0
|
1 #!/usr/bin/env python3
|
|
2
|
|
3 import numpy as np
|
|
4 import subprocess
|
|
5 import math
|
|
6 import time
|
|
7 from operator import itemgetter
|
|
8 from collections import Counter
|
|
9 from itertools import groupby
|
|
10 import os
|
10
|
11 import re
|
0
|
12 import configuration
|
|
13 from tempfile import NamedTemporaryFile
|
|
14 import sys
|
|
15 import warnings
|
|
16 import shutil
|
|
17 from collections import defaultdict
|
|
18
|
|
19 np.set_printoptions(threshold=sys.maxsize)
|
|
20
|
|
21 def alignment_scoring():
|
|
22 ''' Create hash table for alignment similarity counting: for every
|
|
23 combination of aminoacids in alignment assign score from protein
|
|
24 scoring matrix defined in configuration file '''
|
|
25 score_dict = {}
|
|
26 with open(configuration.SC_MATRIX) as smatrix:
|
|
27 count = 1
|
|
28 for line in smatrix:
|
|
29 if not line.startswith("#"):
|
|
30 if count == 1:
|
|
31 aa_all = line.rstrip().replace(" ", "")
|
|
32 else:
|
|
33 count_aa = 1
|
|
34 line = list(filter(None, line.rstrip().split(" ")))
|
|
35 for aa in aa_all:
|
|
36 score_dict["{}{}".format(line[0], aa)] = line[count_aa]
|
|
37 count_aa += 1
|
|
38 count += 1
|
|
39 return score_dict
|
|
40
|
|
41
|
|
42 def characterize_fasta(QUERY, WIN_DOM):
|
|
43 ''' Find the sequences, their lengths, starts, ends and if
|
|
44 they exceed the window '''
|
|
45 with open(QUERY) as query:
|
|
46 headers = []
|
|
47 fasta_lengths = []
|
|
48 seq_starts = []
|
|
49 seq_ends = []
|
|
50 fasta_chunk_len = 0
|
|
51 count_line = 1
|
|
52 for line in query:
|
|
53 line = line.rstrip()
|
|
54 if line.startswith(">"):
|
|
55 headers.append(line.rstrip())
|
|
56 fasta_lengths.append(fasta_chunk_len)
|
|
57 fasta_chunk_len = 0
|
|
58 seq_starts.append(count_line + 1)
|
|
59 seq_ends.append(count_line - 1)
|
|
60 else:
|
|
61 fasta_chunk_len += len(line)
|
|
62 count_line += 1
|
|
63 seq_ends.append(count_line)
|
|
64 seq_ends = seq_ends[1:]
|
|
65 fasta_lengths.append(fasta_chunk_len)
|
|
66 fasta_lengths = fasta_lengths[1:]
|
|
67 # control if there are correct (unique) names for individual seqs:
|
|
68 # LASTAL takes seqs IDs till the first space which can then create problems with ambiguous records
|
|
69 if len(headers) > len(set([header.split(" ")[0] for header in headers
|
|
70 ])):
|
|
71 raise NameError(
|
|
72 '''Sequences in multifasta format are not named correctly:
|
|
73 seq IDs (before the first space) are the same''')
|
|
74
|
|
75 above_win = [idx
|
|
76 for idx, value in enumerate(fasta_lengths) if value > WIN_DOM]
|
|
77 below_win = [idx
|
|
78 for idx, value in enumerate(fasta_lengths)
|
|
79 if value <= WIN_DOM]
|
|
80 lens_above_win = np.array(fasta_lengths)[above_win]
|
|
81 return headers, above_win, below_win, lens_above_win, seq_starts, seq_ends
|
|
82
|
|
83
|
|
84 def split_fasta(QUERY, WIN_DOM, step, headers, above_win, below_win,
|
|
85 lens_above_win, seq_starts, seq_ends):
|
|
86 ''' Create temporary file containing all sequences - the ones that exceed
|
|
87 the window are cut with a set overlap (greater than domain size with a reserve) '''
|
|
88 with open(QUERY, "r") as query:
|
|
89 count_fasta_divided = 0
|
|
90 count_fasta_not_divided = 0
|
|
91 ntf = NamedTemporaryFile(delete=False)
|
|
92 divided = np.array(headers)[above_win]
|
|
93 row_length = configuration.FASTA_LINE
|
|
94 for line in query:
|
|
95 line = line.rstrip()
|
|
96 if line.startswith(">") and line in divided:
|
|
97 stop_line = seq_ends[above_win[
|
|
98 count_fasta_divided]] - seq_starts[above_win[
|
|
99 count_fasta_divided]] + 1
|
|
100 count_line = 0
|
|
101 whole_seq = []
|
|
102 for line2 in query:
|
|
103 whole_seq.append(line2.rstrip())
|
|
104 count_line += 1
|
|
105 if count_line == stop_line:
|
|
106 break
|
|
107 whole_seq = "".join(whole_seq)
|
|
108 ## create list of starting positions for individual parts of a seq with a step given by a window and overlap
|
|
109 windows_starts = list(range(0, lens_above_win[
|
|
110 count_fasta_divided], step))
|
|
111 ## create list of ending positions (starting pos + window), the last element is the whole seq length
|
|
112 windows_ends = [
|
|
113 x + WIN_DOM
|
|
114 if x + WIN_DOM < lens_above_win[count_fasta_divided] else
|
|
115 lens_above_win[count_fasta_divided] for x in windows_starts
|
|
116 ]
|
|
117 count_part = 1
|
|
118 for start_part, end_part in zip(windows_starts, windows_ends):
|
|
119 seq_part = whole_seq[start_part:end_part]
|
|
120 if count_part == len(windows_starts):
|
|
121 ntf.write("{}_DANTE_PART{}_LAST:{}-{}\n{}\n".format(
|
|
122 line.split(" ")[0], count_part, start_part + 1,
|
|
123 end_part, "\n".join([seq_part[i:i + row_length]
|
|
124 for i in range(0, len(
|
|
125 seq_part), row_length)
|
|
126 ])).encode("utf-8"))
|
|
127 else:
|
|
128 ntf.write("{}_DANTE_PART{}:{}-{}\n{}\n".format(
|
|
129 line.split(" ")[0], count_part, start_part + 1,
|
|
130 end_part, "\n".join([seq_part[i:i + row_length]
|
|
131 for i in range(0, len(
|
|
132 seq_part), row_length)
|
|
133 ])).encode("utf-8"))
|
|
134 count_part += 1
|
|
135 count_fasta_divided += 1
|
|
136 elif line.startswith(">") and line not in divided:
|
|
137 length_seq = seq_ends[below_win[
|
|
138 count_fasta_not_divided]] - seq_starts[below_win[
|
|
139 count_fasta_not_divided]] + 1
|
|
140 ntf.write("{}\n{}".format(line, "".join([query.readline(
|
|
141 ) for x in range(length_seq)])).encode("utf-8"))
|
|
142 count_fasta_not_divided += 1
|
|
143 query_temp = ntf.name
|
|
144 ntf.close()
|
|
145 return query_temp
|
|
146
|
|
147
|
|
148 def domain_annotation(elements, CLASSIFICATION):
|
|
149 ''' Assign protein domain to each hit from protein database '''
|
|
150 domains = []
|
|
151 annotations = []
|
|
152 with open(CLASSIFICATION, "r") as cl_tbl:
|
|
153 annotation = {}
|
|
154 for line in cl_tbl:
|
|
155 record = line.rstrip().split("\t")
|
|
156 annotation[record[0]] = record[1:]
|
|
157 for i in range(len(elements)):
|
|
158 domains.append(elements[i].split("__")[0].split("-")[1])
|
|
159 element_name = "__".join(elements[i].split("__")[1:])
|
|
160 if element_name in annotation.keys():
|
|
161 annotations.append("|".join([elements[i].split("__")[0].split("-")[
|
|
162 1], ("|".join(annotation[element_name]))]))
|
|
163 else:
|
|
164 annotations.append("unknown|unknown")
|
|
165 return annotations
|
|
166
|
|
167
|
|
168 def hits_processing(seq_len, start, end, strand):
|
|
169 ''' Gain hits intervals separately for forward and reverse strand '''
|
|
170 reverse_strand_idx = np.where(strand == "-")[0]
|
|
171 if not reverse_strand_idx.any():
|
|
172 start_pos_plus = start + 1
|
|
173 end_pos_plus = end
|
|
174 regions_plus = list(zip(start_pos_plus, end_pos_plus))
|
|
175 regions_minus = []
|
|
176 else:
|
|
177 reverse_strand_idx = reverse_strand_idx[0]
|
|
178 start_pos_plus = start[0:reverse_strand_idx] + 1
|
|
179 end_pos_plus = end[0:reverse_strand_idx]
|
|
180 start_pos_minus = seq_len[0] - end[reverse_strand_idx:] + 1
|
|
181 end_pos_minus = seq_len[0] - start[reverse_strand_idx:]
|
|
182 regions_plus = list(zip(start_pos_plus, end_pos_plus))
|
|
183 regions_minus = list(zip(start_pos_minus, end_pos_minus))
|
|
184 return reverse_strand_idx, regions_plus, regions_minus
|
|
185
|
|
186
|
|
187 def overlapping_regions(input_data):
|
|
188 ''' Join all overlapping intervals(hits) to clusters (potential domains),
|
|
189 get list of start-end positions of individual hits within the interval,
|
|
190 list of minimus and maximums as well as the indices in the original
|
|
191 sequence_hits structure for the hits belonging to the same clusters '''
|
|
192 if input_data:
|
|
193 sorted_idx, sorted_data = zip(*sorted(
|
|
194 [(index, data) for index, data in enumerate(input_data)],
|
|
195 key=itemgetter(1)))
|
|
196 merged_ends = input_data[sorted_idx[0]][1]
|
|
197 intervals = []
|
|
198 data = []
|
|
199 output_intervals = []
|
|
200 output_data = []
|
|
201 for i, j in zip(sorted_idx, sorted_data):
|
|
202 if input_data[i][0] < merged_ends:
|
|
203 merged_ends = max(input_data[i][1], merged_ends)
|
|
204 intervals.append(i)
|
|
205 data.append(j)
|
|
206 else:
|
|
207 output_intervals.append(intervals)
|
|
208 output_data.append(data)
|
|
209 intervals = []
|
|
210 data = []
|
|
211 intervals.append(i)
|
|
212 data.append(j)
|
|
213 merged_ends = input_data[i][1]
|
|
214 output_intervals.append(intervals)
|
|
215 output_data.append(data)
|
|
216 mins = [x[0][0] for x in output_data]
|
|
217 maxs = [max(x, key=itemgetter(1))[1] for x in output_data]
|
|
218 else:
|
|
219 mins = []
|
|
220 maxs = []
|
|
221 output_intervals = []
|
|
222 output_data = []
|
|
223 return mins, maxs, output_data, output_intervals
|
|
224
|
|
225
|
|
226 def annotations_dict(annotations):
|
|
227 ''' Hash table where annotations of the hits within a clusters are the keys.
|
|
228 Each annotation has serial number assigned which indexes the row in the score_table '''
|
|
229 classes_dict = {classes: idx
|
|
230 for idx, classes in enumerate(set(annotations))}
|
|
231 return classes_dict
|
|
232
|
|
233
|
|
234 def score_table(mins, maxs, data, annotations, scores, CLASSIFICATION):
|
|
235 ''' Score table is created based on the annotations occurance in the cluster.
|
|
236 Matrix axis y corresponds to individual annotations (indexed according to classes_dict),
|
10
|
237 axis x represents positions of analyzed seq in a given cluster.
|
|
238 For every hit within cluster, array of scores on the corresponding position
|
|
239 is recorded to the table in case if the score on certain position is so far the highest
|
0
|
240 for the certain position and certain annotation '''
|
|
241 classes_dict = annotations_dict(annotations)
|
|
242 score_matrix = np.zeros((len(classes_dict), maxs - mins + 1), dtype=int)
|
|
243 count = 0
|
|
244 for item in annotations:
|
|
245 saved_scores = score_matrix[classes_dict[item], data[count][0] - mins:
|
|
246 data[count][1] - mins + 1]
|
|
247 new_scores = [scores[count]] * len(saved_scores)
|
|
248 score_matrix[classes_dict[item], data[count][0] - mins:data[count][
|
|
249 1] - mins + 1] = [max(*pos_score)
|
|
250 for pos_score in zip(saved_scores, new_scores)]
|
|
251 count += 1
|
|
252 return score_matrix, classes_dict
|
|
253
|
|
254
|
|
255 def score_matrix_evaluation(score_matrix, classes_dict, THRESHOLD_SCORE):
|
|
256 ''' Score matrix is evaluated based on each position.
|
|
257 For every position the list of annotations with a score which reaches
|
|
258 certain percentage of the overal best score of the cluster are stored '''
|
|
259 ann_per_reg = []
|
|
260 overal_best_score_reg = max((score_matrix.max(axis=1)))
|
|
261 for position in score_matrix.T:
|
|
262 ## score threshold calculated as a percentage of the OVERALL best score in the cluster
|
|
263 threshold = overal_best_score_reg * THRESHOLD_SCORE / 100
|
|
264 above_th = [idx
|
|
265 for idx, score in enumerate(position)
|
|
266 if position[idx] >= threshold]
|
|
267 ## select unique annotations in one position that are above threshold
|
|
268 ann_per_pos = list(set(
|
|
269 [key for key, value in classes_dict.items() if value in above_th]))
|
|
270 ann_per_reg.append(ann_per_pos)
|
|
271 return ann_per_reg
|
|
272
|
|
273
|
|
274 def group_annot_regs(ann_per_reg):
|
|
275 ''' Get list of domains, annotations, longest common annotations and
|
|
276 counts of positions with certain annotation per regions '''
|
|
277 ## tranform list of lists (potential multiple annotations for every position ) to flat list of all annotations
|
|
278 all_annotations = [item for sublist in ann_per_reg for item in sublist]
|
|
279 unique_annotations = list(set(all_annotations))
|
|
280 ann_pos_counts = [all_annotations.count(x) for x in unique_annotations]
|
|
281 unique_annotations = list(set(
|
|
282 [item for sublist in ann_per_reg for item in sublist]))
|
|
283 domain_type = list(set([annotation.split("|")[0]
|
|
284 for annotation in unique_annotations]))
|
|
285 classification_list = [annotation.split("|")
|
|
286 for annotation in unique_annotations]
|
|
287 ann_substring = "|".join(os.path.commonprefix(classification_list))
|
|
288 domain_type = "/".join(domain_type)
|
|
289 return domain_type, ann_substring, unique_annotations, ann_pos_counts
|
|
290
|
|
291
|
|
292 def best_score(scores, region):
|
|
293 ''' From overlapping intervals take the one with the highest score '''
|
|
294 ## if more hits have the same best score take only the first one
|
|
295 best_idx = region[np.where(scores == max(scores))[0][0]]
|
|
296 best_idx_reg = np.where(scores == max(scores))[0][0]
|
|
297 return best_idx, best_idx_reg
|
|
298
|
|
299
|
|
300 def create_gff3(domain_type, ann_substring, unique_annotations, ann_pos_counts,
|
|
301 dom_start, dom_end, step, best_idx, annotation_best,
|
|
302 db_name_best, db_starts_best, db_ends_best, strand, score,
|
10
|
303 seq_id, db_seq, query_seq, domain_size, positions, gff, consensus):
|
0
|
304 ''' Record obtained information about domain corresponding to individual cluster to common gff file '''
|
|
305 best_start = positions[best_idx][0]
|
|
306 best_end = positions[best_idx][1]
|
|
307 best_score = score[best_idx]
|
|
308 ## proportion of length of the best hit to the whole region length found by base
|
|
309 length_proportion = int((best_end - best_start + 1) /
|
|
310 (dom_end - dom_start + 1) * 100)
|
|
311 db_seq_best = db_seq[best_idx]
|
|
312 query_seq_best = query_seq[best_idx]
|
|
313 domain_size_best = domain_size[best_idx]
|
|
314 [percent_ident, align_similarity, relat_align_len, relat_interrupt,
|
|
315 db_len_proportion
|
|
316 ] = filter_params(db_seq_best, query_seq_best, domain_size_best)
|
|
317 ann_substring = "|".join(ann_substring.split("|")[1:])
|
|
318 annotation_best = "|".join([db_name_best] + annotation_best.split("|")[1:])
|
|
319 if "DANTE_PART" in seq_id:
|
|
320 part = int(seq_id.split("DANTE_PART")[1].split(":")[0].split("_")[0])
|
|
321 dom_start = dom_start + (part - 1) * step
|
|
322 dom_end = dom_end + (part - 1) * step
|
|
323 best_start = best_start + (part - 1) * step
|
|
324 best_end = best_end + (part - 1) * step
|
|
325 if ann_substring is '':
|
|
326 ann_substring = "NONE(Annotations from different classes)"
|
|
327 if len(unique_annotations) > 1:
|
|
328 unique_annotations = ",".join(["{}[{}bp]".format(
|
|
329 ann, pos) for ann, pos in zip(unique_annotations, ann_pos_counts)])
|
|
330 else:
|
|
331 unique_annotations = unique_annotations[0]
|
|
332 if __name__ == '__main__':
|
|
333 SOURCE = configuration.SOURCE_DANTE
|
|
334 else:
|
|
335 SOURCE = configuration.SOURCE_PROFREP
|
|
336 if "/" in domain_type:
|
|
337 gff.write(
|
|
338 "{}\t{}\t{}\t{}\t{}\t.\t{}\t{}\tName={};Final_Classification=Ambiguous_domain;Region_Hits_Classifications_={}\n".format(
|
|
339 seq_id, SOURCE, configuration.DOMAINS_FEATURE, dom_start,
|
|
340 dom_end, strand, configuration.PHASE, domain_type,
|
|
341 unique_annotations))
|
|
342 else:
|
|
343 gff.write(
|
10
|
344 "{}\t{}\t{}\t{}\t{}\t{}\t{}\t{}\tName={};Final_Classification={};Region_Hits_Classifications={};Best_Hit={}:{}-{}[{}percent];Best_Hit_DB_Pos={}:{}of{};DB_Seq={};Region_Seq={};Query_Seq={};Identity={};Similarity={};Relat_Length={};Relat_Interruptions={};Hit_to_DB_Length={}\n".format(
|
0
|
345 seq_id, SOURCE, configuration.DOMAINS_FEATURE, dom_start,
|
|
346 dom_end, best_score, strand, configuration.PHASE, domain_type,
|
|
347 ann_substring, unique_annotations, annotation_best, best_start,
|
|
348 best_end, length_proportion, db_starts_best, db_ends_best,
|
10
|
349 domain_size_best, db_seq_best, consensus, query_seq_best, percent_ident,
|
0
|
350 align_similarity, relat_align_len, relat_interrupt,
|
|
351 db_len_proportion))
|
|
352
|
|
353
|
|
354 def filter_params(db, query, protein_len):
|
|
355 ''' Calculate basic statistics of the quality of the alignment '''
|
|
356 score_dict = alignment_scoring()
|
|
357 num_ident = 0
|
|
358 count_interrupt = 0
|
|
359 count_similarity = 0
|
|
360 alignment_len = 0
|
|
361 for i, j in zip(db.upper(), query.upper()):
|
|
362 if i == j and i != "X":
|
|
363 num_ident += 1
|
|
364 if j == "/" or j == "\\" or j == "*":
|
|
365 count_interrupt += 1
|
|
366 if (i.isalpha() or i == "*") and (j.isalpha() or j == "*"):
|
|
367 if int(score_dict["{}{}".format(i, j)]) > 0:
|
|
368 count_similarity += 1
|
|
369 ## gapless alignment length proportional to the domain protein length
|
|
370 relat_align_len = round((len(db) - db.count("-")) / protein_len, 3)
|
|
371 ## proportional identical bases (except of X) to al.length
|
|
372 align_identity = round(num_ident / len(db), 2)
|
|
373 ## proportional count of positive scores from scoring matrix to al. length
|
|
374 align_similarity = round(count_similarity / len(db), 2)
|
|
375 ## number of interruptions per 100 bp
|
|
376 relat_interrupt = round(count_interrupt / math.ceil((len(query) / 100)), 2)
|
|
377 ## Proportion of alignment to the original length of protein domain from database (indels included)
|
|
378 db_len_proportion = round(len(db) / protein_len, 2)
|
|
379 return align_identity, align_similarity, relat_align_len, relat_interrupt, db_len_proportion
|
|
380
|
|
381
|
|
382 def line_generator(tab_pipe, maf_pipe, start):
|
|
383 ''' Yield individual lines of LASTAL stdout for single sequence '''
|
|
384 if hasattr(line_generator, "dom"):
|
|
385 seq_id = line_generator.dom.split("\t")[6]
|
|
386 yield line_generator.dom.encode("utf-8")
|
|
387 del line_generator.dom
|
|
388 line_tab = ""
|
|
389 for line_tab in tab_pipe:
|
|
390 line_tab = line_tab.decode("utf-8")
|
|
391 if not line_tab.startswith('#'):
|
|
392 if start:
|
|
393 if not ('seq_id' in locals() and
|
|
394 seq_id != line_tab.split("\t")[6]):
|
|
395 seq_id = line_tab.split("\t")[6]
|
|
396 start = False
|
|
397 line_maf = [maf_pipe.readline() for line_count in range(4)]
|
|
398 db_seq = line_maf[1].decode("utf-8").rstrip().split(" ")[-1]
|
|
399 alignment_seq = line_maf[2].decode("utf-8").rstrip().split(" ")[-1]
|
|
400 line = "{}\t{}\t{}".format(line_tab, db_seq, alignment_seq)
|
|
401 line_id = line.split("\t")[6]
|
|
402 if seq_id != line_id:
|
|
403 line_generator.dom = line
|
|
404 return
|
|
405 else:
|
|
406 yield line.encode("utf-8")
|
|
407 else:
|
|
408 maf_pipe.readline()
|
|
409 if line_tab == "":
|
|
410 raise RuntimeError
|
|
411 else:
|
|
412 return
|
|
413
|
|
414
|
|
415 def get_version(path, LAST_DB):
|
5
|
416 '''Return version is run from git repository '''
|
|
417 try:
|
|
418 branch = subprocess.check_output("git rev-parse --abbrev-ref HEAD",
|
|
419 shell=True,
|
|
420 cwd=path).decode('ascii').strip()
|
|
421 shorthash = subprocess.check_output("git log --pretty=format:'%h' -n 1 ",
|
|
422 shell=True,
|
|
423 cwd=path).decode('ascii').strip()
|
|
424 revcount = len(subprocess.check_output("git log --oneline",
|
|
425 shell=True,
|
|
426 cwd=path).decode('ascii').split())
|
|
427 version_string = (
|
|
428 "##-----------------------------------------------\n"
|
|
429 "##PIPELINE VERSION : "
|
|
430 "{branch}-rv-{revcount}({shorthash})\n"
|
|
431 "##PROTEIN DATABASE VERSION : {PD}\n"
|
|
432 "##-----------------------------------------------\n").format(
|
|
433 branch=branch,
|
|
434 shorthash=shorthash,
|
|
435 revcount=revcount,
|
|
436 PD=os.path.basename(LAST_DB))
|
|
437 except:
|
|
438 version_string = (
|
|
439 "##-----------------------------------------------\n"
|
|
440 "##PROTEIN DATABASE VERSION : {PD}\n"
|
|
441 "##-----------------------------------------------\n").format(
|
|
442 PD=os.path.basename(LAST_DB)
|
|
443 )
|
|
444
|
0
|
445 return version_string
|
|
446
|
|
447
|
|
448 def write_info(dom_gff_tmp, version_string):
|
|
449 dom_gff_tmp.write("{}\n".format(configuration.HEADER_GFF))
|
|
450 dom_gff_tmp.write(version_string)
|
|
451
|
|
452
|
|
453 def domain_search(QUERY, LAST_DB, CLASSIFICATION, OUTPUT_DOMAIN,
|
10
|
454 THRESHOLD_SCORE, WIN_DOM, OVERLAP_DOM, SCORING_MATRIX):
|
0
|
455 ''' Search for protein domains using our protein database and external tool LAST,
|
|
456 stdout is parsed in real time and hits for a single sequence undergo further processing
|
10
|
457 - tabular format(TAB) to get info about position, score, orientation
|
0
|
458 - MAF format to gain alignment and original sequence
|
|
459 '''
|
|
460
|
|
461 step = WIN_DOM - OVERLAP_DOM
|
|
462 [headers, above_win, below_win, lens_above_win, seq_starts, seq_ends
|
|
463 ] = characterize_fasta(QUERY, WIN_DOM)
|
|
464 query_temp = split_fasta(QUERY, WIN_DOM, step, headers, above_win,
|
|
465 below_win, lens_above_win, seq_starts, seq_ends)
|
|
466
|
|
467 ## TAB output contains all the alignment scores, positions, strands...
|
10
|
468 lastal_columns = {
|
|
469 "BL80" : ("score, name_db, start_db, al_size_db, strand_db,"
|
|
470 " seq_size_db, name_q, start_q, al_size_q, strand_q, seq_size_q,"
|
|
471 " block1, block2, block3, db_seq, q_seq"),
|
|
472 "BL62" : ("score, name_db, start_db, al_size_db, strand_db,"
|
|
473 " seq_size_db, name_q, start_q, al_size_q, strand_q,"
|
|
474 " seq_size_q, block1, block2, block3, db_seq, q_seq"),
|
|
475 "MIQS" : ("score, name_db, start_db, al_size_db, strand_db,"
|
|
476 " seq_size_db, name_q, start_q, al_size_q, strand_q,"
|
|
477 " seq_size_q, block1, db_seq, q_seq"),
|
|
478 }
|
0
|
479 tab = subprocess.Popen(
|
10
|
480 "lastal -F15 {} {} -L 10 -m 70 -p {} -e 80 -f TAB".format(LAST_DB,
|
|
481 query_temp,
|
|
482 SCORING_MATRIX),
|
0
|
483 stdout=subprocess.PIPE,
|
|
484 shell=True)
|
|
485 ## MAF output contains alignment sequences
|
|
486 maf = subprocess.Popen(
|
10
|
487 "lastal -F15 {} {} -L 10 -m 70 -p {} -e 80 -f MAF".format(LAST_DB,
|
|
488 query_temp,
|
|
489 SCORING_MATRIX),
|
0
|
490 stdout=subprocess.PIPE,
|
|
491 shell=True)
|
|
492
|
|
493 tab_pipe = tab.stdout
|
|
494 maf_pipe = maf.stdout
|
|
495 maf_pipe.readline()
|
|
496
|
|
497 seq_ids = []
|
|
498 dom_tmp = NamedTemporaryFile(delete=False)
|
|
499 with open(dom_tmp.name, "w") as dom_gff_tmp:
|
|
500 path = os.path.dirname(os.path.realpath(__file__))
|
|
501 version_string = get_version(path, LAST_DB)
|
|
502 write_info(dom_gff_tmp, version_string)
|
|
503 gff = open(dom_tmp.name, "a")
|
|
504 start = True
|
|
505 while True:
|
|
506 try:
|
|
507 with warnings.catch_warnings():
|
|
508 warnings.simplefilter("ignore")
|
|
509 sequence_hits = np.genfromtxt(
|
|
510 line_generator(tab_pipe, maf_pipe, start),
|
10
|
511 names=lastal_columns[SCORING_MATRIX],
|
|
512 usecols=("score, name_q, start_q, al_size_q,"
|
|
513 " strand_q, seq_size_q, name_db, db_seq,"
|
|
514 " q_seq, seq_size_db, start_db, al_size_db"),
|
0
|
515 dtype=None,
|
|
516 comments=None)
|
|
517 except RuntimeError:
|
|
518 break
|
|
519 ## if there are no domains found
|
|
520 if sequence_hits.size is 0:
|
|
521 gff.write("##NO DOMAINS")
|
|
522 return [], [], [], []
|
|
523
|
|
524 ############# PARSING LASTAL OUTPUT ############################
|
|
525 sequence_hits = np.atleast_1d(sequence_hits)
|
|
526 score = sequence_hits['score'].astype("int")
|
|
527 seq_id = sequence_hits['name_q'][0].astype("str")
|
|
528 start_hit = sequence_hits['start_q'].astype("int")
|
|
529 end_hit = start_hit + sequence_hits['al_size_q'].astype("int")
|
|
530 strand = sequence_hits['strand_q'].astype("str")
|
|
531 seq_len = sequence_hits['seq_size_q'].astype("int")
|
|
532 domain_db = sequence_hits['name_db'].astype("str")
|
|
533 db_seq = sequence_hits['db_seq'].astype("str")
|
|
534 query_seq = sequence_hits['q_seq'].astype("str")
|
|
535 domain_size = sequence_hits['seq_size_db'].astype("int")
|
|
536 db_start = sequence_hits['start_db'].astype("int") + 1
|
|
537 db_end = sequence_hits['start_db'].astype("int") + sequence_hits[
|
|
538 'al_size_db'].astype("int")
|
|
539
|
|
540 [reverse_strand_idx, positions_plus, positions_minus
|
|
541 ] = hits_processing(seq_len, start_hit, end_hit, strand)
|
|
542 strand_gff = "+"
|
|
543 [mins_plus, maxs_plus, data_plus, indices_plus
|
|
544 ] = overlapping_regions(positions_plus)
|
|
545 [mins_minus, maxs_minus, data_minus, indices_minus
|
|
546 ] = overlapping_regions(positions_minus)
|
|
547 positions = positions_plus + positions_minus
|
|
548 indices_overal = indices_plus + [x + reverse_strand_idx
|
|
549 for x in indices_minus]
|
|
550 mins = mins_plus + mins_minus
|
|
551 maxs = maxs_plus + maxs_minus
|
|
552 data = data_plus + data_minus
|
|
553 ## process every region (cluster) of overlapping hits sequentially
|
|
554 count_region = 0
|
|
555 for region in indices_overal:
|
|
556 db_names = domain_db[np.array(region)]
|
|
557 db_starts = db_start[np.array(region)]
|
|
558 db_ends = db_end[np.array(region)]
|
|
559 scores = score[np.array(region)]
|
10
|
560 regions_above_threshold = [
|
|
561 region[i]
|
|
562 for i, _ in enumerate(region)
|
|
563 if max(scores) / 100 * THRESHOLD_SCORE < scores[i]
|
|
564 ]
|
|
565 ## sort by score first:
|
|
566 consensus = get_full_translation(
|
|
567 translation_alignments(
|
|
568 query_seq=sortby(query_seq[regions_above_threshold], score[regions_above_threshold], True),
|
|
569 start_hit=sortby(start_hit[regions_above_threshold], score[regions_above_threshold], True),
|
|
570 end_hit=sortby(end_hit[regions_above_threshold], score[regions_above_threshold], True))
|
|
571 )
|
|
572
|
0
|
573 annotations = domain_annotation(db_names, CLASSIFICATION)
|
|
574 [score_matrix, classes_dict] = score_table(
|
|
575 mins[count_region], maxs[count_region], data[count_region],
|
|
576 annotations, scores, CLASSIFICATION)
|
|
577 ann_per_reg = score_matrix_evaluation(score_matrix, classes_dict,
|
|
578 THRESHOLD_SCORE)
|
|
579 [domain_type, ann_substring, unique_annotations, ann_pos_counts
|
|
580 ] = group_annot_regs(ann_per_reg)
|
|
581 [best_idx, best_idx_reg] = best_score(scores, region)
|
|
582 annotation_best = annotations[best_idx_reg]
|
|
583 db_name_best = db_names[best_idx_reg]
|
|
584 db_starts_best = db_starts[best_idx_reg]
|
|
585 db_ends_best = db_ends[best_idx_reg]
|
|
586 if count_region == len(indices_plus):
|
|
587 strand_gff = "-"
|
|
588 create_gff3(domain_type, ann_substring, unique_annotations,
|
10
|
589 ann_pos_counts, min(start_hit[regions_above_threshold])-1,
|
|
590 max(end_hit[regions_above_threshold]),
|
0
|
591 step, best_idx, annotation_best, db_name_best,
|
|
592 db_starts_best, db_ends_best, strand_gff, score,
|
10
|
593 seq_id, db_seq, query_seq, domain_size, positions, gff, consensus)
|
0
|
594 count_region += 1
|
|
595 seq_ids.append(seq_id)
|
|
596 os.unlink(query_temp)
|
|
597 gff.close()
|
|
598 dom_tmp.close()
|
|
599 ## if any sequence from input data was split into windows, merge and adjust the data from individual windows
|
|
600 if any("DANTE_PART" in x for x in seq_ids):
|
|
601 adjust_gff(OUTPUT_DOMAIN, dom_tmp.name, WIN_DOM, OVERLAP_DOM, step)
|
|
602 ## otherwise use the temporary output as the final domains gff
|
|
603 else:
|
|
604 shutil.copy2(dom_tmp.name, OUTPUT_DOMAIN)
|
|
605 os.unlink(dom_tmp.name)
|
|
606
|
10
|
607 def sortby(a, by, reverse=False):
|
|
608 ''' sort according values in the by list '''
|
|
609 a_sorted = [i[0] for i in
|
|
610 sorted(
|
|
611 zip(a, by),
|
|
612 key=lambda i: i[1],
|
|
613 reverse=reverse
|
|
614 )]
|
|
615 return a_sorted
|
|
616
|
|
617
|
|
618 def a2nnn(s):
|
|
619 s1 = "".join([c if c in ['/', '\\'] else c + c + c
|
|
620 for c in s.replace("-", "")])
|
|
621 # collapse frameshifts (/)
|
|
622 s2 = re.sub("[a-zA-Z*]{2}//[a-zA-Z*]{2}", "//", s1)
|
|
623 s3 = re.sub("[a-zA-Z*]/[a-zA-Z*]", "/", s2)
|
|
624 return (s3)
|
|
625
|
|
626
|
|
627
|
|
628 def rle(s):
|
|
629 '''run length encoding but max is set to 3 (codon)'''
|
|
630 prev = ""
|
|
631 count = 1
|
|
632 char = []
|
|
633 length = []
|
|
634 L = 0
|
|
635 for n in s:
|
|
636 if n == prev and count < (3 - L):
|
|
637 count += 1
|
|
638 else:
|
|
639 char.append(prev)
|
|
640 length.append(count)
|
|
641 L = 1 if prev == "/" else 0
|
|
642 prev = n
|
|
643 count = 1
|
|
644 char.append(prev)
|
|
645 length.append(count)
|
|
646 sequence = char[1:]
|
|
647 return sequence, length[1:]
|
|
648
|
|
649 def get_full_translation(translations):
|
|
650 '''get one full length translation from multiple partial
|
|
651 aligned translation '''
|
|
652 # find minimal set of alignements
|
|
653 minimal_set = []
|
|
654 not_filled_prev = len(translations[0])
|
|
655 for s in translations:
|
|
656 minimal_set.append(s)
|
|
657 # check defined position - is there only '-' character?
|
|
658 not_filled = sum([set(i) == {"-"} for i in zip(*minimal_set)])
|
|
659 if not_filled == 0:
|
|
660 break
|
|
661 if not_filled == not_filled_prev:
|
|
662 # last added sequence did not improve coverage - remove it.
|
|
663 minimal_set.pop()
|
|
664 not_filled_prev = not_filled
|
|
665 # merge translations
|
|
666 final_translation = minimal_set[0]
|
|
667 # record positions of joins to correct frameshifts reportings
|
|
668 position_of_joins = set()
|
|
669 position_of_joins_rle = set()
|
|
670 if len(minimal_set) > 1: # translation must be merged
|
|
671 for s in minimal_set[1:]:
|
|
672 s1 = re.search(r"-\w", final_translation)
|
|
673 s2 = re.search(r"\w-", final_translation)
|
|
674 if s1:
|
|
675 position_of_joins.add(s1.start())
|
|
676 if s2:
|
|
677 position_of_joins.add((s2.end() - 1))
|
|
678 final_translation = "".join(
|
|
679 [b if a == "-" else a for a, b in zip(final_translation, s)])
|
|
680 translation_rle = rle(final_translation)
|
|
681 cumsumed_positions = np.cumsum(translation_rle[1])
|
|
682 for p in position_of_joins:
|
|
683 position_of_joins_rle.add(sum(cumsumed_positions <= p))
|
|
684 # insert /\ when necessary
|
|
685 for p in position_of_joins_rle:
|
|
686 if translation_rle[0][p] not in ['/',"//","\\", "\\\\"]:
|
|
687 if translation_rle[1][p] == 2:
|
|
688 translation_rle[0][p] = translation_rle[0][p] + "/"
|
|
689 if translation_rle[1][p] == 1:
|
|
690 translation_rle[0][p] = "\\"
|
|
691 consensus = "".join(translation_rle[0])
|
|
692 return consensus
|
|
693
|
|
694
|
|
695 # helper function for debugging
|
|
696 def translation_alignments(query_seq, start_hit, end_hit):
|
|
697 pstart = min(start_hit)
|
|
698 pend = max(end_hit)
|
|
699 nnn = list()
|
|
700 for s, start, end in zip(query_seq, start_hit, end_hit):
|
|
701 nnn.append("-" * (start - pstart) + a2nnn(s) + "-" * (pend - end))
|
|
702 return (nnn)
|
|
703
|
|
704
|
0
|
705
|
|
706 def adjust_gff(OUTPUT_DOMAIN, gff, WIN_DOM, OVERLAP_DOM, step):
|
|
707 ''' Original gff file is adjusted in case of containing cut parts
|
|
708 - for consecutive sequences overlap is divided to half with first half
|
|
709 of records(domains) belonging to the first sequence and second to the following one.
|
|
710 Duplicate domains going through the middle of the overlap are removed.
|
|
711 First and the last part (marked as LAST) of a certain sequence are
|
|
712 handled separately as the are overlapped from one side only '''
|
|
713
|
|
714 seq_id_all = []
|
|
715 class_dict = defaultdict(int)
|
|
716 seen = set()
|
|
717 with open(OUTPUT_DOMAIN, "w") as adjusted_gff:
|
|
718 with open(gff, "r") as primary_gff:
|
|
719 start = True
|
|
720 for line in primary_gff:
|
|
721 if line.startswith("#"):
|
|
722 adjusted_gff.write(line)
|
|
723 else:
|
|
724 split_line = line.split("\t")
|
|
725 classification = split_line[-1].split(";")[1].split("=")[1]
|
|
726 if start:
|
|
727 seq_id_all.append(split_line[0].split("_DANTE_PART")[
|
|
728 0])
|
|
729 start = False
|
|
730 seq_id = split_line[0].split("_DANTE_PART")[0]
|
|
731 if "DANTE_PART" in line:
|
|
732 line_without_id = "\t".join(split_line[1:])
|
|
733 part = int(split_line[0].split("_DANTE_PART")[1].split(
|
|
734 ":")[0].split("_")[0])
|
|
735 if seq_id != seq_id_all[-1]:
|
|
736 seq_id_all.append(seq_id)
|
|
737
|
|
738 ## first part of the sequence
|
|
739 if part == 1:
|
|
740 cut_end = WIN_DOM - OVERLAP_DOM / 2
|
|
741 if int(split_line[3]) <= cut_end <= int(split_line[
|
|
742 4]):
|
|
743 if line_without_id not in seen:
|
|
744 adjusted_gff.write("{}\t{}".format(
|
|
745 seq_id, line_without_id))
|
|
746 class_dict[classification] += 1
|
|
747 seen.add(line_without_id)
|
|
748 elif int(split_line[4]) < cut_end:
|
|
749 adjusted_gff.write("{}\t{}".format(
|
|
750 seq_id, line_without_id))
|
|
751 class_dict[classification] += 1
|
|
752
|
|
753 ## last part of the sequence
|
|
754 elif "LAST" in split_line[0]:
|
|
755 cut_start = OVERLAP_DOM / 2 + (part - 1) * step
|
|
756 if int(split_line[3]) <= cut_start <= int(
|
|
757 split_line[4]):
|
|
758 if line_without_id not in seen:
|
|
759 adjusted_gff.write("{}\t{}".format(
|
|
760 seq_id, line_without_id))
|
|
761 class_dict[classification] += 1
|
|
762 seen.add(line_without_id)
|
|
763 elif int(split_line[3]) > cut_start:
|
|
764 adjusted_gff.write("{}\t{}".format(
|
|
765 seq_id, line_without_id))
|
|
766 class_dict[classification] += 1
|
|
767
|
|
768 ## middle part of the sequence
|
|
769 else:
|
|
770 cut_start = OVERLAP_DOM / 2 + (part - 1) * step
|
|
771 cut_end = WIN_DOM - OVERLAP_DOM / 2 + (part -
|
|
772 1) * step
|
|
773 if int(split_line[3]) <= cut_start <= int(
|
|
774 split_line[4]) or int(split_line[
|
|
775 3]) <= cut_end <= int(split_line[4]):
|
|
776 if line_without_id not in seen:
|
|
777 adjusted_gff.write("{}\t{}".format(
|
|
778 seq_id, line_without_id))
|
|
779 class_dict[classification] += 1
|
|
780 seen.add(line_without_id)
|
|
781 elif int(split_line[3]) > cut_start and int(
|
|
782 split_line[4]) < cut_end:
|
|
783 adjusted_gff.write("{}\t{}".format(
|
|
784 seq_id, line_without_id))
|
|
785 class_dict[classification] += 1
|
|
786 ## not divived
|
|
787 else:
|
|
788 if seq_id != seq_id_all[-1]:
|
|
789 seq_id_all.append(seq_id)
|
|
790 adjusted_gff.write(line)
|
|
791 class_dict[classification] += 1
|
|
792
|
|
793
|
|
794 def main(args):
|
|
795
|
|
796 t = time.time()
|
|
797
|
|
798 QUERY = args.query
|
|
799 LAST_DB = args.protein_database
|
|
800 CLASSIFICATION = args.classification
|
|
801 OUTPUT_DOMAIN = args.domain_gff
|
|
802 NEW_LDB = args.new_ldb
|
|
803 OUTPUT_DIR = args.output_dir
|
|
804 THRESHOLD_SCORE = args.threshold_score
|
|
805 WIN_DOM = args.win_dom
|
|
806 OVERLAP_DOM = args.overlap_dom
|
10
|
807 SCORING_MATRIX = args.scoring_matrix
|
|
808 configuration.SC_MATRIX = configuration.SC_MATRIX_SKELETON.format(SCORING_MATRIX)
|
0
|
809
|
|
810 if OUTPUT_DOMAIN is None:
|
|
811 OUTPUT_DOMAIN = configuration.DOMAINS_GFF
|
|
812 if os.path.isdir(LAST_DB):
|
|
813 LAST_DB = os.path.join(LAST_DB, configuration.LAST_DB_FILE)
|
|
814 if os.path.isdir(CLASSIFICATION):
|
|
815 CLASSIFICATION = os.path.join(CLASSIFICATION, configuration.CLASS_FILE)
|
|
816
|
|
817 if NEW_LDB:
|
|
818 subprocess.call("lastdb -p -cR01 {} {}".format(LAST_DB, LAST_DB),
|
|
819 shell=True)
|
|
820
|
|
821 if OUTPUT_DIR and not os.path.exists(OUTPUT_DIR):
|
|
822 os.makedirs(OUTPUT_DIR)
|
|
823
|
|
824 if not os.path.isabs(OUTPUT_DOMAIN):
|
|
825 if OUTPUT_DIR is None:
|
|
826 OUTPUT_DIR = configuration.TMP
|
|
827 if not os.path.exists(OUTPUT_DIR):
|
|
828 os.makedirs(OUTPUT_DIR)
|
|
829 OUTPUT_DOMAIN = os.path.join(OUTPUT_DIR,
|
|
830 os.path.basename(OUTPUT_DOMAIN))
|
|
831 domain_search(QUERY, LAST_DB, CLASSIFICATION, OUTPUT_DOMAIN,
|
10
|
832 THRESHOLD_SCORE, WIN_DOM, OVERLAP_DOM, SCORING_MATRIX)
|
0
|
833
|
|
834 print("ELAPSED_TIME_DOMAINS = {} s".format(time.time() - t))
|
|
835
|
|
836
|
|
837 if __name__ == "__main__":
|
|
838 import argparse
|
|
839 from argparse import RawDescriptionHelpFormatter
|
|
840
|
|
841 class CustomFormatter(argparse.ArgumentDefaultsHelpFormatter,
|
|
842 argparse.RawDescriptionHelpFormatter):
|
|
843 pass
|
|
844
|
|
845 parser = argparse.ArgumentParser(
|
|
846 description=
|
|
847 '''Script performs similarity search on given DNA sequence(s) in (multi)fasta against our protein domains database of all Transposable element for certain group of organisms (Viridiplantae or Metazoans). Domains are subsequently annotated and classified - in case certain domain has multiple annotations assigned, classifation is derived from the common classification level of all of them. Domains search is accomplished engaging LASTAL alignment tool.
|
|
848
|
|
849 DEPENDENCIES:
|
|
850 - python 3.4 or higher with packages:
|
|
851 -numpy
|
|
852 - lastal 744 or higher [http://last.cbrc.jp/]
|
|
853 - configuration.py module
|
|
854
|
|
855 EXAMPLE OF USAGE:
|
|
856
|
|
857 ./protein_domains_pd.py -q PATH_TO_INPUT_SEQ -pdb PATH_TO_PROTEIN_DB -cs PATH_TO_CLASSIFICATION_FILE
|
|
858
|
|
859 When running for the first time with a new database use -nld option allowing lastal to create indexed database files:
|
|
860
|
|
861 -nld True
|
|
862
|
|
863 ''',
|
|
864 epilog="""""",
|
|
865 formatter_class=CustomFormatter)
|
|
866
|
|
867 requiredNamed = parser.add_argument_group('required named arguments')
|
|
868 requiredNamed.add_argument(
|
|
869 "-q",
|
|
870 "--query",
|
|
871 type=str,
|
|
872 required=True,
|
|
873 help=
|
|
874 'input DNA sequence to search for protein domains in a fasta format. Multifasta format allowed.')
|
|
875 requiredNamed.add_argument('-pdb',
|
|
876 "--protein_database",
|
|
877 type=str,
|
|
878 required=True,
|
|
879 help='protein domains database file')
|
|
880 requiredNamed.add_argument('-cs',
|
|
881 '--classification',
|
|
882 type=str,
|
|
883 required=True,
|
|
884 help='protein domains classification file')
|
|
885 parser.add_argument("-oug",
|
|
886 "--domain_gff",
|
|
887 type=str,
|
|
888 help="output domains gff format")
|
|
889 parser.add_argument(
|
|
890 "-nld",
|
|
891 "--new_ldb",
|
|
892 type=str,
|
|
893 default=False,
|
|
894 help=
|
|
895 "create indexed database files for lastal in case of working with new protein db")
|
|
896 parser.add_argument(
|
|
897 "-dir",
|
|
898 "--output_dir",
|
|
899 type=str,
|
|
900 help="specify if you want to change the output directory")
|
|
901 parser.add_argument(
|
10
|
902 "-M",
|
|
903 "--scoring_matrix",
|
|
904 type=str,
|
|
905 default="BL80",
|
|
906 choices=['BL80', 'BL62', 'MIQS'],
|
|
907 help="specify scoring matrix to use for similarity search (BL80, BL62, MIQS)")
|
|
908 parser.add_argument(
|
0
|
909 "-thsc",
|
|
910 "--threshold_score",
|
|
911 type=int,
|
|
912 default=80,
|
|
913 help=
|
|
914 "percentage of the best score in the cluster to be tolerated when assigning annotations per base")
|
|
915 parser.add_argument(
|
|
916 "-wd",
|
|
917 "--win_dom",
|
|
918 type=int,
|
|
919 default=10000000,
|
|
920 help="window to process large input sequences sequentially")
|
|
921 parser.add_argument("-od",
|
|
922 "--overlap_dom",
|
|
923 type=int,
|
|
924 default=10000,
|
|
925 help="overlap of sequences in two consecutive windows")
|
|
926
|
|
927 args = parser.parse_args()
|
|
928 main(args)
|