0
|
1
|
|
2 from __future__ import division
|
|
3 import os
|
|
4 import sys
|
|
5 import pandas as pd
|
|
6 import collections
|
|
7 import pickle as pk
|
|
8 import argparse
|
|
9 from sklearn.cluster import KMeans
|
|
10 import matplotlib.pyplot as plt
|
|
11
|
|
12 ########################## argparse ###########################################
|
|
13
|
|
14 def process_args(args):
|
|
15 parser = argparse.ArgumentParser(usage = '%(prog)s [options]',
|
|
16 description = 'process some value\'s' +
|
|
17 ' genes to create class.')
|
|
18 parser.add_argument('-rs', '--rules_selector',
|
|
19 type = str,
|
|
20 default = 'HMRcore',
|
|
21 choices = ['HMRcore', 'Recon', 'Custom'],
|
|
22 help = 'chose which type of dataset you want use')
|
|
23 parser.add_argument('-cr', '--custom',
|
|
24 type = str,
|
|
25 help='your dataset if you want custom rules')
|
|
26 parser.add_argument('-ch', '--cond_hier',
|
|
27 type = str,
|
|
28 default = 'no',
|
|
29 choices = ['no', 'yes'],
|
|
30 help = 'chose if you wanna hierical dendrogram')
|
|
31 parser.add_argument('-lk', '--k_min',
|
|
32 type = int,
|
|
33 help = 'min number of cluster')
|
|
34 parser.add_argument('-uk', '--k_max',
|
|
35 type = int,
|
|
36 help = 'max number of cluster')
|
|
37 parser.add_argument('-li', '--linkage',
|
|
38 type = str,
|
|
39 choices = ['single', 'complete', 'average'],
|
|
40 help='linkage hierarchical cluster')
|
|
41 parser.add_argument('-d', '--data',
|
|
42 type = str,
|
|
43 required = True,
|
|
44 help = 'input dataset')
|
|
45 parser.add_argument('-n', '--none',
|
|
46 type = str,
|
|
47 default = 'true',
|
|
48 choices = ['true', 'false'],
|
|
49 help = 'compute Nan values')
|
|
50 parser.add_argument('-td', '--tool_dir',
|
|
51 type = str,
|
|
52 required = True,
|
|
53 help = 'your tool directory')
|
|
54 parser.add_argument('-na', '--name',
|
|
55 type = str,
|
|
56 help = 'name of dataset')
|
|
57 parser.add_argument('-de', '--dendro',
|
|
58 help = "Dendrogram out")
|
|
59 parser.add_argument('-ol', '--out_log',
|
|
60 help = "Output log")
|
|
61 parser.add_argument('-el', '--elbow',
|
|
62 help = "Out elbow")
|
|
63 args = parser.parse_args()
|
|
64 return args
|
|
65
|
|
66 ########################### warning ###########################################
|
|
67
|
|
68 def warning(s):
|
|
69 args = process_args(sys.argv)
|
|
70 with open(args.out_log, 'a') as log:
|
|
71 log.write(s)
|
|
72
|
|
73 ############################ dataset input ####################################
|
|
74
|
|
75 def read_dataset(data, name):
|
|
76 try:
|
|
77 dataset = pd.read_csv(data, sep = '\t', header = 0)
|
|
78 except pd.errors.EmptyDataError:
|
|
79 sys.exit('Execution aborted: wrong format of '+name+'\n')
|
|
80 if len(dataset.columns) < 2:
|
|
81 sys.exit('Execution aborted: wrong format of '+name+'\n')
|
|
82 return dataset
|
|
83
|
|
84 ############################ dataset name #####################################
|
|
85
|
|
86 def name_dataset(name_data, count):
|
|
87 if str(name_data) == 'Dataset':
|
|
88 return str(name_data) + '_' + str(count)
|
|
89 else:
|
|
90 return str(name_data)
|
|
91
|
|
92 ############################ load id e rules ##################################
|
|
93
|
|
94 def load_id_rules(reactions):
|
|
95 ids, rules = [], []
|
|
96 for key, value in reactions.items():
|
|
97 ids.append(key)
|
|
98 rules.append(value)
|
|
99 return (ids, rules)
|
|
100
|
|
101 ############################ check_methods ####################################
|
|
102
|
|
103 def gene_type(l, name):
|
|
104 if check_hgnc(l):
|
|
105 return 'hugo_id'
|
|
106 elif check_ensembl(l):
|
|
107 return 'ensembl_gene_id'
|
|
108 elif check_symbol(l):
|
|
109 return 'symbol'
|
|
110 elif check_entrez(l):
|
|
111 return 'entrez_id'
|
|
112 else:
|
|
113 sys.exit('Execution aborted:\n' +
|
|
114 'gene ID type in ' + name + ' not supported. Supported ID' +
|
|
115 'types are: HUGO ID, Ensemble ID, HUGO symbol, Entrez ID\n')
|
|
116
|
|
117 def check_hgnc(l):
|
|
118 if len(l) > 5:
|
|
119 if (l.upper()).startswith('HGNC:'):
|
|
120 return l[5:].isdigit()
|
|
121 else:
|
|
122 return False
|
|
123 else:
|
|
124 return False
|
|
125
|
|
126 def check_ensembl(l):
|
|
127 if len(l) == 15:
|
|
128 if (l.upper()).startswith('ENS'):
|
|
129 return l[4:].isdigit()
|
|
130 else:
|
|
131 return False
|
|
132 else:
|
|
133 return False
|
|
134
|
|
135 def check_symbol(l):
|
|
136 if len(l) > 0:
|
|
137 if l[0].isalpha() and l[1:].isalnum():
|
|
138 return True
|
|
139 else:
|
|
140 return False
|
|
141 else:
|
|
142 return False
|
|
143
|
|
144 def check_entrez(l):
|
|
145 if len(l) > 0:
|
|
146 return l.isdigit()
|
|
147 else:
|
|
148 return False
|
|
149
|
|
150 def check_bool(b):
|
|
151 if b == 'true':
|
|
152 return True
|
|
153 elif b == 'false':
|
|
154 return False
|
|
155
|
|
156 ############################ make recon #######################################
|
|
157
|
|
158 def check_and_doWord(l):
|
|
159 tmp = []
|
|
160 tmp_genes = []
|
|
161 count = 0
|
|
162 while l:
|
|
163 if count >= 0:
|
|
164 if l[0] == '(':
|
|
165 count += 1
|
|
166 tmp.append(l[0])
|
|
167 l.pop(0)
|
|
168 elif l[0] == ')':
|
|
169 count -= 1
|
|
170 tmp.append(l[0])
|
|
171 l.pop(0)
|
|
172 elif l[0] == ' ':
|
|
173 l.pop(0)
|
|
174 else:
|
|
175 word = []
|
|
176 while l:
|
|
177 if l[0] in [' ', '(', ')']:
|
|
178 break
|
|
179 else:
|
|
180 word.append(l[0])
|
|
181 l.pop(0)
|
|
182 word = ''.join(word)
|
|
183 tmp.append(word)
|
|
184 if not(word in ['or', 'and']):
|
|
185 tmp_genes.append(word)
|
|
186 else:
|
|
187 return False
|
|
188 if count == 0:
|
|
189 return (tmp, tmp_genes)
|
|
190 else:
|
|
191 return False
|
|
192
|
|
193 def brackets_to_list(l):
|
|
194 tmp = []
|
|
195 while l:
|
|
196 if l[0] == '(':
|
|
197 l.pop(0)
|
|
198 tmp.append(resolve_brackets(l))
|
|
199 else:
|
|
200 tmp.append(l[0])
|
|
201 l.pop(0)
|
|
202 return tmp
|
|
203
|
|
204 def resolve_brackets(l):
|
|
205 tmp = []
|
|
206 while l[0] != ')':
|
|
207 if l[0] == '(':
|
|
208 l.pop(0)
|
|
209 tmp.append(resolve_brackets(l))
|
|
210 else:
|
|
211 tmp.append(l[0])
|
|
212 l.pop(0)
|
|
213 l.pop(0)
|
|
214 return tmp
|
|
215
|
|
216 def priorityAND(l):
|
|
217 tmp = []
|
|
218 flag = True
|
|
219 while l:
|
|
220 if len(l) == 1:
|
|
221 if isinstance(l[0], list):
|
|
222 tmp.append(priorityAND(l[0]))
|
|
223 else:
|
|
224 tmp.append(l[0])
|
|
225 l = l[1:]
|
|
226 elif l[0] == 'or':
|
|
227 tmp.append(l[0])
|
|
228 flag = False
|
|
229 l = l[1:]
|
|
230 elif l[1] == 'or':
|
|
231 if isinstance(l[0], list):
|
|
232 tmp.append(priorityAND(l[0]))
|
|
233 else:
|
|
234 tmp.append(l[0])
|
|
235 tmp.append(l[1])
|
|
236 flag = False
|
|
237 l = l[2:]
|
|
238 elif l[1] == 'and':
|
|
239 tmpAnd = []
|
|
240 if isinstance(l[0], list):
|
|
241 tmpAnd.append(priorityAND(l[0]))
|
|
242 else:
|
|
243 tmpAnd.append(l[0])
|
|
244 tmpAnd.append(l[1])
|
|
245 if isinstance(l[2], list):
|
|
246 tmpAnd.append(priorityAND(l[2]))
|
|
247 else:
|
|
248 tmpAnd.append(l[2])
|
|
249 l = l[3:]
|
|
250 while l:
|
|
251 if l[0] == 'and':
|
|
252 tmpAnd.append(l[0])
|
|
253 if isinstance(l[1], list):
|
|
254 tmpAnd.append(priorityAND(l[1]))
|
|
255 else:
|
|
256 tmpAnd.append(l[1])
|
|
257 l = l[2:]
|
|
258 elif l[0] == 'or':
|
|
259 flag = False
|
|
260 break
|
|
261 if flag == True: #se ci sono solo AND nella lista
|
|
262 tmp.extend(tmpAnd)
|
|
263 elif flag == False:
|
|
264 tmp.append(tmpAnd)
|
|
265 return tmp
|
|
266
|
|
267 def checkRule(l):
|
|
268 if len(l) == 1:
|
|
269 if isinstance(l[0], list):
|
|
270 if checkRule(l[0]) is False:
|
|
271 return False
|
|
272 elif len(l) > 2:
|
|
273 if checkRule2(l) is False:
|
|
274 return False
|
|
275 else:
|
|
276 return False
|
|
277 return True
|
|
278
|
|
279 def checkRule2(l):
|
|
280 while l:
|
|
281 if len(l) == 1:
|
|
282 return False
|
|
283 elif isinstance(l[0], list) and l[1] in ['and', 'or']:
|
|
284 if checkRule(l[0]) is False:
|
|
285 return False
|
|
286 if isinstance(l[2], list):
|
|
287 if checkRule(l[2]) is False:
|
|
288 return False
|
|
289 l = l[3:]
|
|
290 elif l[1] in ['and', 'or']:
|
|
291 if isinstance(l[2], list):
|
|
292 if checkRule(l[2]) is False:
|
|
293 return False
|
|
294 l = l[3:]
|
|
295 elif l[0] in ['and', 'or']:
|
|
296 if isinstance(l[1], list):
|
|
297 if checkRule(l[1]) is False:
|
|
298 return False
|
|
299 l = l[2:]
|
|
300 else:
|
|
301 return False
|
|
302 return True
|
|
303
|
|
304 def do_rules(rules):
|
|
305 split_rules = []
|
|
306 err_rules = []
|
|
307 tmp_gene_in_rule = []
|
|
308 for i in range(len(rules)):
|
|
309 tmp = list(rules[i])
|
|
310 if tmp:
|
|
311 tmp, tmp_genes = check_and_doWord(tmp)
|
|
312 tmp_gene_in_rule.extend(tmp_genes)
|
|
313 if tmp is False:
|
|
314 split_rules.append([])
|
|
315 err_rules.append(rules[i])
|
|
316 else:
|
|
317 tmp = brackets_to_list(tmp)
|
|
318 if checkRule(tmp):
|
|
319 split_rules.append(priorityAND(tmp))
|
|
320 else:
|
|
321 split_rules.append([])
|
|
322 err_rules.append(rules[i])
|
|
323 else:
|
|
324 split_rules.append([])
|
|
325 if err_rules:
|
|
326 warning('Warning: wrong format rule in ' + str(err_rules) + '\n')
|
|
327 return (split_rules, list(set(tmp_gene_in_rule)))
|
|
328
|
|
329 def make_recon(data):
|
|
330 try:
|
|
331 import cobra as cb
|
|
332 import warnings
|
|
333 with warnings.catch_warnings():
|
|
334 warnings.simplefilter('ignore')
|
|
335 recon = cb.io.read_sbml_model(data)
|
|
336 react = recon.reactions
|
|
337 rules = [react[i].gene_reaction_rule for i in range(len(react))]
|
|
338 ids = [react[i].id for i in range(len(react))]
|
|
339 except cb.io.sbml3.CobraSBMLError:
|
|
340 try:
|
|
341 data = (pd.read_csv(data, sep = '\t', dtype = str)).fillna('')
|
|
342 if len(data.columns) < 2:
|
|
343 sys.exit('Execution aborted: wrong format of ' +
|
|
344 'custom GPR rules\n')
|
|
345 if not len(data.columns) == 2:
|
|
346 warning('WARNING: more than 2 columns in custom GPR rules.\n' +
|
|
347 'Extra columns have been disregarded\n')
|
|
348 ids = list(data.iloc[:, 0])
|
|
349 rules = list(data.iloc[:, 1])
|
|
350 except pd.errors.EmptyDataError:
|
|
351 sys.exit('Execution aborted: wrong format of custom GPR rules\n')
|
|
352 except pd.errors.ParserError:
|
|
353 sys.exit('Execution aborted: wrong format of custom GPR rules\n')
|
|
354 split_rules, tmp_genes = do_rules(rules)
|
|
355 gene_in_rule = {}
|
|
356 for i in tmp_genes:
|
|
357 gene_in_rule[i] = 'ok'
|
|
358 return (ids, split_rules, gene_in_rule)
|
|
359
|
|
360 ############################ resolve_methods ##################################
|
|
361
|
|
362 def replace_gene_value(l, d):
|
|
363 tmp = []
|
|
364 err = []
|
|
365 while l:
|
|
366 if isinstance(l[0], list):
|
|
367 tmp_rules, tmp_err = replace_gene_value(l[0], d)
|
|
368 tmp.append(tmp_rules)
|
|
369 err.extend(tmp_err)
|
|
370 else:
|
|
371 value = replace_gene(l[0],d)
|
|
372 tmp.append(value)
|
|
373 if value == None:
|
|
374 err.append(l[0])
|
|
375 l = l[1:]
|
|
376 return (tmp, err)
|
|
377
|
|
378 def replace_gene(l, d):
|
|
379 if l =='and' or l == 'or':
|
|
380 return l
|
|
381 else:
|
|
382 value = d.get(l, None)
|
|
383 if not(value == None or isinstance(value, (int, float))):
|
|
384 sys.exit('Execution aborted: ' + value + ' value not valid\n')
|
|
385 return value
|
|
386
|
|
387 def compute(val1, op, val2, cn):
|
|
388 if val1 != None and val2 != None:
|
|
389 if op == 'and':
|
|
390 return min(val1, val2)
|
|
391 else:
|
|
392 return val1 + val2
|
|
393 elif op == 'and':
|
|
394 if cn is True:
|
|
395 if val1 != None:
|
|
396 return val1
|
|
397 elif val2 != None:
|
|
398 return val2
|
|
399 else:
|
|
400 return None
|
|
401 else:
|
|
402 return None
|
|
403 else:
|
|
404 if val1 != None:
|
|
405 return val1
|
|
406 elif val2 != None:
|
|
407 return val2
|
|
408 else:
|
|
409 return None
|
|
410
|
|
411 def control(ris, l, cn):
|
|
412 if len(l) == 1:
|
|
413 if isinstance(l[0], (float, int)) or l[0] == None:
|
|
414 return l[0]
|
|
415 elif isinstance(l[0], list):
|
|
416 return control(None, l[0], cn)
|
|
417 else:
|
|
418 return False
|
|
419 elif len(l) > 2:
|
|
420 return control_list(ris, l, cn)
|
|
421 else:
|
|
422 return False
|
|
423
|
|
424 def control_list(ris, l, cn):
|
|
425 while l:
|
|
426 if len(l) == 1:
|
|
427 return False
|
|
428 elif (isinstance(l[0], (float, int)) or
|
|
429 l[0] == None) and l[1] in ['and', 'or']:
|
|
430 if isinstance(l[2], (float, int)) or l[2] == None:
|
|
431 ris = compute(l[0], l[1], l[2], cn)
|
|
432 elif isinstance(l[2], list):
|
|
433 tmp = control(None, l[2], cn)
|
|
434 if tmp is False:
|
|
435 return False
|
|
436 else:
|
|
437 ris = compute(l[0], l[1], tmp, cn)
|
|
438 else:
|
|
439 return False
|
|
440 l = l[3:]
|
|
441 elif l[0] in ['and', 'or']:
|
|
442 if isinstance(l[1], (float, int)) or l[1] == None:
|
|
443 ris = compute(ris, l[0], l[1], cn)
|
|
444 elif isinstance(l[1], list):
|
|
445 tmp = control(None,l[1], cn)
|
|
446 if tmp is False:
|
|
447 return False
|
|
448 else:
|
|
449 ris = compute(ris, l[0], tmp, cn)
|
|
450 else:
|
|
451 return False
|
|
452 l = l[2:]
|
|
453 elif isinstance(l[0], list) and l[1] in ['and', 'or']:
|
|
454 if isinstance(l[2], (float, int)) or l[2] == None:
|
|
455 tmp = control(None, l[0], cn)
|
|
456 if tmp is False:
|
|
457 return False
|
|
458 else:
|
|
459 ris = compute(tmp, l[1], l[2], cn)
|
|
460 elif isinstance(l[2], list):
|
|
461 tmp = control(None, l[0], cn)
|
|
462 tmp2 = control(None, l[2], cn)
|
|
463 if tmp is False or tmp2 is False:
|
|
464 return False
|
|
465 else:
|
|
466 ris = compute(tmp, l[1], tmp2, cn)
|
|
467 else:
|
|
468 return False
|
|
469 l = l[3:]
|
|
470 else:
|
|
471 return False
|
|
472 return ris
|
|
473
|
|
474 ############################ gene #############################################
|
|
475
|
|
476 def data_gene(gene, type_gene, name, gene_custom):
|
|
477 args = process_args(sys.argv)
|
|
478 for i in range(len(gene)):
|
|
479 tmp = gene.iloc[i, 0]
|
|
480 if tmp.startswith(' ') or tmp.endswith(' '):
|
|
481 gene.iloc[i, 0] = (tmp.lstrip()).rstrip()
|
|
482 gene_dup = [item for item, count in
|
|
483 collections.Counter(gene[gene.columns[0]]).items() if count > 1]
|
|
484 pat_dup = [item for item, count in
|
|
485 collections.Counter(list(gene.columns)).items() if count > 1]
|
|
486 if gene_dup:
|
|
487 if gene_custom == None:
|
|
488 if args.rules_selector == 'HMRcore':
|
|
489 gene_in_rule = pk.load(open(args.tool_dir +
|
|
490 '/local/HMRcore_genes.p', 'rb'))
|
|
491 elif args.rules_selector == 'Recon':
|
|
492 gene_in_rule = pk.load(open(args.tool_dir +
|
|
493 '/local/Recon_genes.p', 'rb'))
|
|
494 gene_in_rule = gene_in_rule.get(type_gene)
|
|
495 else:
|
|
496 gene_in_rule = gene_custom
|
|
497 tmp = []
|
|
498 for i in gene_dup:
|
|
499 if gene_in_rule.get(i) == 'ok':
|
|
500 tmp.append(i)
|
|
501 if tmp:
|
|
502 sys.exit('Execution aborted because gene ID '
|
|
503 + str(tmp) + ' in ' + name + ' is duplicated\n')
|
|
504 if pat_dup:
|
|
505 sys.exit('Execution aborted: duplicated label\n'
|
|
506 + str(pat_dup) + 'in ' + name + '\n')
|
|
507 return (gene.set_index(gene.columns[0])).to_dict()
|
|
508
|
|
509 ############################ resolve ##########################################
|
|
510
|
|
511 def resolve(genes, rules, ids, resolve_none, name):
|
|
512 resolve_rules = {}
|
|
513 not_found = []
|
|
514 flag = False
|
|
515 for key, value in genes.items():
|
|
516 tmp_resolve = []
|
|
517 for i in range(len(rules)):
|
|
518 tmp = rules[i]
|
|
519 if tmp:
|
|
520 tmp, err = replace_gene_value(tmp, value)
|
|
521 if err:
|
|
522 not_found.extend(err)
|
|
523 ris = control(None, tmp, resolve_none)
|
|
524 if ris is False or ris == None:
|
|
525 tmp_resolve.append(None)
|
|
526 else:
|
|
527 tmp_resolve.append(ris)
|
|
528 flag = True
|
|
529 else:
|
|
530 tmp_resolve.append(None)
|
|
531 resolve_rules[key] = tmp_resolve
|
|
532 if flag is False:
|
|
533 sys.exit('Execution aborted: no computable score' +
|
|
534 ' (due to missing gene values) for class '
|
|
535 + name + ', the class has been disregarded\n')
|
|
536 return (resolve_rules, list(set(not_found)))
|
|
537
|
|
538 ################################# clustering ##################################
|
|
539
|
|
540 def f_cluster(resolve_rules):
|
|
541 os.makedirs('cluster_out')
|
|
542 args = process_args(sys.argv)
|
|
543 cluster_data = pd.DataFrame.from_dict(resolve_rules, orient = 'index')
|
|
544 for i in cluster_data.columns:
|
|
545 tmp = cluster_data[i][0]
|
|
546 if tmp == None:
|
|
547 cluster_data = cluster_data.drop(columns=[i])
|
|
548 distorsion = []
|
|
549 for i in range(args.k_min, args.k_max+1):
|
|
550 tmp_kmeans = KMeans(n_clusters = i,
|
|
551 n_init = 100,
|
|
552 max_iter = 300,
|
|
553 random_state = 0).fit(cluster_data)
|
|
554 distorsion.append(tmp_kmeans.inertia_)
|
|
555 predict = tmp_kmeans.predict(cluster_data)
|
|
556 predict = [x+1 for x in predict]
|
|
557 classe = (pd.DataFrame(zip(cluster_data.index, predict))).astype(str)
|
|
558 dest = 'cluster_out/K=' + str(i) + '_' + args.name+'.tsv'
|
|
559 classe.to_csv(dest, sep = '\t', index = False,
|
|
560 header = ['Patient_ID', 'Class'])
|
|
561 plt.figure(0)
|
|
562 plt.plot(range(args.k_min, args.k_max+1), distorsion, marker = 'o')
|
|
563 plt.xlabel('Number of cluster')
|
|
564 plt.ylabel('Distorsion')
|
|
565 plt.savefig(args.elbow, dpi = 240, format = 'pdf')
|
|
566 if args.cond_hier == 'yes':
|
|
567 import scipy.cluster.hierarchy as hier
|
|
568 lin = hier.linkage(cluster_data, args.linkage)
|
|
569 plt.figure(1)
|
|
570 plt.figure(figsize=(10, 5))
|
|
571 hier.dendrogram(lin, leaf_font_size = 2, labels = cluster_data.index)
|
|
572 plt.savefig(args.dendro, dpi = 480, format = 'pdf')
|
|
573 return None
|
|
574
|
|
575 ################################# main ########################################
|
|
576
|
|
577 def main():
|
|
578 args = process_args(sys.argv)
|
|
579 if args.k_min > args.k_max:
|
|
580 sys.exit('Execution aborted: max cluster > min cluster')
|
|
581 if args.rules_selector == 'HMRcore':
|
|
582 recon = pk.load(open(args.tool_dir + '/local/HMRcore_rules.p', 'rb'))
|
|
583 elif args.rules_selector == 'Recon':
|
|
584 recon = pk.load(open(args.tool_dir + '/local/Recon_rules.p', 'rb'))
|
|
585 elif args.rules_selector == 'Custom':
|
|
586 ids, rules, gene_in_rule = make_recon(args.custom)
|
|
587 resolve_none = check_bool(args.none)
|
|
588 dataset = read_dataset(args.data, args.name)
|
|
589 dataset.iloc[:, 0] = (dataset.iloc[:, 0]).astype(str)
|
|
590 type_gene = gene_type(dataset.iloc[0, 0], args.name)
|
|
591 if args.rules_selector != 'Custom':
|
|
592 genes = data_gene(dataset, type_gene, args.name, None)
|
|
593 ids, rules = load_id_rules(recon.get(type_gene))
|
|
594 elif args.rules_selector == 'Custom':
|
|
595 genes = data_gene(dataset, type_gene, args.name, gene_in_rule)
|
|
596 resolve_rules, err = resolve(genes, rules, ids, resolve_none, args.name)
|
|
597 if err:
|
|
598 warning('WARNING: gene\n' + str(err) + '\nnot found in class '
|
|
599 + args.name + ', the expression level for this gene ' +
|
|
600 'will be considered NaN\n')
|
|
601 f_cluster(resolve_rules)
|
|
602 warning('Execution succeeded')
|
|
603 return None
|
|
604
|
|
605 ###############################################################################
|
|
606
|
|
607 if __name__ == "__main__":
|
|
608 main() |