1 | #!/usr/bin/env python |
---|
2 | #Guruprasad Ananda |
---|
3 | """ |
---|
4 | This tool computes microsatellite mutability for the orthologous microsatellites fetched from 'Extract Orthologous Microsatellites from pair-wise alignments' tool. |
---|
5 | """ |
---|
6 | from galaxy import eggs |
---|
7 | import sys, string, re, commands, tempfile, os, fileinput |
---|
8 | from galaxy.tools.util.galaxyops import * |
---|
9 | from bx.intervals.io import * |
---|
10 | from bx.intervals.operations import quicksect |
---|
11 | |
---|
12 | fout = open(sys.argv[2],'w') |
---|
13 | p_group = int(sys.argv[3]) #primary "group-by" feature |
---|
14 | p_bin_size = int(sys.argv[4]) |
---|
15 | s_group = int(sys.argv[5]) #sub-group by feature |
---|
16 | s_bin_size = int(sys.argv[6]) |
---|
17 | mono_threshold = 9 |
---|
18 | non_mono_threshold = 4 |
---|
19 | p_group_cols = [p_group, p_group+7] |
---|
20 | s_group_cols = [s_group, s_group+7] |
---|
21 | num_generations = int(sys.argv[7]) |
---|
22 | region = sys.argv[8] |
---|
23 | int_file = sys.argv[9] |
---|
24 | if int_file != "None": #User has specified an interval file |
---|
25 | try: |
---|
26 | fint = open(int_file, 'r') |
---|
27 | dbkey_i = sys.argv[10] |
---|
28 | chr_col_i, start_col_i, end_col_i, strand_col_i = parse_cols_arg( sys.argv[11] ) |
---|
29 | except: |
---|
30 | stop_err("Unable to open input Interval file") |
---|
31 | |
---|
32 | def stop_err(msg): |
---|
33 | sys.stderr.write(msg) |
---|
34 | sys.exit() |
---|
35 | |
---|
36 | def reverse_complement(text): |
---|
37 | DNA_COMP = string.maketrans( "ACGTacgt", "TGCAtgca" ) |
---|
38 | comp = [ch for ch in text.translate(DNA_COMP)] |
---|
39 | comp.reverse() |
---|
40 | return "".join(comp) |
---|
41 | |
---|
42 | def get_unique_elems(elems): |
---|
43 | seen=set() |
---|
44 | return[x for x in elems if x not in seen and not seen.add(x)] |
---|
45 | |
---|
46 | def get_binned_lists(uniqlist, binsize): |
---|
47 | binnedlist=[] |
---|
48 | uniqlist.sort() |
---|
49 | start = int(uniqlist[0]) |
---|
50 | bin_ind=0 |
---|
51 | l_ind=0 |
---|
52 | binnedlist.append([]) |
---|
53 | while l_ind < len(uniqlist): |
---|
54 | elem = int(uniqlist[l_ind]) |
---|
55 | if elem in range(start,start+binsize): |
---|
56 | binnedlist[bin_ind].append(elem) |
---|
57 | else: |
---|
58 | start += binsize |
---|
59 | bin_ind += 1 |
---|
60 | binnedlist.append([]) |
---|
61 | binnedlist[bin_ind].append(elem) |
---|
62 | l_ind += 1 |
---|
63 | return binnedlist |
---|
64 | |
---|
65 | def fetch_weight(H,C,t): |
---|
66 | if (H-(C-H)) < t: |
---|
67 | return 2.0 |
---|
68 | else: |
---|
69 | return 1.0 |
---|
70 | |
---|
71 | def mutabilityEstimator(repeats1,repeats2,thresholds): |
---|
72 | mut_num = 0.0 #Mutability Numerator |
---|
73 | mut_den = 0.0 #Mutability denominator |
---|
74 | for ind,H in enumerate(repeats1): |
---|
75 | C = repeats2[ind] |
---|
76 | t = thresholds[ind] |
---|
77 | w = fetch_weight(H,C,t) |
---|
78 | mut_num += ((H-C)*(H-C)*w) |
---|
79 | mut_den += w |
---|
80 | return [mut_num, mut_den] |
---|
81 | |
---|
82 | def output_writer(blk, blk_lines): |
---|
83 | global winspecies, speciesind |
---|
84 | all_elems_1=[] |
---|
85 | all_elems_2=[] |
---|
86 | all_s_elems_1=[] |
---|
87 | all_s_elems_2=[] |
---|
88 | for bline in blk_lines: |
---|
89 | if not(bline): |
---|
90 | continue |
---|
91 | items = bline.split('\t') |
---|
92 | seq1 = items[1] |
---|
93 | start1 = items[2] |
---|
94 | end1 = items[3] |
---|
95 | seq2 = items[8] |
---|
96 | start2 = items[9] |
---|
97 | end2 = items[10] |
---|
98 | if p_group_cols[0] == 6: |
---|
99 | items[p_group_cols[0]] = int(items[p_group_cols[0]]) |
---|
100 | items[p_group_cols[1]] = int(items[p_group_cols[1]]) |
---|
101 | if s_group_cols[0] == 6: |
---|
102 | items[s_group_cols[0]] = int(items[s_group_cols[0]]) |
---|
103 | items[s_group_cols[1]] = int(items[s_group_cols[1]]) |
---|
104 | all_elems_1.append(items[p_group_cols[0]]) #primary col elements for species 1 |
---|
105 | all_elems_2.append(items[p_group_cols[1]]) #primary col elements for species 2 |
---|
106 | if s_group_cols[0] != -1: #sub-group is not None |
---|
107 | all_s_elems_1.append(items[s_group_cols[0]]) #secondary col elements for species 1 |
---|
108 | all_s_elems_2.append(items[s_group_cols[1]]) #secondary col elements for species 2 |
---|
109 | uniq_elems_1 = get_unique_elems(all_elems_1) |
---|
110 | uniq_elems_2 = get_unique_elems(all_elems_2) |
---|
111 | if s_group_cols[0] != -1: |
---|
112 | uniq_s_elems_1 = get_unique_elems(all_s_elems_1) |
---|
113 | uniq_s_elems_2 = get_unique_elems(all_s_elems_2) |
---|
114 | mut1={} |
---|
115 | mut2={} |
---|
116 | count1 = {} |
---|
117 | count2 = {} |
---|
118 | """ |
---|
119 | if p_group_cols[0] == 7: #i.e. the option chosen is group-by unit(AG, GTC, etc) |
---|
120 | uniq_elems_1 = get_unique_units(j.sort(lambda x, y: len(x)-len(y))) |
---|
121 | """ |
---|
122 | if p_group_cols[0] == 6: #i.e. the option chosen is group-by repeat number. |
---|
123 | uniq_elems_1 = get_binned_lists(uniq_elems_1,p_bin_size) |
---|
124 | uniq_elems_2 = get_binned_lists(uniq_elems_2,p_bin_size) |
---|
125 | |
---|
126 | if s_group_cols[0] == 6: #i.e. the option chosen is subgroup-by repeat number. |
---|
127 | uniq_s_elems_1 = get_binned_lists(uniq_s_elems_1,s_bin_size) |
---|
128 | uniq_s_elems_2 = get_binned_lists(uniq_s_elems_2,s_bin_size) |
---|
129 | |
---|
130 | for pitem1 in uniq_elems_1: |
---|
131 | #repeats1 = [] |
---|
132 | #repeats2 = [] |
---|
133 | thresholds = [] |
---|
134 | if s_group_cols[0] != -1: #Sub-group by feature is not None |
---|
135 | for sitem1 in uniq_s_elems_1: |
---|
136 | repeats1 = [] |
---|
137 | repeats2 = [] |
---|
138 | if type(sitem1) == type(''): |
---|
139 | sitem1 = sitem1.strip() |
---|
140 | for bline in blk_lines: |
---|
141 | belems = bline.split('\t') |
---|
142 | if type(pitem1) == list: |
---|
143 | if p_group_cols[0] == 6: |
---|
144 | belems[p_group_cols[0]] = int(belems[p_group_cols[0]]) |
---|
145 | if belems[p_group_cols[0]] in pitem1: |
---|
146 | if belems[s_group_cols[0]]==sitem1: |
---|
147 | repeats1.append(int(belems[6])) |
---|
148 | repeats2.append(int(belems[13])) |
---|
149 | if belems[4] == 'mononucleotide': |
---|
150 | thresholds.append(mono_threshold) |
---|
151 | else: |
---|
152 | thresholds.append(non_mono_threshold) |
---|
153 | mut1[str(pitem1)+'\t'+str(sitem1)]=mutabilityEstimator(repeats1,repeats2,thresholds) |
---|
154 | if region == 'align': |
---|
155 | count1[str(pitem1)+'\t'+str(sitem1)]=min(sum(repeats1),sum(repeats2)) |
---|
156 | else: |
---|
157 | if winspecies == 1: |
---|
158 | count1["%s\t%s" %(pitem1,sitem1)]=sum(repeats1) |
---|
159 | elif winspecies == 2: |
---|
160 | count1["%s\t%s" %(pitem1,sitem1)]=sum(repeats2) |
---|
161 | else: |
---|
162 | if type(sitem1) == list: |
---|
163 | if s_group_cols[0] == 6: |
---|
164 | belems[s_group_cols[0]] = int(belems[s_group_cols[0]]) |
---|
165 | if belems[p_group_cols[0]]==pitem1 and belems[s_group_cols[0]] in sitem1: |
---|
166 | repeats1.append(int(belems[6])) |
---|
167 | repeats2.append(int(belems[13])) |
---|
168 | if belems[4] == 'mononucleotide': |
---|
169 | thresholds.append(mono_threshold) |
---|
170 | else: |
---|
171 | thresholds.append(non_mono_threshold) |
---|
172 | mut1["%s\t%s" %(pitem1,sitem1)]=mutabilityEstimator(repeats1,repeats2,thresholds) |
---|
173 | if region == 'align': |
---|
174 | count1[str(pitem1)+'\t'+str(sitem1)]=min(sum(repeats1),sum(repeats2)) |
---|
175 | else: |
---|
176 | if winspecies == 1: |
---|
177 | count1[str(pitem1)+'\t'+str(sitem1)]=sum(repeats1) |
---|
178 | elif winspecies == 2: |
---|
179 | count1[str(pitem1)+'\t'+str(sitem1)]=sum(repeats2) |
---|
180 | else: |
---|
181 | if belems[p_group_cols[0]]==pitem1 and belems[s_group_cols[0]]==sitem1: |
---|
182 | repeats1.append(int(belems[6])) |
---|
183 | repeats2.append(int(belems[13])) |
---|
184 | if belems[4] == 'mononucleotide': |
---|
185 | thresholds.append(mono_threshold) |
---|
186 | else: |
---|
187 | thresholds.append(non_mono_threshold) |
---|
188 | mut1["%s\t%s" %(pitem1,sitem1)]=mutabilityEstimator(repeats1,repeats2,thresholds) |
---|
189 | if region == 'align': |
---|
190 | count1[str(pitem1)+'\t'+str(sitem1)]=min(sum(repeats1),sum(repeats2)) |
---|
191 | else: |
---|
192 | if winspecies == 1: |
---|
193 | count1["%s\t%s" %(pitem1,sitem1)]=sum(repeats1) |
---|
194 | elif winspecies == 2: |
---|
195 | count1["%s\t%s" %(pitem1,sitem1)]=sum(repeats2) |
---|
196 | else: #Sub-group by feature is None |
---|
197 | for bline in blk_lines: |
---|
198 | belems = bline.split('\t') |
---|
199 | if type(pitem1) == list: |
---|
200 | #print >>sys.stderr, "item: " + str(item1) |
---|
201 | if p_group_cols[0] == 6: |
---|
202 | belems[p_group_cols[0]] = int(belems[p_group_cols[0]]) |
---|
203 | if belems[p_group_cols[0]] in pitem1: |
---|
204 | repeats1.append(int(belems[6])) |
---|
205 | repeats2.append(int(belems[13])) |
---|
206 | if belems[4] == 'mononucleotide': |
---|
207 | thresholds.append(mono_threshold) |
---|
208 | else: |
---|
209 | thresholds.append(non_mono_threshold) |
---|
210 | else: |
---|
211 | if belems[p_group_cols[0]]==pitem1: |
---|
212 | repeats1.append(int(belems[6])) |
---|
213 | repeats2.append(int(belems[13])) |
---|
214 | if belems[4] == 'mononucleotide': |
---|
215 | thresholds.append(mono_threshold) |
---|
216 | else: |
---|
217 | thresholds.append(non_mono_threshold) |
---|
218 | mut1["%s" %(pitem1)]=mutabilityEstimator(repeats1,repeats2,thresholds) |
---|
219 | if region == 'align': |
---|
220 | count1["%s" %(pitem1)]=min(sum(repeats1),sum(repeats2)) |
---|
221 | else: |
---|
222 | if winspecies == 1: |
---|
223 | count1[str(pitem1)]=sum(repeats1) |
---|
224 | elif winspecies == 2: |
---|
225 | count1[str(pitem1)]=sum(repeats2) |
---|
226 | |
---|
227 | for pitem2 in uniq_elems_2: |
---|
228 | #repeats1 = [] |
---|
229 | #repeats2 = [] |
---|
230 | thresholds = [] |
---|
231 | if s_group_cols[0] != -1: #Sub-group by feature is not None |
---|
232 | for sitem2 in uniq_s_elems_2: |
---|
233 | repeats1 = [] |
---|
234 | repeats2 = [] |
---|
235 | if type(sitem2)==type(''): |
---|
236 | sitem2 = sitem2.strip() |
---|
237 | for bline in blk_lines: |
---|
238 | belems = bline.split('\t') |
---|
239 | if type(pitem2) == list: |
---|
240 | if p_group_cols[0] == 6: |
---|
241 | belems[p_group_cols[1]] = int(belems[p_group_cols[1]]) |
---|
242 | if belems[p_group_cols[1]] in pitem2 and belems[s_group_cols[1]]==sitem2: |
---|
243 | repeats2.append(int(belems[13])) |
---|
244 | repeats1.append(int(belems[6])) |
---|
245 | if belems[4] == 'mononucleotide': |
---|
246 | thresholds.append(mono_threshold) |
---|
247 | else: |
---|
248 | thresholds.append(non_mono_threshold) |
---|
249 | mut2["%s\t%s" %(pitem2,sitem2)]=mutabilityEstimator(repeats2,repeats1,thresholds) |
---|
250 | #count2[str(pitem2)+'\t'+str(sitem2)]=len(repeats2) |
---|
251 | if region == 'align': |
---|
252 | count2["%s\t%s" %(pitem2,sitem2)]=min(sum(repeats1),sum(repeats2)) |
---|
253 | else: |
---|
254 | if winspecies == 1: |
---|
255 | count2["%s\t%s" %(pitem2,sitem2)]=len(repeats2) |
---|
256 | elif winspecies == 2: |
---|
257 | count2["%s\t%s" %(pitem2,sitem2)]=len(repeats1) |
---|
258 | else: |
---|
259 | if type(sitem2) == list: |
---|
260 | if s_group_cols[0] == 6: |
---|
261 | belems[s_group_cols[1]] = int(belems[s_group_cols[1]]) |
---|
262 | if belems[p_group_cols[1]]==pitem2 and belems[s_group_cols[1]] in sitem2: |
---|
263 | repeats2.append(int(belems[13])) |
---|
264 | repeats1.append(int(belems[6])) |
---|
265 | if belems[4] == 'mononucleotide': |
---|
266 | thresholds.append(mono_threshold) |
---|
267 | else: |
---|
268 | thresholds.append(non_mono_threshold) |
---|
269 | mut2["%s\t%s" %(pitem2,sitem2)]=mutabilityEstimator(repeats2,repeats1,thresholds) |
---|
270 | if region == 'align': |
---|
271 | count2["%s\t%s" %(pitem2,sitem2)]=min(sum(repeats1),sum(repeats2)) |
---|
272 | else: |
---|
273 | if winspecies == 1: |
---|
274 | count2["%s\t%s" %(pitem2,sitem2)]=len(repeats2) |
---|
275 | elif winspecies == 2: |
---|
276 | count2["%s\t%s" %(pitem2,sitem2)]=len(repeats1) |
---|
277 | else: |
---|
278 | if belems[p_group_cols[1]]==pitem2 and belems[s_group_cols[1]]==sitem2: |
---|
279 | repeats1.append(int(belems[13])) |
---|
280 | repeats2.append(int(belems[6])) |
---|
281 | if belems[4] == 'mononucleotide': |
---|
282 | thresholds.append(mono_threshold) |
---|
283 | else: |
---|
284 | thresholds.append(non_mono_threshold) |
---|
285 | mut2["%s\t%s" %(pitem2,sitem2)]=mutabilityEstimator(repeats2,repeats1,thresholds) |
---|
286 | if region == 'align': |
---|
287 | count2["%s\t%s" %(pitem2,sitem2)]=min(sum(repeats1),sum(repeats2)) |
---|
288 | else: |
---|
289 | if winspecies == 1: |
---|
290 | count2["%s\t%s" %(pitem2,sitem2)]=len(repeats2) |
---|
291 | elif winspecies == 2: |
---|
292 | count2["%s\t%s" %(pitem2,sitem2)]=len(repeats1) |
---|
293 | else: #Sub-group by feature is None |
---|
294 | for bline in blk_lines: |
---|
295 | belems = bline.split('\t') |
---|
296 | if type(pitem2) == list: |
---|
297 | if p_group_cols[0] == 6: |
---|
298 | belems[p_group_cols[1]] = int(belems[p_group_cols[1]]) |
---|
299 | if belems[p_group_cols[1]] in pitem2: |
---|
300 | repeats2.append(int(belems[13])) |
---|
301 | repeats1.append(int(belems[6])) |
---|
302 | if belems[4] == 'mononucleotide': |
---|
303 | thresholds.append(mono_threshold) |
---|
304 | else: |
---|
305 | thresholds.append(non_mono_threshold) |
---|
306 | else: |
---|
307 | if belems[p_group_cols[1]]==pitem2: |
---|
308 | repeats2.append(int(belems[13])) |
---|
309 | repeats1.append(int(belems[6])) |
---|
310 | if belems[4] == 'mononucleotide': |
---|
311 | thresholds.append(mono_threshold) |
---|
312 | else: |
---|
313 | thresholds.append(non_mono_threshold) |
---|
314 | mut2["%s" %(pitem2)]=mutabilityEstimator(repeats2,repeats1,thresholds) |
---|
315 | if region == 'align': |
---|
316 | count2["%s" %(pitem2)]=min(sum(repeats1),sum(repeats2)) |
---|
317 | else: |
---|
318 | if winspecies == 1: |
---|
319 | count2["%s" %(pitem2)]=sum(repeats2) |
---|
320 | elif winspecies == 2: |
---|
321 | count2["%s" %(pitem2)]=sum(repeats1) |
---|
322 | for key in mut1.keys(): |
---|
323 | if key in mut2.keys(): |
---|
324 | mut = (mut1[key][0]+mut2[key][0])/(mut1[key][1]+mut2[key][1]) |
---|
325 | count = count1[key] |
---|
326 | del mut2[key] |
---|
327 | else: |
---|
328 | unit_found = False |
---|
329 | if p_group_cols[0] == 7 or s_group_cols[0] == 7: #if it is Repeat Unit (AG, GCT etc.) check for reverse-complements too |
---|
330 | if p_group_cols[0] == 7: |
---|
331 | this,other = 0,1 |
---|
332 | else: |
---|
333 | this,other = 1,0 |
---|
334 | groups1 = key.split('\t') |
---|
335 | mutn = mut1[key][0] |
---|
336 | mutd = mut1[key][1] |
---|
337 | count = 0 |
---|
338 | for key2 in mut2.keys(): |
---|
339 | groups2 = key2.split('\t') |
---|
340 | if groups1[other] == groups2[other]: |
---|
341 | if groups1[this] in groups2[this]*2 or reverse_complement(groups1[this]) in groups2[this]*2: |
---|
342 | #mut = (mut1[key][0]+mut2[key2][0])/(mut1[key][1]+mut2[key2][1]) |
---|
343 | mutn += mut2[key2][0] |
---|
344 | mutd += mut2[key2][1] |
---|
345 | count += int(count2[key2]) |
---|
346 | unit_found = True |
---|
347 | del mut2[key2] |
---|
348 | #break |
---|
349 | if unit_found: |
---|
350 | mut = mutn/mutd |
---|
351 | else: |
---|
352 | mut = mut1[key][0]/mut1[key][1] |
---|
353 | count = count1[key] |
---|
354 | mut = "%.2e" %(mut/num_generations) |
---|
355 | if region == 'align': |
---|
356 | print >>fout, str(blk) + '\t'+seq1 + '\t' + seq2 + '\t' +key.strip()+ '\t'+str(mut) + '\t'+ str(count) |
---|
357 | elif region == 'win': |
---|
358 | fout.write("%s\t%s\t%s\t%s\n" %(blk,key.strip(),mut,count)) |
---|
359 | fout.flush() |
---|
360 | |
---|
361 | #catch any remaining repeats, for instance if the orthologous position contained different repeat units |
---|
362 | for remaining_key in mut2.keys(): |
---|
363 | mut = mut2[remaining_key][0]/mut2[remaining_key][1] |
---|
364 | mut = "%.2e" %(mut/num_generations) |
---|
365 | count = count2[remaining_key] |
---|
366 | if region == 'align': |
---|
367 | print >>fout, str(blk) + '\t'+seq1 + '\t'+seq2 + '\t'+remaining_key.strip()+ '\t'+str(mut)+ '\t'+ str(count) |
---|
368 | elif region == 'win': |
---|
369 | fout.write("%s\t%s\t%s\t%s\n" %(blk,remaining_key.strip(),mut,count)) |
---|
370 | fout.flush() |
---|
371 | #print >>fout, blk + '\t'+remaining_key.strip()+ '\t'+str(mut)+ '\t'+ str(count) |
---|
372 | |
---|
373 | def counter(node, start, end, report_func): |
---|
374 | if start <= node.start < end and start < node.end <= end: |
---|
375 | report_func(node) |
---|
376 | if node.right: |
---|
377 | counter(node.right, start, end, report_func) |
---|
378 | if node.left: |
---|
379 | counter(node.left, start, end, report_func) |
---|
380 | elif node.start < start and node.right: |
---|
381 | counter(node.right, start, end, report_func) |
---|
382 | elif node.start >= end and node.left and node.left.maxend > start: |
---|
383 | counter(node.left, start, end, report_func) |
---|
384 | |
---|
385 | |
---|
386 | def main(): |
---|
387 | infile = sys.argv[1] |
---|
388 | |
---|
389 | for i, line in enumerate( file ( infile )): |
---|
390 | line = line.rstrip('\r\n') |
---|
391 | if len( line )>0 and not line.startswith( '#' ): |
---|
392 | elems = line.split( '\t' ) |
---|
393 | break |
---|
394 | if i == 30: |
---|
395 | break # Hopefully we'll never get here... |
---|
396 | |
---|
397 | if len( elems ) != 15: |
---|
398 | stop_err( "This tool only works on tabular data output by 'Extract Orthologous Microsatellites from pair-wise alignments' tool. The data in your input dataset is either missing or not formatted properly." ) |
---|
399 | global winspecies, speciesind |
---|
400 | if region == 'win': |
---|
401 | if dbkey_i in elems[1]: |
---|
402 | winspecies = 1 |
---|
403 | speciesind = 1 |
---|
404 | elif dbkey_i in elems[8]: |
---|
405 | winspecies = 2 |
---|
406 | speciesind = 8 |
---|
407 | else: |
---|
408 | stop_err("The species build corresponding to your interval file is not present in the Microsatellite file.") |
---|
409 | |
---|
410 | fin = open(infile, 'r') |
---|
411 | skipped = 0 |
---|
412 | blk=0 |
---|
413 | win=0 |
---|
414 | linestr="" |
---|
415 | |
---|
416 | if region == 'win': |
---|
417 | |
---|
418 | msats = NiceReaderWrapper( fileinput.FileInput( infile ), |
---|
419 | chrom_col = speciesind, |
---|
420 | start_col = speciesind+1, |
---|
421 | end_col = speciesind+2, |
---|
422 | strand_col = -1, |
---|
423 | fix_strand = True) |
---|
424 | msatTree = quicksect.IntervalTree() |
---|
425 | for item in msats: |
---|
426 | if type( item ) is GenomicInterval: |
---|
427 | msatTree.insert( item, msats.linenum, item.fields ) |
---|
428 | |
---|
429 | for iline in fint: |
---|
430 | try: |
---|
431 | iline = iline.rstrip('\r\n') |
---|
432 | if not(iline) or iline == "": |
---|
433 | continue |
---|
434 | ielems = iline.strip("\r\n").split('\t') |
---|
435 | ichr = ielems[chr_col_i] |
---|
436 | istart = int(ielems[start_col_i]) |
---|
437 | iend = int(ielems[end_col_i]) |
---|
438 | isrc = "%s.%s" %(dbkey_i,ichr) |
---|
439 | if isrc not in msatTree.chroms: |
---|
440 | continue |
---|
441 | result = [] |
---|
442 | root = msatTree.chroms[isrc] #root node for the chrom |
---|
443 | counter(root, istart, iend, lambda node: result.append( node )) |
---|
444 | if not(result): |
---|
445 | continue |
---|
446 | tmpfile1 = tempfile.NamedTemporaryFile('wb+') |
---|
447 | for node in result: |
---|
448 | tmpfile1.write("%s\n" % "\t".join( node.other )) |
---|
449 | |
---|
450 | tmpfile1.seek(0) |
---|
451 | output_writer(iline, tmpfile1.readlines()) |
---|
452 | except: |
---|
453 | skipped+=1 |
---|
454 | if skipped: |
---|
455 | print "Skipped %d intervals as invalid." %(skipped) |
---|
456 | elif region == 'align': |
---|
457 | if s_group_cols[0] != -1: |
---|
458 | print >>fout, "#Window\tSpecies_1\tSpecies_2\tGroupby_Feature\tSubGroupby_Feature\tMutability\tCount" |
---|
459 | else: |
---|
460 | print >>fout, "#Window\tSpecies_1\tWindow_Start\tWindow_End\tSpecies_2\tGroupby_Feature\tMutability\tCount" |
---|
461 | prev_bnum = -1 |
---|
462 | try: |
---|
463 | for line in fin: |
---|
464 | line = line.strip("\r\n") |
---|
465 | if not(line) or line == "": |
---|
466 | continue |
---|
467 | elems = line.split('\t') |
---|
468 | try: |
---|
469 | assert int(elems[0]) |
---|
470 | assert len(elems) == 15 |
---|
471 | except: |
---|
472 | continue |
---|
473 | new_bnum = int(elems[0]) |
---|
474 | if new_bnum != prev_bnum: |
---|
475 | if prev_bnum != -1: |
---|
476 | output_writer(prev_bnum, linestr.strip().replace('\r','\n').split('\n')) |
---|
477 | linestr = line + "\n" |
---|
478 | else: |
---|
479 | linestr += line |
---|
480 | linestr += "\n" |
---|
481 | prev_bnum = new_bnum |
---|
482 | output_writer(prev_bnum, linestr.strip().replace('\r','\n').split('\n')) |
---|
483 | except Exception, ea: |
---|
484 | print >>sys.stderr, ea |
---|
485 | skipped += 1 |
---|
486 | if skipped: |
---|
487 | print "Skipped %d lines as invalid." %(skipped) |
---|
488 | if __name__ == "__main__": |
---|
489 | main() |
---|