir + 'Allen/metadata.csv'
ontology_file = data_dir + 'Allen/cell_type_ontology'
ontology_nlp_file = None
feature, label, genes = parse_csv(feature_file, label_file, nsample = nsample, label_key='cell_type_accession_label', exclude_non_ontology = True, exclude_non_leaf_ontology = True, cell_ontology_file=ontology_file)
elif 'allen' in dname:
feature_file = data_dir + 'Allen/features.pkl'
label_file = data_dir + 'Allen/labels.pkl'
gene_file = data_dir + 'Allen/genes.pkl'
ontology_file = data_dir + 'Allen/cell_type_ontology'
ontology_nlp_file = None
feature, label, genes = parse_pkl(feature_file, label_file, gene_file, nsample = nsample, exclude_non_leaf_ontology = True, cell_ontology_file=ontology_file)
elif 'krasnow' in dname:
tech = dname.split('_')[1]
feature_file = data_dir + 'Krasnow/'+tech+'_features.pkl'
label_file = data_dir + 'Krasnow/'+tech+'_labels.pkl'
gene_file = data_dir + 'Krasnow/'+tech+'_genes.pkl'
ontology_file = ontology_dir + '/cell_ontology/cl.ontology'
ontology_nlp_file = ontology_dir + '/cell_ontology/cl.ontology.nlp'
cl_obo_file = ontology_dir + '/cell_ontology/cl.obo'
feature, label, genes = parse_pkl(feature_file, label_file, gene_file, nsample = nsample, exclude_non_leaf_ontology = True, cell_ontology_file=ontology_file)
else:
sys.exit('wrong dname '+dname)
if read_tissue:
return feature, label, genes, tissues, ontology_nlp_file, ontology_file
else:
return feature, label, genes, ontology_nlp_file, ontology_file
def parse_krasnow(feature_file, label_file, gene_file, seed = 1, nsample = 1000,exclude_non_leaf_ontology = True, exclude_non_ontology = True, cell_ontology_file=None):
np.random.seed(seed)
if feature_file.endswith('.pkl'):
features = pickle.load(open(feature_file, 'rb'))
labels = pickle.load(open(label_file, 'rb'))
genes = pickle.load(open(gene_file, 'rb'))
ncell, ngene = np.shape(features)
assert(ncell == len(labels))
assert(ngene == len(genes))
index = np.random.choice(ncell,min(nsample,ncell),replace=False)
features = features[index, :]
labels = labels[index]
if exclude_non_leaf_ontology:
new_ids, exclude_terms = exclude_parent_child_nodes(cell_ontology_file, labels)
#print (len(exclude_terms),'non leaf terms are excluded')
features = features[new_ids, :]
labels = labels[new_ids]
genes = [x.upper() for x in genes]
genes = np.array(genes)
return features, labels, genes
def parse_pkl(feature_file, label_file, gene_file, seed = 1, nsample = 10000000,exclude_non_leaf_ontology = True, cell_ontology_file=None):
np.random.seed(seed)
if feature_file.endswith('.pkl'):
features = pickle.load(open(feature_file, 'rb'))
labels = pickle.load(open(label_file, 'rb'))
genes = pickle.load(open(gene_file, 'rb'))
ncell, ngene = np.shape(features)
assert(ncell == len(labels))
assert(ngene == len(genes))
index = np.random.choice(ncell,ncell,replace=False)
features = features[index, :]
labels = labels[index]
if exclude_non_leaf_ontology:
new_ids, exclude_terms = exclude_parent_child_nodes(cell_ontology_file, labels)
#print (len(exclude_terms),'non leaf terms are excluded')
features = features[new_ids, :]
labels = labels[new_ids]
genes = [x.upper() for x in genes]
genes = np.array(genes)
return features, labels, genes
def select_high_var_genes(train_X, test_X, ngene = 200):
mat = np.vstack((train_X, test_X))
#mat = mat.todense()
gstd = np.std(mat, axis=0)
best_genes = np.argsort(gstd*-1)
best_genes = best_genes[:ngene]
return train_X[:, best_genes], test_X[:, best_genes]
def emb_cells(train_X, test_X, dim=20):
if dim==-1:
return np.log1p(train_X.todense()), np.log1p(test_X.todense())
train_X = np.log1p(train_X)
test_X = np.log1p(test_X)
train_X = preprocessing.normalize(train_X, axis=1)
test_X = preprocessing.normalize(test_X, axis=1)
ntrain = np.shape(train_X)[0]
mat = sparse.vstack((train_X, test_X))
U, s, Vt = pca(mat, k=dim) # Automatically centers.
X = U[:, range(dim)] * s[range(dim)]
return X[:ntrain,:], X[ntrain:,:]
def write_markers(fname, markers):
## Write marker genes to file
fmarker_genes = open(fname,'w')
for t in markers:
fmarker_genes.write(t+'\t')
g2pv = sorted(markers[t].items(), key=lambda item: item[1])
for g,pv in g2pv:
fmarker_genes.write(g+'(pv:'+'{:.2e}'.format(pv)+')\t')
fmarker_genes.write('\n')
fmarker_genes.close()
def calculate_markers(cell2term, cell2gene, genes, terms, topk_cells=500, only_over_expressed = True, return_k_genes = 100):
ncell, nterm = np.shape(cell2term)
ngene = np.shape(cell2gene)[1]
assert(ncell == np.shape(cell2gene)[0])
markers = collections.defaultdict(dict)
for t in range(nterm):
scs = np.argsort(cell2term[:,t])
k_bot_cells = scs[:topk_cells]
k_top_cells = scs[ncell-topk_cells:]
pv = scipy.stats.ttest_ind(cell2gene[k_top_cells,:], cell2gene[k_bot_cells,:], axis=0)[1] #* ngene
top_mean = np.mean(cell2gene[k_top_cells,:],axis=0)
bot_mean = np.mean(cell2gene[k_bot_cells,:],axis=0)
if only_over_expressed:
for g in range(ngene):
if top_mean[g] < bot_mean[g]:
pv[g] = 1.
pv_sort = list(np.argsort(pv))
#for i in range(return_k_genes):
#markers[terms[t]][genes[pv_sort[i]]] = pv[pv_sort[i]]
markers[terms[t]] = pv
for i,p in enumerate(pv):
if np.isnan(p):
pv[i] = 1.
#markers[terms[t]][str(pv_sort[i])] = pv[pv_sort[i]]
return markers
def peak_h5ad(file):
'''
peak the number of cells, classes, genes in h5ad file
'''
x = read_h5ad(file)
#print (np.shape(x.X))
#print (x.X[:10][:10])
#print (x.obs.keys())
ncell, ngene = np.shape(x.X)
nclass = len(np.unique(x.obs['free_annotation']))
#print (np.unique(x.obs['free_annotation']))
f2name = {}
sel_cell = 0.
for i in range(ncell):
if x.obs['method'][i]!='10x':
continue
free = x.obs['free_annotation'][i]
name = x.obs['cell_ontology_class'][i]
f2name[free] = name
sel_cell += 1
#return f2name
#for key in x.obs.keys():
# print (key, np.unique(x.obs[key]))
return sel_cell, ngene, nclass
#for i in range(10):
# print (x.obs['method'][i], x.obs['channel_no_10x'][i])
#for key in x.obs.keys():
# print (key, np.unique(x.obs[key]))
#return index
def get_onotlogy_parents(GO_net, g):
term_valid = set()
ngh_GO = set()
ngh_GO.add(g)
while len(ngh_GO) > 0:
for GO in list(ngh_GO):
for GO1 in GO_net[GO]:
ngh_GO.add(GO1)
ngh_GO.remove(GO)
term_valid.add(GO)
return term_valid
def exclude_non_ontology_term(cl_obo_file, labels, label_key):
co2name, name2co = get_ontology_name(cl_obo_file)
new_labs = []
new_ids = []
if label_key!='cell_ontology_class' and label_key!='cell_ontology_id':
use_co = False
for kk in np.unique(labels):
if kk.lower().startswith('cl:'):
use_co = True
break
else:
if label_key == 'cell_ontology_class':
use_co = False
else:
use_co = True
for i in range(len(labels)):
l = labels[i]
if not use_co:
if l.lower() in name2co.keys():
new_labs.append(name2co[l.lower()])
new_ids.append(i)
else:
if l.lower() in co2name.keys():
new_labs.append(l.lower())
new_ids.append(i)
new_labs = np.array(new_labs)
new_ids = np.array(new_ids)
return new_ids, new_labs
def parse_raw_h5ad(file,seed=1,nsample=1e10,tissue_key='tissue',label_key='cell_ontology_class', read_tissue = True, batch_key = '', filter_key={}, cell_ontology_file = None, exclude_non_leaf_ontology = True, exclude_non_ontology=True, cl_obo_file = None):
np.random.seed(seed)
x = read_h5ad(file)
ncell = np.shape(x.raw.X)[0]
select_cells = set(range(ncell))
for key in filter_key:
value = filter_key[key]
select_cells = select_cells & set(np.where(np.array(x.obs[key])==value)[0])
select_cells = sorted(select_cells)
feature = x.raw.X[select_cells, :]
labels = np.array(x.obs[label_key].tolist())[select_cells]
if read_tissue:
tissues = np.array(x.obs[tissue_key].tolist())[select_cells]
if batch_key=='' or batch_key not in x.obs.keys():
batch_labels = np.ones(len(labels))
else:
batch_labels = np.array(x.obs[batch_key].tolist())[select_cells]
genes = x.var.index
ncell = len(select_cells)
if exclude_non_ontology:
new_ids, labels = exclude_non_ontology_term(cl_obo_file, labels, label_key)
feature = feature[new_ids, :]
batch_labels = batch_labels[new_ids]
if exclude_non_leaf_ontology:
new_ids, exclude_terms = exclude_parent_child_nodes(cell_ontology_file, labels)
#print (len(exclude_terms),'non leaf terms are excluded')
feature = feature[new_ids, :]
batch_labels = batch_labels[new_ids]
labels = labels[new_ids]
if read_tissue:
tissues = tissues[new_ids]
ncell = len(labels)
index = np.random.choice(ncell,min(nsample,ncell),replace=False)
batch_labels = batch_labels[index]
feature = feature[index, :] # cell by gene matrix
labels = labels[index]
if read_tissue:
tissues = tissues[index]
genes = x.var.index
corrected_feature = run_scanorama_same_genes(feature, batch_labels)
corrected_feature = corrected_feature.toarray()
genes = [x.upper() for x in genes]
genes = np.array(genes)
if read_tissue:
assert(len(tissues) == len(labels))
return corrected_feature, labels, genes, tissues
else:
return corrected_feature, labels, genes
def select_cells_based_on_keys(x, features, tissues = None, labels = None, filter_key = None):
ncell = np.shape(x.X)[0]
select_cells = set(range(ncell))
for key in filter_key:
value = filter_key[key]
select_cells = select_cells & set(np.where(np.array(x.obs[key])==value)[0])
select_cells = sorted(select_cells)
features = features[select_cells,: ]
if labels is not None:
labels = labels[select_cells]
if tissues is not None:
tissues = tissues[select_cells]
x = x[select_cells,:]
return features, labels, tissues, x
def find_marker_genes(train_X, pred_Y_all, genes, i2l, topk = 50):
cor = corr2_coeff(pred_Y_all[:,:].T, train_X[:,:].T)
cor = np.nan_to_num(cor) # cell type to gene
nl = len(i2l)
c2g = {}
for i in range(nl):
gl = np.argsort(cor[i,:]*-1)
c2g[i2l[i]] = {}
for j in range(topk):
c2g[i2l[i]][genes[gl[j]]] = cor[i, gl[j]]
return c2g, cor
def use_pretrained_model(OnClass, genes, test_X, models = []):
last_l2i = {}
last_i2l = {}
pred_Y_all_models = 0.
ngene = len(genes)
for model in models:
OnClass.BuildModel(OnClass.co2emb, ngene = ngene, use_pretrain = model)
print ('Build model finished for ',model)
pred_Y_seen, pred_Y_all, pred_label = OnClass.Predict(test_X, test_genes = genes)
print ('Predict for ',model)
pred_Y_all = pred_Y_all.T / (pred_Y_all.T.sum(axis=1)[:, np.newaxis] + 1)
pred_Y_all = pred_Y_all.T
if len(last_l2i)>0:
new_ct_ind = []
for i in range(len(last_i2l)):
l = last_i2l[i]
new_ct_ind.append(OnClass.co2i[l])
pred_Y_all = pred_Y_all[:, np.array(new_ct_ind)]
pred_Y_all_models += pred_Y_all
else:
last_l2i = OnClass.co2i
last_i2l = OnClass.i2co
pred_Y_all_models = pred_Y_all
return pred_Y_all_models
def read_data(feature_file, cell_ontology_ids, exclude_non_leaf_ontology = False, ct_mapping_key = {}, tissue_key = None, seed = 1, filter_key = None, AnnData_label_key=None, nlp_mapping = True, nlp_mapping_cutoff = 0.8, co2emb = None, label_file=None, cl_obo_file = None, cell_ontology_file = None):
np.random.seed(seed)
x = read_h5ad(feature_file)
ncell = np.shape(x.X)[0]
dataset = x.X.toarray()
genes = np.array([x.upper() for x in x.var.index])
if tissue_key is not None:
tissues = np.array(x.obs[tissue_key].tolist())
else:
tissues = None
if AnnData_label_key is None and label_file is None:
print ('no label file is provided')
labels = None
dataset, labels, tissues, x = select_cells_based_on_keys(x, dataset, labels = labels, tissues = tissues, filter_key = filter_key)
return dataset, genes, labels, tissues, x
if AnnData_label_key is not None:
labels = x.obs[AnnData_label_key].tolist()
else:
fin = open(label_file)
labels = []
for line in fin:
labels.append(line.strip())
fin.close()
labels = np.array(labels)
dataset, labels, tissues, x = select_cells_based_on_keys(x, dataset, labels = labels, tissues = tissues, filter_key = filter_key)
ind, labels, unfound_labs = map_and_select_labels(labels, cell_ontology_ids, cl_obo_file, ct_mapping_key = ct_mapping_key, nlp_mapping = nlp_mapping, co2emb = co2emb, nlp_mapping_cutoff = nlp_mapping_cutoff, cl_obo_file = cl_obo_file)
if tissue_key is not None:
tissues = tissues[ind]
dataset = dataset[ind, :]
x = x[ind, :]
if exclude_non_leaf_ontology:
new_ids, exclude_terms = exclude_parent_child_nodes(cell_ontology_file, labels)
tissues = tissues[new_ids]
dataset = dataset[new_ids, :]
labels = labels[new_ids]
x = x[new_ids, :]
ncell = np.shape(dataset)[0]
index = np.random.choice(ncell,ncell,replace=False)
dataset = dataset[index, :] # cell by gene matrix
labels = labels[index]
if tissue_key is not None:
tissues = tissues[index]
return dataset, genes, labels, tissues, x
def exact_match_co_name_2_co_id(labels, lab2co, cl_obo_file = None):
if cl_obo_file is None:
return lab2co
co2name, name2co = get_ontology_name(obo_file = cl_obo_file)
for label in labels:
if label.lower() in name2co:
lab2co[label.lower()] = name2co[label.lower()]
for name in name2co:
lab2co[name.lower()] = name2co[name]
return lab2co
def map_and_select_labels(labels, cell_ontology_ids, obo_file, ct_mapping_key = {}, nlp_mapping = True, nlp_mapping_cutoff = 0.8, co2emb = None, cl_obo_file = None):
lab2co = {}
if nlp_mapping:
if co2emb is None:
sys.exit('Please provide cell type embedding to do NLP-based mapping.')
lab2co = fine_nearest_co_using_nlp(np.unique(labels), co2emb, obo_file,nlp_mapping_cutoff = nlp_mapping_cutoff)
lab2co = exact_match_co_name_2_co_id(np.unique(labels), lab2co, cl_obo_file = cl_obo_file)
for ct in ct_mapping_key:
lab2co[ct_mapping_key[ct]] = lab2co[ct]
ind = []
lab_id = []
unfound_labs = set()
for i,l in enumerate(labels):
if l in cell_ontology_ids:
ind.append(i)
lab_id.append(l)
elif l.lower() in lab2co:
ind.append(i)
lab_id.append(lab2co[l.lower()])
else:
unfound_labs.add(l)
frac = len(ind) * 1. / len(labels)
ind = np.array(ind)
labels = np.array(lab_id)
unfound_labs = set(unfound_labs)
warn_message = 'Warning: Only: %f precentage of labels are in the Cell Ontology. The remaining cells are excluded! Consider using NLP mapping and choose a small mapping cutoff (nlp_mapping_cutoff)' % (frac * 100)
if frac < 0.5:
print (warn_message)
print ('Here are unfound labels:',unfound_labs)
return ind, labels, unfound_labs
def parse_h5ad(file,seed=1,nsample=1e10,label_key='cell_ontology_class', read_tissue = False, batch_key = '', filter_key={}, cell_ontology_file = None, exclude_non_leaf_ontology = True, exclude_non_ontology=True, cl_obo_file = None):
'''
read h5ad file
feature: cell by gene expression
label: cell ontology class
genes: gene names HGNC
'''
np.random.seed(seed)
x = read_h5ad(file)
ncell = np.shape(x.X)[0]
select_cells = set(range(ncell))
for key in filter_key:
value = filter_key[key]
select_cells = select_cells & set(np.where(np.array(x.obs[key])==value)[0])
select_cells = sorted(select_cells)
feature = x.X[select_cells, :]
labels = np.array(x.obs[label_key].tolist())[select_cells]
if read_tissue:
tissues = np.array(x.obs['tissue'].tolist())[select_cells]
if batch_key=='' or batch_key not in x.obs.keys():
batch_labels = np.ones(len(labels))
else:
batch_labels = np.array(x.obs[batch_key].tolist())[select_cells]
genes = x.var.index
ncell = len(select_cells)
if exclude_non_ontology:
new_ids, labels = exclude_non_ontology_term(cl_obo_file, labels, label_key)
feature = feature[new_ids, :]
batch_labels = batch_labels[new_ids]
if exclude_non_leaf_ontology:
new_ids, exclude_terms = exclude_parent_child_nodes(cell_ontology_file, labels)
#print (len(exclude_terms),'non leaf terms are excluded')
feature = feature[new_ids, :]
batch_labels = batch_labels[new_ids]
labels = labels[new_ids]
if read_tissue:
tissues = tissues[new_ids]
ncell = len(labels)
index = np.random.choice(ncell,min(nsample,ncell),replace=False)
batch_labels = batch_labels[index]
feature = feature[index, :] # cell by gene matrix
labels = labels[index]
if read_tissue:
tissues = tissues[index]
genes = x.var.index
#corrected_feature = run_scanorama_same_genes(feature, batch_labels)
corrected_feature = feature.toarray()
genes = [x.upper() for x in genes]
genes = np.array(genes)
if read_tissue:
assert(len(tissues) == len(labels))
return corrected_feature, labels, genes, tissues
else:
return corrected_feature, labels, genes
def exclude_parent_child_nodes(cell_ontology_file,labels):
uniq_labels = np.unique(labels)
excludes = set()
net = collections.defaultdict(dict)
fin = open(cell_ontology_file)
for line in fin:
s,p = line.strip().split('\t')
net[s][p] = 1 #p is parent
fin.close()
for n in list(net.keys()):
ngh = get_ontology_parents(net, n)
for n1 in ngh:
net[n][n1] = 1
for l1 in uniq_labels:
for l2 in uniq_labels:
if l1 in net[l2] and l1!=l2: #l1 is l2 parent
excludes.add(l1)
#print (excludes)
new_ids = []
for i in range(len(labels)):
if labels[i] not in excludes:
new_ids.append(i)
new_ids = np.array(new_ids)
return new_ids, excludes
def corr2_coeff(A, B):
# Rowwise mean of input arrays & subtract from input arrays themeselves
A_mA = A - A.mean(1)[:, None]
B_mB = B - B.mean(1)[:, None]
# Sum of squares across rows
ssA = (A_mA**2).sum(1)
ssB = (B_mB**2).sum(1)
# Finally get corr coeff
return np.dot(A_mA, B_mB.T) / np.sqrt(np.dot(ssA[:, None],ssB[None]))
def extract_data_based_on_class(feats, labels, sel_labels):
ind = []
for l in sel_labels:
id = np.where(labels == l)[0]
ind.extend(id)
np.random.shuffle(ind)
X = feats[ind,:]
Y = labels[ind]
return X, Y, ind
def SplitTrainTest(all_X, all_Y, all_tissues = None, random_state=10, nfold_cls = 0.3, nfold_sample = 0.2, nmin_size=10):
np.random.seed(random_state)
cls = np.unique(all_Y)
cls2ct = Counter(all_Y)
ncls = len(cls)
test_cls = list(np.random.choice(cls, int(ncls * nfold_cls), replace=False))
for c in cls2ct:
if cls2ct[c] < nmin_size:
test_cls.append(c)
test_cls = np.unique(test_cls)
#add rare class to test, since they cannot be split into train and test by using train_test_split(stratify=True)
train_cls = [x for x in cls if x not in test_cls]
train_cls = np.array(train_cls)
train_X, train_Y, train_ind = extract_data_based_on_class(all_X, all_Y, train_cls)
test_X, test_Y, test_ind = extract_data_based_on_class(all_X, all_Y, test_cls)
if all_tissues is not None:
train_tissues = all_tissues[train_ind]
test_tissues = all_tissues[test_ind]
train_X_train, train_X_test, train_Y_train, train_Y_test, train_tissues_train, train_tissues_test = train_test_split(
train_X, train_Y, train_tissues, test_size=nfold_sample, stratify = train_Y,random_state=random_state)
test_tissues = np.concatenate((test_tissues, train_tissues_test))
train_tissues = train_tissues_train
else:
train_X_train, train_X_test, train_Y_train, train_Y_test = train_test_split(
train_X, train_Y, test_size=nfold_sample, stratify = train_Y,random_state=random_state)
test_X = np.vstack((test_X, train_X_test))
test_Y = np.concatenate((test_Y, train_Y_test))
train_X = train_X_train
train_Y = train_Y_train
if all_tissues is not None:
return train_X, train_Y, train_tissues, test_X, test_Y, test_tissues
else:
return train_X, train_Y, test_X, test_Y
'''
def SplitTrainTest(all_X, all_Y, all_tissues = None, random_state=10, nfold_cls = 0.3, nfold_sample = 0.2, nmin_size=10):
np.random.seed(random_state)
cls = np.unique(all_Y)
cls2ct = Counter(all_Y)
ncls = len(cls)
rare_cls = []
not_rare_cls = []
for c in cls2ct:
if cls2ct[c] < 2:
continue
elif cls2ct[c] < nmin_size:
rare_cls.append(c)
else:
not_rare_cls.append(c)
cls = np.concatenate((rare_cls, not_rare_cls))
ncls = len(cls)
rare_cls = np.array(rare_cls)
not_rare_cls = np.array(not_rare_cls)
train_non_rare_cls = list(np.random.choice(not_rare_cls, int(len(not_rare_cls) * (1 - nfold_cls)), replace=False))
train_cls = np.concatenate((train_non_rare_cls, rare_cls))
test_cls = [x for x in cls if x not in train_cls]
test_cls = np.array(test_cls)
assert(len(test_cls) + len(train_cls) == ncls)
assert(len(set(test_cls) & set(train_cls)) == 0)
#add rare class to test, since they cannot be split into train and test by using train_test_split(stratify=True)
train_X, train_Y, train_ind = extract_data_based_on_class(all_X, all_Y, train_cls)
test_X, test_Y, test_ind = extract_data_based_on_class(all_X, all_Y, test_cls)
if all_tissues is not None:
train_tissues = all_tissues[train_ind]
test_tissues = all_tissues[test_ind]
train_X_train, train_X_test, train_Y_train, train_Y_test, train_tissues_train, train_tissues_test = train_test_split(
train_X, train_Y, train_tissues, test_size=nfold_sample, stratify = train_Y,random_state=random_state)
test_tissues = np.concatenate((test_tissues, train_tissues_test))
train_tissues = train_tissues_train
else:
train_X_train, train_X_test, train_Y_train, train_Y_test = train_test_split(
train_X, train_Y, test_size=nfold_sample, stratify = train_Y,random_state=random_state)
test_X = np.vstack((test_X, train_X_test))
test_Y = np.concatenate((test_Y, train_Y_test))
train_X = train_X_train
train_Y = train_Y_train
if all_tissues is not None:
return train_X, train_Y, train_tissues, test_X, test_Y, test_tissues
else:
return train_X, train_Y, test_X, test_Y
'''
def LeaveOneOutTrainTest(all_X, all_Y, test_Y, all_tissues = None, random_state=10, nfold_sample = 0.2, nmin_size=10):
np.random.seed(random_state)
cls = np.unique(all_Y)
cls2ct = Counter(all_Y)
ncls = len(cls)
test_cls = [test_Y]
test_cls = np.unique(test_cls)
#add rare class to test, since they cannot be split into train and test by using train_test_split(stratify=True)
train_cls = [x for x in cls if x not in test_cls]
train_cls = np.array(train_cls)
train_X, train_Y, train_ind = extract_data_based_on_class(all_X, all_Y, train_cls)
test_X, test_Y, test_ind = extract_data_based_on_class(all_X, all_Y, test_cls)
if all_tissues is not None:
train_tissues = all_tissues[train_ind]
test_tissues = all_tissues[test_ind]
train_X_train, train_X_test, train_Y_train, train_Y_test, train_tissues_train, train_tissues_test = train_test_split(
train_X, train_Y, train_tissues, test_size=nfold_sample, stratify = train_Y,random_state=random_state)
test_tissues = np.concatenate((test_tissues, train_tissues_test))
train_tissues = train_tissues_train
else:
train_X_train, train_X_test, train_Y_train, train_Y_test = train_test_split(
train_X, train_Y, test_size=nfold_sample, stratify = train_Y,random_state=random_state)
test_X = np.vstack((test_X, train_X_test))
test_Y = np.concatenate((test_Y, train_Y_test))
train_X = train_X_train
train_Y = train_Y_train
if all_tissues is not None:
return train_X, train_Y, train_tissues, test_X, test_Y, test_tissues
else:
return train_X, train_Y, test_X, test_Y
def renorm(X):
Y = X.copy()
Y = Y.astype(float)
ngene,nsample = Y.shape
s = np.sum(Y, axis=0)
#print s.shape()
for i in range(nsample):
if s[i]==0:
s[i] = 1
if i < ngene:
Y[i,i] = 1
else:
for j in range(ngene):
Y[j,i] = 1. / ngene
Y[:,i] = Y[:,i]/s[i]
return Y
def RandomWalkRestart(A, rst_prob, delta = 1e-4, reset=None, max_iter=50,use_torch=False,return_torch=False):
if use_torch:
device = torch.device("cuda:0")
nnode = A.shape[0]
#print nnode
if reset is None:
reset = np.eye(nnode)
nsample,nnode = reset.shape
#print nsample,nnode
P = renorm(A)
P = P.T
norm_reset = renorm(reset.T)
norm_reset = norm_reset.T
if use_torch:
norm_reset = torch.from_numpy(norm_reset).float().to(device)
P = torch.from_numpy(P).float().to(device)
Q = norm_reset
for i in range(1,max_iter):
#Q = gnp.garray(Q)
#P = gnp.garray(P)
if use_torch:
Q_new = rst_prob*norm_reset + (1-rst_prob) * torch.mm(Q, P)#.as_numpy_array()
delta = torch.norm(Q-Q_new, 2)
else:
Q_new = rst_prob*norm_reset + (1-rst_prob) * np.dot(Q, P)#.as_numpy_array()
delta = np.linalg.norm(Q-Q_new, 'fro')
Q = Q_new
#print (i,Q)
sys.stdout.flush()
if delta < 1e-4:
break
if use_torch and not return_torch:
Q = Q.cpu().numpy()
return Q
def DCA_vector(Q, dim):
nnode = Q.shape[0]
alpha = 1. / (nnode **2)
Q = np.log(Q + alpha) - np.log(alpha);
#Q = Q * Q';
[U, S, V] = svds(Q, dim);
S = np.diag(S)
X = np.dot(U, np.sqrt(S))
Y = np.dot(np.sqrt(S), V)
Y = np.transpose(Y)
return X,U,S,V,Y
def read_cell_ontology_nlp(l2i, ontology_nlp_file, ontology_nlp_emb_file):
ncls = len(l2i)
net = np.zeros((ncls, ncls))
bin_net = np.zeros((ncls, ncls))
fin = open(ontology_nlp_file)
for line in fin:
s,p,wt = line.upper().strip().split('\t')
wt = float(wt)
net[l2i[s], l2i[p]] = np.exp(wt)
net[l2i[p], l2i[s]] = np.exp(wt)
bin_net[l2i[s], l2i[p]] = 1
bin_net[l2i[p], l2i[s]] = 1
fin.close()
l2vec = {}
fin = open(ontology_nlp_emb_file)
for line in fin:
w = line.upper().strip().split('\t')
l2vec[w[0]] = []
dim = len(w)-1
for i in range(1,len(w)):
l2vec[w[0]].append(float(w[i]))
fin.close()
l2vec_mat = np.zeros((ncls, dim))
for l in l2vec:
if l.upper() not in l2i:
continue
l2vec_mat[l2i[l.upper()],:] = l2vec[l]
'''
net_sum = np.sum(net,axis=0)
for i in range(ncls):
if net_sum[i] == 0:
net[i,i] = 1.
net[:,i] /= np.sum(net[:,i])
#net = net / net.sum(axis=1)[:, np.newaxis]