git-svn-id: http://svn.tuebingen.mpg.de/ag-raetsch/projects/QPalma@8622 e1793c9e...
[qpalma.git] / scripts / PipelineHeuristic.py
1 #!/usr/bin/env python
2 # -*- coding: utf-8 -*-
3
4 import cPickle
5 import sys
6 import pdb
7 import os
8 import os.path
9 import math
10 import resource
11
12 from qpalma.DataProc import *
13 from qpalma.computeSpliceWeights import *
14 from qpalma.set_param_palma import *
15 from qpalma.computeSpliceAlignWithQuality import *
16 from qpalma.penalty_lookup_new import *
17 from qpalma.compute_donacc import *
18 from qpalma.TrainingParam import Param
19 from qpalma.Plif import Plf
20
21 from qpalma.tools.splicesites import getDonAccScores
22 from qpalma.Configuration import *
23
24 from compile_dataset import getSpliceScores, get_seq_and_scores
25
26 from numpy.matlib import mat,zeros,ones,inf
27 from numpy import inf,mean
28
29 from qpalma.parsers import PipelineReadParser
30
31
32 def unbracket_est(est):
33 new_est = ''
34 e = 0
35
36 while True:
37 if e >= len(est):
38 break
39
40 if est[e] == '[':
41 new_est += est[e+2]
42 e += 4
43 else:
44 new_est += est[e]
45 e += 1
46
47 return "".join(new_est).lower()
48
49
50 class PipelineHeuristic:
51 """
52 This class wraps the filter which decides whether an alignment found by
53 vmatch is spliced an should be then newly aligned using QPalma or not.
54 """
55
56 def __init__(self,run_fname,data_fname,param_fname):
57 """
58 We need a run object holding information about the nr. of support points
59 etc.
60 """
61
62 run = cPickle.load(open(run_fname))
63 self.run = run
64
65 start = cpu()
66
67 self.data_fname = data_fname
68
69 self.param = cPickle.load(open(param_fname))
70
71 # Set the parameters such as limits penalties for the Plifs
72 [h,d,a,mmatrix,qualityPlifs] = set_param_palma(self.param,True,run)
73
74 self.h = h
75 self.d = d
76 self.a = a
77 self.mmatrix = mmatrix
78 self.qualityPlifs = qualityPlifs
79
80 # when we look for alternative alignments with introns this value is the
81 # mean of intron size
82 self.intron_size = 90
83
84 self.read_size = 36
85
86 self.original_reads = {}
87
88 for line in open('/fml/ag-raetsch/share/projects/qpalma/solexa/allReads.pipeline'):
89 line = line.strip()
90 id,seq,q1,q2,q3 = line.split()
91 id = int(id)
92 self.original_reads[id] = seq
93
94 lengthSP = run['numLengthSuppPoints']
95 donSP = run['numDonSuppPoints']
96 accSP = run['numAccSuppPoints']
97 mmatrixSP = run['matchmatrixRows']*run['matchmatrixCols']
98 numq = run['numQualSuppPoints']
99 totalQualSP = run['totalQualSuppPoints']
100
101 currentPhi = zeros((run['numFeatures'],1))
102 currentPhi[0:lengthSP] = mat(h.penalties[:]).reshape(lengthSP,1)
103 currentPhi[lengthSP:lengthSP+donSP] = mat(d.penalties[:]).reshape(donSP,1)
104 currentPhi[lengthSP+donSP:lengthSP+donSP+accSP] = mat(a.penalties[:]).reshape(accSP,1)
105 currentPhi[lengthSP+donSP+accSP:lengthSP+donSP+accSP+mmatrixSP] = mmatrix[:]
106
107 totalQualityPenalties = self.param[-totalQualSP:]
108 currentPhi[lengthSP+donSP+accSP+mmatrixSP:] = totalQualityPenalties[:]
109 self.currentPhi = currentPhi
110
111 # we want to identify spliced reads
112 # so true pos are spliced reads that are predicted "spliced"
113 self.true_pos = 0
114
115 # as false positives we count all reads that are not spliced but predicted
116 # as "spliced"
117 self.false_pos = 0
118
119 self.true_neg = 0
120 self.false_neg = 0
121
122 # total time spend for get seq and scores
123 self.get_time = 0.0
124 self.calcAlignmentScoreTime = 0.0
125 self.alternativeScoresTime = 0.0
126
127 self.count_time = 0.0
128 self.read_parsing = 0.0
129 self.main_loop = 0.0
130 self.splice_site_time = 0.0
131 self.computeSpliceAlignWithQualityTime = 0.0
132 self.computeSpliceWeightsTime = 0.0
133 self.DotProdTime = 0.0
134 self.array_stuff = 0.0
135 stop = cpu()
136
137 self.init_time = stop-start
138
139 def filter(self):
140 """
141 This method...
142 """
143 run = self.run
144
145 start = cpu()
146
147 rrp = PipelineReadParser(self.data_fname)
148 all_remapped_reads = rrp.parse()
149
150 stop = cpu()
151
152 self.read_parsing = stop-start
153
154 ctr = 0
155 unspliced_ctr = 0
156 spliced_ctr = 0
157
158 print 'Starting filtering...'
159 _start = cpu()
160
161 for readId,currentReadLocations in all_remapped_reads.items():
162 for location in currentReadLocations[:1]:
163
164 id = location['id']
165 chr = location['chr']
166 pos = location['pos']
167 strand = location['strand']
168 mismatch = location['mismatches']
169 length = location['length']
170 off = location['offset']
171 seq = location['seq']
172 prb = location['prb']
173 cal_prb = location['cal_prb']
174 chastity = location['chastity']
175
176 id = int(id)
177
178 if strand == '-':
179 continue
180
181 if ctr == 100:
182 break
183
184 #if pos > 10000000:
185 # continue
186
187 unb_seq = unbracket_est(seq)
188 effective_len = len(unb_seq)
189
190 genomicSeq_start = pos
191 genomicSeq_stop = pos+effective_len-1
192
193 start = cpu()
194 #print genomicSeq_start,genomicSeq_stop
195 currentDNASeq, currentAcc, currentDon = get_seq_and_scores(chr,strand,genomicSeq_start,genomicSeq_stop,run['dna_flat_files'])
196 stop = cpu()
197 self.get_time += stop-start
198
199 dna = currentDNASeq
200 exons = zeros((2,1))
201 exons[0,0] = 0
202 exons[1,0] = effective_len
203 est = unb_seq
204 original_est = seq
205 quality = prb
206
207 #pdb.set_trace()
208
209 currentVMatchAlignment = dna, exons, est, original_est, quality,\
210 currentAcc, currentDon
211 vMatchScore = self.calcAlignmentScore(currentVMatchAlignment)
212
213 alternativeAlignmentScores = self.calcAlternativeAlignments(location)
214
215 start = cpu()
216 # found no alternatives
217 if alternativeAlignmentScores == []:
218 continue
219
220 maxAlternativeAlignmentScore = max(alternativeAlignmentScores)
221 #print 'vMatchScore/alternativeScore: %f %f ' % (vMatchScore,maxAlternativeAlignmentScore)
222 #print 'all candidates %s' % str(alternativeAlignmentScores)
223
224 new_id = id - 1000000300000
225
226 unspliced = False
227 # unspliced
228 if new_id > 0:
229 unspliced = True
230
231 # Seems that according to our learned parameters VMatch found a good
232 # alignment of the current read
233 if maxAlternativeAlignmentScore < vMatchScore:
234 unspliced_ctr += 1
235
236 if unspliced:
237 self.true_neg += 1
238 else:
239 self.false_neg += 1
240
241 # We found an alternative alignment considering splice sites that scores
242 # higher than the VMatch alignment
243 else:
244 spliced_ctr += 1
245
246 if unspliced:
247 self.false_pos += 1
248 else:
249 self.true_pos += 1
250
251 ctr += 1
252 stop = cpu()
253 self.count_time = stop-start
254
255 _stop = cpu()
256 self.main_loop = _stop-_start
257
258 print 'Unspliced/Splice: %d %d'%(unspliced_ctr,spliced_ctr)
259 print 'True pos / false pos : %d %d'%(self.true_pos,self.false_pos)
260 print 'True neg / false neg : %d %d'%(self.true_neg,self.false_neg)
261
262
263 def findHighestScoringSpliceSites(self,currentAcc,currentDon):
264
265 acc = []
266 for idx,score in enumerate(currentAcc):
267 if score > -inf:
268 acc.append((idx,score))
269 if idx>self.read_size:
270 break
271
272 acc.sort(lambda x,y: x[0]-y[0])
273
274 don = []
275 for idx,score in enumerate(currentDon):
276 if score > -inf:
277 don.append((idx,score))
278 if idx>self.read_size:
279 break
280
281 don.sort(lambda x,y: x[0]-y[0])
282
283 return don,acc
284
285
286 def calcAlternativeAlignments(self,location):
287 """
288 Given an alignment proposed by Vmatch this function calculates possible
289 alternative alignments taking into account for example matched
290 donor/acceptor positions.
291 """
292
293 run = self.run
294
295 id = location['id']
296 chr = location['chr']
297 pos = location['pos']
298 strand = location['strand']
299 seq = location['seq']
300 #orig_seq = location['orig_seq']
301 prb = location['prb']
302 cal_prb = location['cal_prb']
303
304 orig_seq = self.original_reads[int(id)]
305
306 unb_seq = unbracket_est(seq)
307 effective_len = len(unb_seq)
308
309 genomicSeq_start = pos
310 genomicSeq_stop = pos+self.intron_size*2+self.read_size*2
311
312 start = cpu()
313 currentDNASeq, currentAcc, currentDon = get_seq_and_scores(chr,strand,genomicSeq_start,genomicSeq_stop,run['dna_flat_files'])
314 stop = cpu()
315 self.get_time += stop-start
316
317 dna = currentDNASeq
318
319 start = cpu()
320 alt_don,alt_acc = self.findHighestScoringSpliceSites(currentAcc,currentDon)
321 stop = cpu()
322 self.splice_site_time = stop-start
323
324 alternativeScores = []
325
326 exons = zeros((2,2),dtype=numpy.int)
327 est = unb_seq
328 original_est = seq
329 quality = prb
330
331 # inlined
332 h = self.h
333 d = self.d
334 a = self.a
335 mmatrix = self.mmatrix
336 qualityPlifs = self.qualityPlifs
337 # inlined
338
339 IntronScore = calculatePlif(h, [self.intron_size])[0]
340 dummyAcceptorScore = calculatePlif(a, [0.25])[0]
341 dummyDonorScore = calculatePlif(d, [0.25])[0]
342 print IntronScore,dummyAcceptorScore,dummyDonorScore
343
344 _start = cpu()
345 for (don_pos,don_score) in alt_don:
346 start = cpu()
347
348 acc_pos = don_pos + self.intron_size
349
350 exons[0,0] = 0
351 exons[0,1] = don_pos
352 exons[1,0] = acc_pos+1
353 exons[1,1] = acc_pos+1+(self.read_size-don_pos)
354
355 _dna = dna[:int(exons[1,1])]
356 _dna = _dna[:exons[1,0]] + orig_seq[don_pos:]
357
358 _currentAcc = currentAcc[:int(exons[1,1])]
359 _currentAcc = [0.25]*len(_currentAcc)
360
361 _currentDon = currentDon[:int(exons[1,1])]
362 _currentDon = [0.25]*len(_currentDon)
363
364 currentVMatchAlignment = _dna, exons, est, original_est, quality,\
365 _currentAcc, _currentDon
366
367 stop = cpu()
368 self.array_stuff += stop - start
369
370 #alternativeScore = self.calcAlignmentScore(currentVMatchAlignment)
371 #alternativeScores.append(self.calcAlignmentScore(currentVMatchAlignment))
372
373 # Lets start calculation
374 dna, exons, est, original_est, quality, acc_supp, don_supp =\
375 currentVMatchAlignment
376
377 # Berechne die Parameter des wirklichen Alignments (but with untrained d,a,h ...)
378 trueSpliceAlign, trueWeightMatch, trueWeightQuality ,dna_calc =\
379 computeSpliceAlignWithQuality(dna, exons, est, original_est,\
380 quality, qualityPlifs, run)
381
382 stop = cpu()
383 self.computeSpliceAlignWithQualityTime += stop-start
384 start = cpu()
385
386 # Calculate the weights
387 trueWeightDon, trueWeightAcc, trueWeightIntron =\
388 computeSpliceWeights(d, a, h, trueSpliceAlign, don_supp, acc_supp)
389
390 stop = cpu()
391 self.computeSpliceWeightsTime += stop-start
392
393 start = cpu()
394
395 trueWeight = numpy.vstack([trueWeightIntron, trueWeightDon, trueWeightAcc, trueWeightMatch, trueWeightQuality])
396
397 # Calculate w'phi(x,y) the total score of the alignment
398 alternativeScores.append((trueWeight.T * self.currentPhi)[0,0])
399
400 # remove mismatching positions in the second exon
401 original_est_cut=''
402
403 est_ptr=0
404 dna_ptr=0
405 ptr=0
406 while ptr<len(original_est):
407
408 if original_est[ptr]=='[':
409 dnaletter=original_est[ptr+1]
410 estletter=original_est[ptr+2]
411 if est_ptr<=exons[0,1]:
412 original_est_cut+=original_est[ptr:ptr+4]
413 else:
414 original_est_cut+=estletter # EST letter
415 ptr+=4
416 else:
417 dnaletter=original_est[ptr]
418 estletter=dnaletter
419
420 original_est_cut+=estletter # EST letter
421 ptr+=1
422
423 if estletter=='-':
424 dna_ptr+=1
425 elif dnaletter=='-':
426 est_ptr+=1
427 else:
428 dna_ptr+=1
429 est_ptr+=1
430
431 assert(dna_ptr<=len(dna))
432 assert(est_ptr<=len(est))
433
434 # new score
435 DonorScore = calculatePlif(d, [don_score])[0]
436
437 score = computeSpliceAlignScoreWithQuality(original_est_cut, quality, qualityPlifs, run, self.currentPhi)
438 score += dummyAcceptorScore + IntronScore + DonorScore
439
440 print 'diff %f,%f,%f' % ((trueWeight.T * self.currentPhi)[0,0] - score,(trueWeight.T * self.currentPhi)[0,0], score)
441
442 stop = cpu()
443 self.DotProdTime += stop-start
444
445 _stop = cpu()
446 self.alternativeScoresTime += _stop-_start
447
448 return alternativeScores
449
450
451 def calcAlignmentScore(self,alignment):
452 """
453 Given an alignment (dna,exons,est) and the current parameter for QPalma
454 this function calculates the dot product of the feature representation of
455 the alignment and the parameter vector i.e the alignment score.
456 """
457
458 start = cpu()
459 run = self.run
460
461 # Lets start calculation
462 dna, exons, est, original_est, quality, acc_supp, don_supp = alignment
463
464 score = computeSpliceAlignScoreWithQuality(original_est, quality, self.qualityPlifs, run, self.currentPhi)
465
466 stop = cpu()
467 self.calcAlignmentScoreTime += stop-start
468
469 return score
470
471
472 def cpu():
473 return (resource.getrusage(resource.RUSAGE_SELF).ru_utime+\
474 resource.getrusage(resource.RUSAGE_SELF).ru_stime)
475
476
477 if __name__ == '__main__':
478 #run_fname = sys.argv[1]
479 #data_fname = sys.argv[2]
480 #param_filename = sys.argv[3]
481
482 dir = '/fml/ag-raetsch/home/fabio/tmp/QPalma_test/run_+_quality_+_splicesignals_+_intron_len'
483 jp = os.path.join
484
485 run_fname = jp(dir,'run_object.pickle')
486 #data_fname = '/fml/ag-raetsch/share/projects/qpalma/solexa/current_data/map.vm_unspliced_flag'
487
488 #data_fname = '/fml/ag-raetsch/share/projects/qpalma/solexa/pipeline_data/map.vm_2k'
489 data_fname = '/fml/ag-raetsch/share/projects/qpalma/solexa/pipeline_data/map.vm_100'
490
491 param_fname = jp(dir,'param_500.pickle')
492
493 ph1 = PipelineHeuristic(run_fname,data_fname,param_fname)
494
495 start = cpu()
496 ph1.filter()
497 stop = cpu()
498
499 print 'total time elapsed: %f' % (stop-start)
500 print 'time spend for get seq: %f' % ph1.get_time
501 print 'time spend for calcAlignmentScoreTime: %f' % ph1.calcAlignmentScoreTime
502 print 'time spend for alternativeScoresTime: %f' % ph1.alternativeScoresTime
503 print 'time spend for count time: %f' % ph1.count_time
504 print 'time spend for init time: %f' % ph1.init_time
505 print 'time spend for read_parsing time: %f' % ph1.read_parsing
506 print 'time spend for main_loop time: %f' % ph1.main_loop
507 print 'time spend for splice_site_time time: %f' % ph1.splice_site_time
508
509 print 'time spend for computeSpliceAlignWithQualityTime time: %f'% ph1.computeSpliceAlignWithQualityTime
510 print 'time spend for computeSpliceWeightsTime time: %f'% ph1.computeSpliceWeightsTime
511 print 'time spend for DotProdTime time: %f'% ph1.DotProdTime
512 print 'time spend forarray_stuff time: %f'% ph1.array_stuff
513 #import cProfile
514 #cProfile.run('ph1.filter()')
515
516 #import hotshot
517 #p = hotshot.Profile('profile.log')
518 #p.runcall(ph1.filter)