0
|
1 // report the total depth for the segment
|
|
2 // Format: ./a.out alignments.bam splice_site
|
|
3 #include <stdio.h>
|
|
4 #include <string.h>
|
|
5 #include <stdlib.h>
|
|
6
|
|
7 #define __STDC_FORMAT_MACROS
|
|
8 #include <inttypes.h>
|
|
9
|
|
10 #include <algorithm>
|
|
11 #include <vector>
|
|
12 #include <math.h>
|
|
13
|
|
14 #include "alignments.hpp"
|
|
15 #include "blocks.hpp"
|
|
16 #include "stats.hpp"
|
|
17
|
|
18 #define ABS(x) ((x)<0?-(x):(x))
|
|
19
|
|
20 char usage[] = "./subexon-info alignment.bam intron.splice [options]\n"
|
|
21 "options:\n"
|
|
22 "\t--minDepth INT: the minimum coverage depth considered as part of a subexon (default: 2)\n"
|
|
23 "\t--noStats: do not compute the statistical scores (default: not used)\n" ;
|
|
24 char buffer[4096] ;
|
|
25
|
|
26 int gMinDepth ;
|
|
27
|
|
28 bool CompSplitSite( struct _splitSite a, struct _splitSite b )
|
|
29 {
|
|
30 if ( a.chrId < b.chrId )
|
|
31 return true ;
|
|
32 else if ( a.chrId > b.chrId )
|
|
33 return false ;
|
|
34 else if ( a.pos != b.pos )
|
|
35 return a.pos < b.pos ;
|
|
36 else if ( a.type != b.type )
|
|
37 return a.type < b.type ; // We want the start of exons comes first, since we are scanning the genome from left to right,
|
|
38 // so we can terminate the extension early and create a single-base subexon later.
|
|
39 else
|
|
40 return a.oppositePos < b.oppositePos ;
|
|
41 }
|
|
42
|
|
43 bool CompBlocksByAvgDepth( struct _block a, struct _block b )
|
|
44 {
|
|
45 double avgA = a.depthSum / (double)( a.end - a.start + 1 ) ;
|
|
46 double avgB = b.depthSum / (double)( b.end - b.start + 1 ) ;
|
|
47 return avgA < avgB ;
|
|
48 }
|
|
49
|
|
50 bool CompBlocksByRatio( struct _block a, struct _block b )
|
|
51 {
|
|
52 return a.ratio < b.ratio ;
|
|
53 }
|
|
54
|
|
55 int CompDouble( const void *p1, const void *p2 )
|
|
56 {
|
|
57 double a = *(double *)p1 ;
|
|
58 double b = *(double *)p2 ;
|
|
59 if ( a > b )
|
|
60 return 1 ;
|
|
61 else if ( a < b )
|
|
62 return -1 ;
|
|
63 else
|
|
64 return 0 ;
|
|
65 }
|
|
66
|
|
67 // Clean up the split sites;
|
|
68 void FilterAndSortSplitSites( std::vector<struct _splitSite> &sites )
|
|
69 {
|
|
70 std::sort( sites.begin(), sites.end(), CompSplitSite ) ;
|
|
71 int i, j, k, l ;
|
|
72 int size = sites.size() ;
|
|
73
|
|
74 for ( i = 0 ; i < size ; )
|
|
75 {
|
|
76 for ( j = i + 1 ; j < size ; ++j )
|
|
77 if ( sites[j].chrId != sites[i].chrId || sites[j].pos != sites[i].pos || sites[j].type != sites[i].type )
|
|
78 break ;
|
|
79 int maxSupport = 0 ;
|
|
80 for ( k = i ; k < j ; ++k )
|
|
81 if ( sites[k].support > maxSupport )
|
|
82 maxSupport = sites[k].support ;
|
|
83
|
|
84 int strandCnt[2] = {0, 0} ;
|
|
85 char strand = sites[i].strand ;
|
|
86 for ( k = i ; k < j ; ++k )
|
|
87 {
|
|
88 if ( sites[k].strand == '-' )
|
|
89 strandCnt[0] += sites[k].support ;
|
|
90 else if ( sites[k].strand == '+' )
|
|
91 strandCnt[1] += sites[k].support ;
|
|
92
|
|
93 }
|
|
94 if ( strandCnt[0] > strandCnt[1] )
|
|
95 strand = '-' ;
|
|
96 else if ( strandCnt[1] > strandCnt[0] )
|
|
97 strand = '+' ;
|
|
98
|
|
99 bool allOneExceptMax = false ;
|
|
100 if ( maxSupport >= 20 )
|
|
101 {
|
|
102 allOneExceptMax = true ;
|
|
103 for ( k = i ; k < j ; ++k )
|
|
104 {
|
|
105 if ( sites[k].support == maxSupport )
|
|
106 continue ;
|
|
107 if ( sites[k].support > 1 )
|
|
108 {
|
|
109 allOneExceptMax = false ;
|
|
110 break ;
|
|
111 }
|
|
112 }
|
|
113 }
|
|
114
|
|
115 for ( k = i ; k < j ; ++k )
|
|
116 {
|
|
117 if ( ( sites[k].support < 0.01 * maxSupport && sites[k].support <= 3 )
|
|
118 || sites[k].support < 0.001 * maxSupport || sites[k].strand != strand // The introns from the different strand are filtered.
|
|
119 || ( sites[k].support < 0.02 * maxSupport && sites[k].mismatchSum >= 2 * sites[k].support )
|
|
120 || ( allOneExceptMax && sites[k].support == 1 )
|
|
121 || ( sites[k].support <= 2 && sites[k].mismatchSum >= 2 * sites[k].support )
|
|
122 || ( maxSupport >= 2 && sites[k].support == 1 && ( ABS( sites[k].oppositePos - sites[k].pos - 1 ) >= 10000 || sites[k].mismatchSum != 0 ) ) )
|
|
123 {
|
|
124 for ( l = i - 1 ; l >= 0 && sites[l].chrId == sites[i].chrId && sites[l].pos >= sites[k].oppositePos ; --l )
|
|
125 {
|
|
126 if ( sites[l].pos == sites[k].oppositePos && sites[l].oppositePos == sites[k].pos )
|
|
127 {
|
|
128 sites[l].support = -1 ;
|
|
129 sites[k].support = -1 ;
|
|
130 break ;
|
|
131 }
|
|
132 }
|
|
133
|
|
134 for ( l = j ; l < size && sites[l].chrId == sites[i].chrId && sites[l].pos <= sites[k].oppositePos ; ++l )
|
|
135 {
|
|
136 if ( sites[l].pos == sites[k].oppositePos && sites[l].oppositePos == sites[k].pos )
|
|
137 {
|
|
138 sites[l].support = -1 ;
|
|
139 sites[k].support = -1 ;
|
|
140 break ;
|
|
141 }
|
|
142 }
|
|
143 }
|
|
144 /*else if ( sites[k].support <= 1 && sites[k].oppositePos - sites[k].pos + 1 >= 30000 )
|
|
145 {
|
|
146 for ( l = j ; l < size && sites[l].chrId == sites[i].chrId && sites[l].pos <= sites[k].oppositePos ; ++l )
|
|
147 {
|
|
148 if ( sites[l].pos == sites[k].oppositePos && sites[l].oppositePos == sites[k].pos )
|
|
149 {
|
|
150 if ( l - j >= 20 )
|
|
151 {
|
|
152 sites[l].support = -1 ;
|
|
153 sites[k].support = -1 ;
|
|
154 break ;
|
|
155 }
|
|
156 }
|
|
157 }
|
|
158
|
|
159 }*/
|
|
160 }
|
|
161 i = j ;
|
|
162 }
|
|
163
|
|
164 k = 0 ;
|
|
165 for ( i = 0 ; i < size ; ++i )
|
|
166 {
|
|
167 if ( sites[i].support > 0 )
|
|
168 {
|
|
169 sites[k] = sites[i] ;
|
|
170 if ( sites[k].strand == '?' )
|
|
171 sites[k].strand = '.' ;
|
|
172 ++k ;
|
|
173 }
|
|
174 }
|
|
175 sites.resize( k ) ;
|
|
176 }
|
|
177
|
|
178 // Remove the same sites.
|
|
179 void KeepUniqSplitSites( std::vector< struct _splitSite> &sites )
|
|
180 {
|
|
181 int i, j ;
|
|
182 int size = sites.size() ;
|
|
183 int k = 0 ;
|
|
184 for ( i = 0 ; i < size ; )
|
|
185 {
|
|
186 for ( j = i + 1 ; j < size ; ++j )
|
|
187 if ( sites[j].chrId != sites[i].chrId || sites[j].pos != sites[i].pos || sites[j].type != sites[i].type )
|
|
188 break ;
|
|
189 sites[k] = sites[i] ;
|
|
190 ++k ;
|
|
191 /*else
|
|
192 {
|
|
193 if ( sites[i].type != sites[k-1].type )
|
|
194 {
|
|
195 printf( "%d\n", sites[i].pos ) ;
|
|
196 }
|
|
197 }*/
|
|
198 i = j ;
|
|
199 }
|
|
200 sites.resize( k ) ;
|
|
201
|
|
202 // For the sites that corresponds to the start of an exon, we remove the adjust to it and
|
|
203 /*for ( i = 1 ; i < size ; ++i )
|
|
204 {
|
|
205 if ( sites[i].pos == sites[i - 1].pos && sites[i - 1].type == 2 )
|
|
206 {
|
|
207 if ( sites[i].type == 1 )
|
|
208 ++sites[i].pos ;
|
|
209 }
|
|
210 }*/
|
|
211 }
|
|
212
|
|
213 // Filter split sites that are extremely close to each other but on different strand.
|
|
214 void FilterNearSplitSites( std::vector< struct _splitSite> &sites )
|
|
215 {
|
|
216 int i, j ;
|
|
217 int size = sites.size() ;
|
|
218 int k = 0 ;
|
|
219 for ( i = 0 ; i < size - 1 ; ++i )
|
|
220 {
|
|
221 if ( sites[i].support < 0 || sites[i].type != sites[i + 1].type || sites[i].chrId != sites[i + 1].chrId )
|
|
222 continue ;
|
|
223 if ( sites[i + 1].pos - sites[i].pos <= 7 &&
|
|
224 ( sites[i + 1].strand != sites[i].strand ||
|
|
225 sites[i].strand == '?' ) )
|
|
226 {
|
|
227 int tag = i ;
|
|
228 if ( sites[i + 1].support < sites[i].support )
|
|
229 tag = i + 1 ;
|
|
230 sites[tag].support = -1 ;
|
|
231 int direction ;
|
|
232 if ( sites[tag].oppositePos < sites[tag].pos )
|
|
233 direction = -1 ;
|
|
234 else
|
|
235 direction = 1 ;
|
|
236
|
|
237 for ( j = tag ; j >= 0 && j < size ; j += direction )
|
|
238 if ( sites[j].pos == sites[tag].oppositePos && sites[j].oppositePos == sites[tag].pos )
|
|
239 {
|
|
240 sites[j].support = -1 ;
|
|
241 break ;
|
|
242 }
|
|
243 }
|
|
244 }
|
|
245
|
|
246 for ( i = 0 ; i < size ; ++i )
|
|
247 {
|
|
248 if ( sites[i].support > 0 )
|
|
249 {
|
|
250 sites[k] = sites[i] ;
|
|
251 ++k ;
|
|
252 }
|
|
253 }
|
|
254 sites.resize( k ) ;
|
|
255 }
|
|
256
|
|
257 void FilterRepeatSplitSites( std::vector<struct _splitSite> &sites )
|
|
258 {
|
|
259 int i, j ;
|
|
260 int size = sites.size() ;
|
|
261 int k = 0 ;
|
|
262 for ( i = 0 ; i < size ; )
|
|
263 {
|
|
264 for ( j = i + 1 ; j < size ; ++j )
|
|
265 {
|
|
266 if ( sites[j].pos != sites[i].pos || sites[j].type != sites[i].type || sites[i].chrId != sites[j].chrId )
|
|
267 break ;
|
|
268 }
|
|
269 int max = -1 ;
|
|
270 int maxtag = 0 ;
|
|
271 for ( k = i ; k < j ; ++k )
|
|
272 {
|
|
273 if ( sites[k].uniqSupport > max )
|
|
274 {
|
|
275 max = sites[k].uniqSupport ;
|
|
276 maxtag = k ;
|
|
277 }
|
|
278 else if ( sites[k].uniqSupport == max && sites[k].support > sites[maxtag].support )
|
|
279 {
|
|
280 maxtag = k ;
|
|
281 }
|
|
282 }
|
|
283
|
|
284 if ( max > -1 )
|
|
285 {
|
|
286 if ( sites[maxtag].uniqSupport > sites[maxtag].support * 0.1 )
|
|
287 {
|
|
288 for ( k = i ; k < j ; ++k )
|
|
289 if ( sites[k].uniqSupport < 0.05 * sites[k].support )
|
|
290 {
|
|
291 sites[k].support = -1 ;
|
|
292
|
|
293 int direction ;
|
|
294 if ( sites[k].oppositePos < sites[k].pos )
|
|
295 direction = -1 ;
|
|
296 else
|
|
297 direction = 1 ;
|
|
298 int l ;
|
|
299 for ( l = k ; l >= 0 && l < size ; l += direction )
|
|
300 if ( sites[l].pos == sites[k].oppositePos && sites[l].oppositePos == sites[k].pos )
|
|
301 {
|
|
302 sites[l].support = -1 ;
|
|
303 break ;
|
|
304 }
|
|
305 }
|
|
306 }
|
|
307 else
|
|
308 {
|
|
309 for ( k = i ; k < j ; ++k )
|
|
310 {
|
|
311 if ( sites[k].support <= 10 )
|
|
312 {
|
|
313 sites[k].support = -1 ;
|
|
314
|
|
315 int direction ;
|
|
316 if ( sites[k].oppositePos < sites[k].pos )
|
|
317 direction = -1 ;
|
|
318 else
|
|
319 direction = 1 ;
|
|
320 int l ;
|
|
321 for ( l = k ; l >= 0 && l < size ; l += direction )
|
|
322 if ( sites[l].pos == sites[k].oppositePos && sites[l].oppositePos == sites[k].pos )
|
|
323 {
|
|
324 sites[l].support = -1 ;
|
|
325 break ;
|
|
326 }
|
|
327 }
|
|
328 }
|
|
329 }
|
|
330 }
|
|
331
|
|
332 i = j ;
|
|
333 }
|
|
334
|
|
335 k = 0 ;
|
|
336 for ( i = 0 ; i < size ; ++i )
|
|
337 {
|
|
338 if ( sites[i].support > 0 )
|
|
339 {
|
|
340 sites[k] = sites[i] ;
|
|
341 ++k ;
|
|
342 }
|
|
343 }
|
|
344 sites.resize( k ) ;
|
|
345
|
|
346 }
|
|
347
|
|
348
|
|
349 // When k=1, the gamma distribution becomes exponential distribution, and can be optimized analytically..
|
|
350 // Maximize: sum_i z_i( log( 1/theta e^{-x_i / theta} )
|
|
351 /*double ThetaOfExponentialDistribution( double *x, double *z, int n )
|
|
352 {
|
|
353 double sumZ = 0;
|
|
354 double sumZX = 0 ;
|
|
355 for ( i = 0 ; i < n ; ++i )
|
|
356 {
|
|
357 sumZ += z[i] ;
|
|
358 sumZX += z[i] * x[i] ;
|
|
359 }
|
|
360 }*/
|
|
361
|
|
362 // for boundK, if it is positive, it represent the upper bound. If it is negative, -boundK will be the lower bound for k.
|
|
363 // if boundK==0, there is no extra bound.
|
|
364 // The same logic for boundProduct, which bounds k*theta
|
|
365 void GradientDescentGammaDistribution( double &k, double &theta, double initK, double initTheta, double lowerBoundK, double upperBoundK,
|
|
366 double lowerBoundMean, double upperBoundMean, double *x, double *z, int n )
|
|
367 {
|
|
368 int i ;
|
|
369 k = initK ;
|
|
370 theta = initTheta ;
|
|
371 double c = 0.5 ;
|
|
372 int iterCnt = 1 ;
|
|
373
|
|
374 double sumZ = 0 ;
|
|
375 double sumZX = 0 ;
|
|
376 double sumZLogX = 0 ;
|
|
377 double Hessian[2][2] ; // 0 for k, 1 for theta
|
|
378 double inverseHessian[2][2] ;
|
|
379 int tmp ;
|
|
380
|
|
381 double positiveKRecord = -5, positiveThetaRecord = -5 ; // record the value of k, theta when theta, k becomes non-positive.
|
|
382
|
|
383 for ( i = 0 ; i < n ; ++i )
|
|
384 {
|
|
385 sumZ += z[i] ;
|
|
386 sumZX += z[i] * x[i] ;
|
|
387 sumZLogX += z[i] * log( x[i] ) ;
|
|
388 }
|
|
389
|
|
390 while ( 1 )
|
|
391 {
|
|
392 double gradK = 0 ;
|
|
393 double gradTheta = 0 ;
|
|
394
|
|
395 double prevK = k ;
|
|
396 double prevTheta = theta ;
|
|
397 double digammaK = digammal( k ) ;
|
|
398
|
|
399 gradK = sumZ * ( -log( theta ) - digammaK ) + sumZLogX ;
|
|
400 gradTheta = -sumZ * ( k / theta ) + sumZX / ( theta * theta ) ;
|
|
401
|
|
402 Hessian[0][0] = -sumZ * trigamma( k, &tmp ) ;
|
|
403 Hessian[0][1] = -sumZ / theta ; // \partial l / ( \partial k \partial theta)
|
|
404 Hessian[1][0] = -sumZ / theta ;
|
|
405 Hessian[1][1] = sumZ * k / ( theta * theta ) - 2 * sumZX / ( theta * theta * theta ) ;
|
|
406
|
|
407 double det = Hessian[0][0] * Hessian[1][1] - Hessian[0][1] * Hessian[1][0] ;
|
|
408 /*printf( "%s iter %d:\n", __func__, iterCnt ) ;
|
|
409 printf( "%lf %lf %lf %lf\n", sumZ, k, theta, sumZX ) ;
|
|
410 printf( "%lf %lf %lf\n", gradK, gradTheta, det ) ;
|
|
411 printf( "%lf %lf %lf %lf\n", Hessian[0][0], Hessian[0][1], Hessian[1][0], Hessian[1][1] ) ;*/
|
|
412 if ( det <= 1e-4 && det >=-1e-4 )
|
|
413 {
|
|
414 k = k + c / iterCnt * gradK ;
|
|
415 theta = theta + c / iterCnt * gradTheta ;
|
|
416 }
|
|
417 else
|
|
418 {
|
|
419 inverseHessian[0][0] = Hessian[1][1] / det ;
|
|
420 inverseHessian[0][1] = -Hessian[0][1] / det ;
|
|
421 inverseHessian[1][0] = -Hessian[1][0] / det ;
|
|
422 inverseHessian[1][1] = Hessian[0][0] / det ;
|
|
423 //printf( "%lf %lf %lf %lf: %lf\n=====\n", inverseHessian[0][0], inverseHessian[0][1], inverseHessian[1][0], inverseHessian[1][1],
|
|
424 // Hessian[1][0] * inverseHessian[0][1] + Hessian[1][1] * inverseHessian[1][1] ) ;
|
|
425 double step = 0.5 ;
|
|
426 k = k - step * ( inverseHessian[0][0] * gradK + inverseHessian[0][1] * gradTheta ) ;
|
|
427 theta = theta - step * ( inverseHessian[1][0] * gradK + inverseHessian[1][1] * gradTheta ) ;
|
|
428
|
|
429 bool flag = false ;
|
|
430 if ( k <= 1e-6 )
|
|
431 {
|
|
432 step = ( prevK - 1e-6 ) / ( inverseHessian[0][0] * gradK + inverseHessian[0][1] * gradTheta ) ;
|
|
433
|
|
434 if ( ABS( theta - positiveThetaRecord ) < 1e-5 )
|
|
435 flag = true ;
|
|
436 positiveThetaRecord = theta ;
|
|
437 }
|
|
438 if ( theta <= 1e-6 )
|
|
439 {
|
|
440 double tmp = ( prevTheta - 1e-6 ) / ( inverseHessian[1][0] * gradK + inverseHessian[1][1] * gradTheta ) ;
|
|
441 if ( tmp < step )
|
|
442 step = tmp ;
|
|
443
|
|
444 if ( ABS( k - positiveKRecord ) < 1e-5 )
|
|
445 flag = true ;
|
|
446 positiveKRecord = k ;
|
|
447 }
|
|
448
|
|
449 if ( step != 0.5 )
|
|
450 {
|
|
451 k = prevK - step * ( inverseHessian[0][0] * gradK + inverseHessian[0][1] * gradTheta ) ;
|
|
452 theta = prevTheta - step * ( inverseHessian[1][0] * gradK + inverseHessian[1][1] * gradTheta ) ;
|
|
453 }
|
|
454
|
|
455 /*if ( flag )
|
|
456 {
|
|
457 k = prevK ;
|
|
458 theta = prevTheta ;
|
|
459 break ;
|
|
460 }*/
|
|
461 }
|
|
462
|
|
463 if ( upperBoundK > 0 && k > upperBoundK )
|
|
464 k = upperBoundK ;
|
|
465 else if ( lowerBoundK > 0 && k < lowerBoundK )
|
|
466 k = lowerBoundK ;
|
|
467
|
|
468 if ( upperBoundMean > 0 && k * theta > upperBoundMean )
|
|
469 {
|
|
470 theta = upperBoundMean / k ;
|
|
471 }
|
|
472 else if ( lowerBoundMean > 0 && k * theta < lowerBoundMean )
|
|
473 {
|
|
474 theta = lowerBoundMean / k ;
|
|
475 }
|
|
476
|
|
477 if ( k <= 1e-6 )
|
|
478 {
|
|
479 k = 1e-6 ;
|
|
480 }
|
|
481
|
|
482 if ( theta <= 1e-6 )
|
|
483 {
|
|
484 theta = 1e-6 ;
|
|
485 }
|
|
486
|
|
487 double diff = ABS( prevK - k ) + ABS( prevTheta - theta ) ;
|
|
488 if ( diff < 1e-5 )
|
|
489 break ;
|
|
490
|
|
491 ++iterCnt ;
|
|
492 //if ( det <= 1e-4 && det >=-1e-4 && k >= 5000 ) //&& diff < 1 )
|
|
493 if ( k >= 10000 ) //&& diff < 1 )
|
|
494 {
|
|
495 k = prevK ;
|
|
496 theta = prevTheta ;
|
|
497 break ;
|
|
498 }
|
|
499 if ( iterCnt == 1000 )
|
|
500 break ;
|
|
501 }
|
|
502 }
|
|
503
|
|
504 double MixtureGammaEM( double *x, int n, double &pi, double *k, double *theta, int tries, double meanBound[2], int iter = 1000 )
|
|
505 {
|
|
506 int i ;
|
|
507 double *z = new double[n] ; // the expectation that it assigned to model 0.
|
|
508 double *oneMinusZ = new double[n] ;
|
|
509 int t = 0 ;
|
|
510 double history[5] = {-1, -1, -1, -1, -1} ;
|
|
511 double maxX = -1 ;
|
|
512 double sumX = 0 ;
|
|
513 if ( n <= 0 )
|
|
514 return 0 ;
|
|
515
|
|
516 for ( i = 0 ; i < n ; ++i )
|
|
517 {
|
|
518 sumX += x[i] ;
|
|
519 if ( x[i] > maxX )
|
|
520 maxX = x[i] ;
|
|
521 }
|
|
522 if ( maxX > meanBound[1] && meanBound[1] >= 0 )
|
|
523 maxX = meanBound[1] ;
|
|
524
|
|
525 /*if ( meanBound[1] == -1 )
|
|
526 {
|
|
527 // The EM for coverage
|
|
528 maxX = 10.0 ;
|
|
529 }*/
|
|
530
|
|
531 while ( 1 )
|
|
532 {
|
|
533 double npi, nk[2], ntheta[2] ;
|
|
534 double sum = 0 ;
|
|
535 for ( i = 0 ; i < n ; ++i )
|
|
536 {
|
|
537 //double lf0 = -k[0] * log( theta[0] ) + ( k[0] - 1 ) * log( cov[i]) - cov[i] / theta[0] - lgamma( k[0] );
|
|
538 //double lf1 = -k[1] * log( theta[1] ) + ( k[1] - 1 ) * log( cov[i]) - cov[i] / theta[1] - lgamma( k[1] );
|
|
539 //z[i] = exp( lf0 + log( pi ) ) / ( exp( lf0 + log( pi ) ) + exp( lf1 + log( 1 - pi ) ) ) ;
|
|
540 if ( pi != 0 )
|
|
541 z[i] = MixtureGammaAssignment( x[i], pi, k, theta ) ;
|
|
542 else
|
|
543 z[i] = 0 ;
|
|
544 /*if ( isnan( z[i] ) )
|
|
545 {
|
|
546 printf( "nan: %lf %lf %lf %lf\n", x[i], pi, k, theta ) ;
|
|
547 }*/
|
|
548 oneMinusZ[i] = 1 - z[i] ;
|
|
549 sum += z[i] ;
|
|
550 }
|
|
551
|
|
552 // compute new pi.
|
|
553 npi = sum / n ;
|
|
554
|
|
555 // Use gradient descent to compute new k and theta.
|
|
556 if ( 1 ) //pi > 0 )
|
|
557 {
|
|
558 double bound ;
|
|
559 if ( meanBound[1] != -1 ) // the EM for ratio
|
|
560 {
|
|
561 bound = ( theta[1] * k[1] > 1 ) ? 1 : ( theta[1] * k[1] ) / ( 1 + tries );
|
|
562 GradientDescentGammaDistribution( nk[0], ntheta[0], k[0], theta[0], k[1], -1, -1, bound, x, z, n ) ; // It seems setting an upper bound 1 for k[0] is not a good idea.
|
|
563 }
|
|
564 else
|
|
565 {
|
|
566 bound = ( theta[1] * k[1] > 1 ) ? 1 : ( theta[1] * k[1] ) / ( 1 + tries ) ;
|
|
567 GradientDescentGammaDistribution( nk[0], ntheta[0], k[0], theta[0], k[1], -1, meanBound[0], bound, x, z, n ) ; // It seems setting an upper bound 1 for k[0] is not a good idea.
|
|
568 }
|
|
569 GradientDescentGammaDistribution( nk[1], ntheta[1], k[1], theta[1], -1, k[0], theta[0] * k[0], maxX, x, oneMinusZ, n ) ;
|
|
570 }
|
|
571 else
|
|
572 {
|
|
573 GradientDescentGammaDistribution( nk[1], ntheta[1], k[1], theta[1], 0, 0, 0, 0, x, oneMinusZ, n ) ;
|
|
574 }
|
|
575
|
|
576 double diff ;
|
|
577 if ( isnan( npi ) || isnan( nk[0] ) || isnan( nk[1] ) || isnan( ntheta[0] ) || isnan( ntheta[1] ) )
|
|
578 {
|
|
579 delete[] z ;
|
|
580 delete[] oneMinusZ ;
|
|
581 return -1 ;
|
|
582 }
|
|
583 diff = ABS( nk[0] - k[0] ) + ABS( nk[1] - k[1] )
|
|
584 + ABS( ntheta[0] - theta[0] ) + ABS( ntheta[1] - theta[1] ) ; // pi is fully determined by these 4 parameters.
|
|
585 if ( diff < 1e-4 )
|
|
586 break ;
|
|
587 diff = ABS( nk[0] - history[1] ) + ABS( nk[1] - history[2] )
|
|
588 + ABS( ntheta[0] - history[3] ) + ABS( ntheta[1] - history[4] ) ; // pi is fully determined by these 4 parameters.
|
|
589 if ( diff < 1e-4 )
|
|
590 break ;
|
|
591
|
|
592 history[0] = pi ;
|
|
593 history[1] = k[0] ;
|
|
594 history[2] = k[1] ;
|
|
595 history[3] = theta[0] ;
|
|
596 history[4] = theta[1] ;
|
|
597
|
|
598 pi = npi ;
|
|
599 k[0] = nk[0] ;
|
|
600 k[1] = nk[1] ;
|
|
601 theta[0] = ntheta[0] ;
|
|
602 theta[1] = ntheta[1] ;
|
|
603
|
|
604 /*double logLikelihood = 0 ;
|
|
605 for ( i = 0 ; i < n ; ++i )
|
|
606 logLikelihood += log( pi * exp( LogGammaDensity( x[i], k[0], theta[0]) ) +
|
|
607 (1 - pi ) * exp( LogGammaDensity( x[i], k[1], theta[1] ) ) ) ;*/
|
|
608
|
|
609 //printf( "%d: %lf %lf %lf %lf %lf\n", t, pi, k[0], theta[0], k[1], theta[1] ) ;
|
|
610
|
|
611 ++t ;
|
|
612 if ( iter != -1 && t >= iter )
|
|
613 break ;
|
|
614 }
|
|
615 delete[] z ;
|
|
616 delete[] oneMinusZ ;
|
|
617 return 0 ;
|
|
618 }
|
|
619
|
|
620 bool IsParametersTheSame( double *k, double *theta )
|
|
621 {
|
|
622 if ( ABS( k[0] - k[1] ) < 1e-2 && ABS( theta[0] - theta[1] ) < 1e-2 )
|
|
623 return true ;
|
|
624 return false ;
|
|
625 }
|
|
626
|
|
627 int RatioAndCovEM( double *covRatio, double *cov, int n, double &piRatio, double kRatio[2],
|
|
628 double thetaRatio[2], double &piCov, double kCov[2], double thetaCov[2] )
|
|
629 {
|
|
630 int i ;
|
|
631 piRatio = 0.6 ; // mixture coefficient for model 0 and 1
|
|
632 kRatio[0] = 0.9 ;
|
|
633 kRatio[1] = 0.45 ;
|
|
634 thetaRatio[0] = 0.05 ;
|
|
635 thetaRatio[1] = 1 ;
|
|
636 double meanBound[2] = {-1, 1} ; // [0] is for the lower bound of the noise model, [1] is for the upper bound of the true model
|
|
637
|
|
638 /*double *filteredCovRatio = new double[n] ;// ignore the ratio that is greater than 5.
|
|
639 int m = 0 ;
|
|
640 for ( i = 0 ; i < n ; ++i )
|
|
641 if ( covRatio[i] < 1.0 )
|
|
642 {
|
|
643 filteredCovRatio[m] = covRatio[i] ;
|
|
644 ++m ;
|
|
645 }*/
|
|
646 srand( 17 ) ;
|
|
647 int maxTries = 10 ;
|
|
648 int t = 0 ;
|
|
649 double *buffer = new double[n] ;
|
|
650 for ( i = 0 ; i < n ; ++i )
|
|
651 buffer[i] = covRatio[i] ;
|
|
652 qsort( buffer, n, sizeof( double ), CompDouble ) ;
|
|
653 //covRatio = buffer ;
|
|
654 while ( 1 )
|
|
655 {
|
|
656 //printf( "EM\n" ) ;
|
|
657 MixtureGammaEM( covRatio, n, piRatio, kRatio, thetaRatio, t, meanBound ) ;
|
|
658 //printf( "%lf %lf %lf %lf %lf\n", piRatio, kRatio[0], kRatio[1], thetaRatio[0], thetaRatio[1] ) ;
|
|
659 if ( piRatio > 0.999 || piRatio < 0.001 || IsParametersTheSame( kRatio, thetaRatio ) )
|
|
660 {
|
|
661 ++t ;
|
|
662 if ( t > maxTries )
|
|
663 break ;
|
|
664 piRatio = 0.6 ;
|
|
665 kRatio[0] += ( ( rand() * 0.5 - RAND_MAX ) / (double)RAND_MAX * 0.1 ) ;
|
|
666 if ( kRatio[0] <= 0 )
|
|
667 kRatio[0] = 0.9 ;
|
|
668 kRatio[1] += ( ( rand() * 0.5 - RAND_MAX ) / (double)RAND_MAX * 0.1 ) ;
|
|
669 if ( kRatio[1] <= 0 )
|
|
670 kRatio[1] = 0.45 ;
|
|
671 thetaRatio[0] += ( ( rand() * 0.5 - RAND_MAX ) / (double)RAND_MAX * 0.1 ) ;
|
|
672 if ( thetaRatio[0] <= 0 )
|
|
673 thetaRatio[0] = 0.05 ;
|
|
674 thetaRatio[1] += ( ( rand() * 0.5 - RAND_MAX ) / (double)RAND_MAX * 0.1 ) ;
|
|
675 if ( thetaRatio[1] <= 0 )
|
|
676 thetaRatio[1] = 1 ;
|
|
677 if ( kRatio[0] < kRatio[1] )
|
|
678 {
|
|
679 if ( rand() & 1 )
|
|
680 kRatio[0] = kRatio[1] ;
|
|
681 else
|
|
682 kRatio[1] = kRatio[0] ;
|
|
683 }
|
|
684 if ( kRatio[0] * thetaRatio[0] > kRatio[1] * thetaRatio[1] )
|
|
685 {
|
|
686 thetaRatio[0] = kRatio[1] * thetaRatio[1] / kRatio[0] ;
|
|
687 }
|
|
688 //printf( "%lf %lf %lf %lf %lf\n", piRatio, kRatio[0], kRatio[1], thetaRatio[0], thetaRatio[1] ) ;
|
|
689
|
|
690 continue ;
|
|
691 }
|
|
692
|
|
693 break ;
|
|
694 }
|
|
695 //delete[] filteredCovRatio ;
|
|
696 if ( t > maxTries && piRatio > 0.999 )
|
|
697 {
|
|
698 /*piRatio = 0.6 ; // mixture coefficient for model 0 and 1
|
|
699 kRatio[0] = 0.9 ;
|
|
700 kRatio[1] = 0.45 ;
|
|
701 thetaRatio[0] = 0.05 ;
|
|
702 thetaRatio[1] = 1 ;*/
|
|
703 piRatio = 0.999 ;
|
|
704 }
|
|
705 if ( IsParametersTheSame( kRatio, thetaRatio ) || piRatio <= 1e-3 )
|
|
706 piRatio = 1e-3 ;
|
|
707
|
|
708 piCov = piRatio ; // mixture coefficient for model 0 and 1
|
|
709 kCov[0] = 0.9 ;
|
|
710 kCov[1] = 0.45 ;
|
|
711 thetaCov[0] = 3 ;
|
|
712 thetaCov[1] = 12 ;
|
|
713
|
|
714 // only do one iteration of EM, so that pi does not change?
|
|
715 // But it seems it still better to run full EM.
|
|
716 meanBound[0] = 1.01 ;
|
|
717 meanBound[1] = -1 ;
|
|
718
|
|
719 //printf( "for coverage:\n" ) ;
|
|
720 //piCov = 0.001000 ;
|
|
721 MixtureGammaEM( cov, n, piCov, kCov, thetaCov, 0, meanBound ) ;
|
|
722 //printf( "for coverage done\n" ) ;
|
|
723 piCov = piRatio ;
|
|
724
|
|
725 delete []buffer ;
|
|
726
|
|
727 return 0 ;
|
|
728 }
|
|
729
|
|
730 double GetPValue( double x, double *k, double *theta )
|
|
731 {
|
|
732 int fault ;
|
|
733 double p ;
|
|
734 p = 1 - gammad( x / theta[0], k[0], &fault ) ;
|
|
735 return p ;
|
|
736 }
|
|
737
|
|
738 // if x's value is less than the average of (k0-1)*theta0, then we force x=(k0-1)*theta0,
|
|
739 // the mode of the model 0. Of course, it does not affect when k0<=1 already.
|
|
740 double MixtureGammaAssignmentAdjust( double x, double pi, double* k, double *theta )
|
|
741 {
|
|
742 if ( x < ( k[0] - 1 ) * theta[0] )
|
|
743 {
|
|
744 x = ( k[0] - 1 ) * theta[0] ;
|
|
745 }
|
|
746 return MixtureGammaAssignment( x, pi, k, theta ) ;
|
|
747 }
|
|
748
|
|
749
|
|
750 // Transform the cov number for better fitting
|
|
751 double TransformCov( double c )
|
|
752 {
|
|
753 double ret ;
|
|
754 // original it is c-1.
|
|
755 // Use -2 instead of -1 is that many region covered to 1 reads will be filtered when
|
|
756 // build the subexons.
|
|
757 //
|
|
758 //ret = sqrt( c ) - 1 ;
|
|
759 if ( c <= 2 + 1e-6 )
|
|
760 ret = 1e-6 ;
|
|
761 else
|
|
762 ret = c - 2 ;
|
|
763 return ret ;
|
|
764 //return log( c ) / log( 2.0 ) ;
|
|
765 }
|
|
766
|
|
767 int main( int argc, char *argv[] )
|
|
768 {
|
|
769 int i, j ;
|
|
770 bool noStats = false ;
|
|
771 if ( argc < 3 )
|
|
772 {
|
|
773 fprintf( stderr, usage ) ;
|
|
774 exit( 1 ) ;
|
|
775 }
|
|
776
|
|
777 gMinDepth = 2 ;
|
|
778
|
|
779 for ( i = 3 ; i < argc ; ++i )
|
|
780 {
|
|
781 if ( !strcmp( argv[i], "--noStats" ) )
|
|
782 {
|
|
783 noStats = true ;
|
|
784 continue ;
|
|
785 }
|
|
786 else if ( !strcmp( argv[i], "--minDepth" ) )
|
|
787 {
|
|
788 gMinDepth = atoi( argv[i + 1] ) ;
|
|
789 ++i ;
|
|
790 continue ;
|
|
791 }
|
|
792 else
|
|
793 {
|
|
794 fprintf( stderr, "Unknown argument: %s\n", argv[i] ) ;
|
|
795 return 0 ;
|
|
796 }
|
|
797 }
|
|
798
|
|
799 Alignments alignments ;
|
|
800 alignments.Open( argv[1] ) ;
|
|
801 std::vector<struct _splitSite> splitSites ; // only compromised the
|
|
802 std::vector<struct _splitSite> allSplitSites ;
|
|
803
|
|
804 // read in the splice site
|
|
805 FILE *fp ;
|
|
806 fp = fopen( argv[2], "r" ) ;
|
|
807 char chrom[50] ;
|
|
808 int64_t start, end ;
|
|
809 int support ;
|
|
810 char strand[3] ;
|
|
811 int uniqSupport, secondarySupport, uniqEditDistance, secondaryEditDistance ;
|
|
812 while ( fscanf( fp, "%s %" PRId64 " %" PRId64 " %d %s %d %d %d %d", chrom, &start, &end, &support, strand,
|
|
813 &uniqSupport, &secondarySupport, &uniqEditDistance, &secondaryEditDistance ) != EOF )
|
|
814 {
|
|
815 if ( support <= 0 )
|
|
816 continue ;
|
|
817 //if ( !( uniqSupport >= 1
|
|
818 // || secondarySupport > 10 ) )
|
|
819 //if ( uniqSupport <= 0.01 * ( uniqSupport + secondarySupport ) || ( uniqSupport == 0 && secondarySupport < 20 ) )
|
|
820 //if ( uniqSupport == 0 && secondarySupport <= 10 )
|
|
821 // continue ;
|
|
822 int chrId = alignments.GetChromIdFromName( chrom ) ;
|
|
823 struct _splitSite ss ;
|
|
824 --start ;
|
|
825 --end ;
|
|
826 ss.pos = start ;
|
|
827 ss.chrId = chrId ;
|
|
828 ss.type = 2 ;
|
|
829 ss.oppositePos = end ;
|
|
830 ss.strand = strand[0] ;
|
|
831 ss.support = support ;
|
|
832 ss.uniqSupport = uniqSupport ;
|
|
833 ss.mismatchSum = uniqEditDistance + secondaryEditDistance ;
|
|
834 splitSites.push_back( ss ) ;
|
|
835
|
|
836 ss.pos = end ;
|
|
837 ss.type = 1 ;
|
|
838 ss.oppositePos = start ;
|
|
839 ss.strand = strand[0] ;
|
|
840 ss.support = support ;
|
|
841 ss.uniqSupport = uniqSupport ;
|
|
842 ss.mismatchSum = uniqEditDistance + secondaryEditDistance ;
|
|
843 splitSites.push_back( ss ) ;
|
|
844 }
|
|
845 fclose( fp ) ;
|
|
846 //printf( "ss:%d\n", splitSites.size() ) ;
|
|
847
|
|
848 //printf( "ss:%d\n", splitSites.size() ) ;
|
|
849
|
|
850 alignments.GetGeneralInfo( true ) ;
|
|
851 // Build the blocks
|
|
852 Blocks regions ;
|
|
853 alignments.Rewind() ;
|
|
854 regions.BuildExonBlocks( alignments ) ;
|
|
855 //printf( "%d\n", regions.exonBlocks.size() ) ;
|
|
856
|
|
857 FilterAndSortSplitSites( splitSites ) ;
|
|
858 FilterNearSplitSites( splitSites ) ;
|
|
859 FilterRepeatSplitSites( splitSites ) ;
|
|
860 regions.FilterSplitSitesInRegions( splitSites ) ;
|
|
861 regions.FilterGeneMergeSplitSites( splitSites ) ;
|
|
862
|
|
863
|
|
864 allSplitSites = splitSites ;
|
|
865 KeepUniqSplitSites( splitSites ) ;
|
|
866
|
|
867 //for ( i = 0 ; i < splitSites.size() ; ++i )
|
|
868 // printf( "%d %d\n", splitSites[i].pos + 1, splitSites[i].oppositePos + 1 ) ;
|
|
869 // Split the blocks using split site
|
|
870 regions.SplitBlocks( alignments, splitSites ) ;
|
|
871 //printf( "%d\n", regions.exonBlocks.size() ) ;
|
|
872 /*for ( i = 0 ; i < regions.exonBlocks.size() ; ++i )
|
|
873 {
|
|
874 struct _block &e = regions.exonBlocks[i] ;
|
|
875 printf( "%s %" PRId64 " %" PRId64 " %d %d\n", alignments.GetChromName( e.chrId ), e.start + 1, e.end + 1, e.leftType, e.rightType ) ;
|
|
876 }
|
|
877 return 0 ;*/
|
|
878 // Recompute the coverage for each block.
|
|
879 alignments.Rewind() ;
|
|
880 //printf( "Before computeDepth: %d\n", regions.exonBlocks.size() ) ;
|
|
881
|
|
882 regions.ComputeDepth( alignments ) ;
|
|
883 //printf( "After computeDepth: %d\n", regions.exonBlocks.size() ) ;
|
|
884
|
|
885 // Merge blocks that may have a hollow coverage by accident.
|
|
886 regions.MergeNearBlocks() ;
|
|
887 //printf( "After merge: %d\n", regions.exonBlocks.size() ) ;
|
|
888
|
|
889 // Put the intron informations
|
|
890 regions.AddIntronInformation( allSplitSites, alignments ) ;
|
|
891 //printf( "After add information.\n" ) ;
|
|
892
|
|
893 // Compute the average ratio against the left and right connected subexons.
|
|
894 regions.ComputeRatios() ;
|
|
895 //printf( "After compute ratios.\n" ) ;
|
|
896
|
|
897 //printf( "Finish building regions.\n" ) ;
|
|
898 if ( noStats )
|
|
899 {
|
|
900 // just output the subexons.
|
|
901 if ( realpath( argv[1], buffer ) == NULL )
|
|
902 {
|
|
903 strcpy( buffer, argv[1] ) ;
|
|
904 }
|
|
905 printf( "#%s\n", buffer ) ;
|
|
906 printf( "#fitted_ir_parameter_ratio: pi: -1 k0: -1 theta0: -1 k1: -1 theta1: -1\n" ) ;
|
|
907 printf( "#fitted_ir_parameter_cov: pi: -1 k0: -1 theta0: -1 k1: -1 theta1: -1\n" ) ;
|
|
908
|
|
909 int blockCnt = regions.exonBlocks.size() ;
|
|
910 for ( int i = 0 ; i < blockCnt ; ++i )
|
|
911 {
|
|
912 struct _block &e = regions.exonBlocks[i] ;
|
|
913 double avgDepth = (double)e.depthSum / ( e.end - e.start + 1 ) ;
|
|
914 printf( "%s %" PRId64 " %" PRId64 " %d %d %lf -1 -1 -1 -1 ", alignments.GetChromName( e.chrId ), e.start + 1, e.end + 1, e.leftType, e.rightType, avgDepth ) ;
|
|
915 int prevCnt = e.prevCnt ;
|
|
916 if ( i > 0 && e.start == regions.exonBlocks[i - 1].end + 1 &&
|
|
917 e.leftType == regions.exonBlocks[i - 1].rightType )
|
|
918 {
|
|
919 printf( "%d ", prevCnt + 1 ) ;
|
|
920 for ( j = 0 ; j < prevCnt ; ++j )
|
|
921 printf( "%" PRId64 " ", regions.exonBlocks[ e.prev[j] ].end + 1 ) ;
|
|
922 printf( "%" PRId64 " ", regions.exonBlocks[i - 1].end + 1 ) ;
|
|
923 }
|
|
924 else
|
|
925 {
|
|
926 printf( "%d ", prevCnt ) ;
|
|
927 for ( j = 0 ; j < prevCnt ; ++j )
|
|
928 printf( "%" PRId64 " ", regions.exonBlocks[ e.prev[j] ].end + 1 ) ;
|
|
929 }
|
|
930
|
|
931 int nextCnt = e.nextCnt ;
|
|
932 if ( i < blockCnt - 1 && e.end == regions.exonBlocks[i + 1].start - 1 &&
|
|
933 e.rightType == regions.exonBlocks[i + 1].leftType )
|
|
934 {
|
|
935 printf( "%d %" PRId64 " ", nextCnt + 1, regions.exonBlocks[i + 1].start + 1 ) ;
|
|
936 }
|
|
937 else
|
|
938 printf( "%d ", nextCnt ) ;
|
|
939 for ( j = 0 ; j < nextCnt ; ++j )
|
|
940 printf( "%" PRId64 " ", regions.exonBlocks[ e.next[j] ].start + 1 ) ;
|
|
941 printf( "\n" ) ;
|
|
942
|
|
943 }
|
|
944 return 0 ;
|
|
945 }
|
|
946
|
|
947 // Extract the blocks for different events.
|
|
948 int blockCnt = regions.exonBlocks.size() ;
|
|
949 std::vector<struct _block> irBlocks ; // The regions corresponds to intron retention events.
|
|
950 double *leftClassifier = new double[ blockCnt ] ;
|
|
951 double *rightClassifier = new double[ blockCnt ] ;
|
|
952 std::vector<struct _block> overhangBlocks ; //blocks like (...[ or ]...)
|
|
953 std::vector<struct _block> islandBlocks ; // blocks like (....)
|
|
954
|
|
955 for ( i = 0 ; i < blockCnt ; ++i )
|
|
956 {
|
|
957 int ltype = regions.exonBlocks[i].leftType ;
|
|
958 int rtype = regions.exonBlocks[i].rightType ;
|
|
959 leftClassifier[i] = -1 ;
|
|
960 rightClassifier[i] = -1 ;
|
|
961
|
|
962 //double avgDepth = (double)regions.exonBlocks[i].depthSum / ( regions.exonBlocks[i].end - regions.exonBlocks[i].start + 1 ) ;
|
|
963
|
|
964 if ( ltype == 2 && rtype == 1 )
|
|
965 {
|
|
966 // candidate intron retention.
|
|
967 // Note that when I compute the ratio, it is already made sure that the avgDepth>1.
|
|
968 double ratio = regions.PickLeftAndRightRatio( regions.exonBlocks[i] ) ;
|
|
969
|
|
970 //printf( "%lf %lf\n", regions.exonBlocks[i].leftRatio, regions.exonBlocks[i].rightRatio ) ;
|
|
971 if ( ratio > 0 )
|
|
972 {
|
|
973 regions.exonBlocks[i].ratio = ratio ;
|
|
974 irBlocks.push_back( regions.exonBlocks[i] ) ;
|
|
975 irBlocks[ irBlocks.size() - 1 ].contigId = i ;
|
|
976 }
|
|
977 }
|
|
978 else if ( ( ltype == 0 && rtype == 1 ) || ( ltype == 2 && rtype == 0 ) )
|
|
979 {
|
|
980 // subexons like (...[ or ]...)
|
|
981 double ratio ;
|
|
982 if ( ltype == 0 )
|
|
983 {
|
|
984 ratio = regions.exonBlocks[i].rightRatio ;
|
|
985 }
|
|
986 else if ( ltype == 2 )
|
|
987 {
|
|
988 ratio = regions.exonBlocks[i].leftRatio ;
|
|
989 }
|
|
990 if ( ratio > 0 )
|
|
991 {
|
|
992 regions.exonBlocks[i].ratio = ratio ;
|
|
993 overhangBlocks.push_back( regions.exonBlocks[i] ) ;
|
|
994 overhangBlocks[ overhangBlocks.size() - 1].contigId = i ;
|
|
995 }
|
|
996 }
|
|
997 else if ( ltype == 0 && rtype == 0 )
|
|
998 {
|
|
999 islandBlocks.push_back( regions.exonBlocks[i] ) ;
|
|
1000 islandBlocks[ islandBlocks.size() - 1].contigId = i ;
|
|
1001 }
|
|
1002 }
|
|
1003
|
|
1004 // Compute the histogram for each intron.
|
|
1005 int irBlockCnt = irBlocks.size() ;
|
|
1006 double *cov = new double[irBlockCnt] ;
|
|
1007 double *covRatio = new double[ irBlockCnt ] ;
|
|
1008 double piRatio, kRatio[2], thetaRatio[2] ;
|
|
1009 double piCov, kCov[2], thetaCov[2] ;
|
|
1010 for ( i = 0 ; i < irBlockCnt ; ++i )
|
|
1011 {
|
|
1012 double avgDepth = regions.GetAvgDepth( irBlocks[i] ) ;
|
|
1013 //cov[i] = ( avgDepth - 1 ) / ( flankingAvg - 1 ) ;
|
|
1014 cov[i] = TransformCov( avgDepth ) ;
|
|
1015
|
|
1016 covRatio[i] = regions.PickLeftAndRightRatio( irBlocks[i] ) ;
|
|
1017 //cov[i] = avgDepth / anchorAvg ;
|
|
1018 //printf( "%"PRId64" %d %d: %lf %lf\n", irBlocks[i].depthSum, irBlocks[i].start, irBlocks[i].end, avgDepth, cov[i] ) ;
|
|
1019 }
|
|
1020
|
|
1021 /*fp = fopen( "ratio.out", "r" ) ;
|
|
1022 int irBlockCnt = 0 ;
|
|
1023 double *cov = new double[10000] ;
|
|
1024 while ( 1 )
|
|
1025 {
|
|
1026 double r ;
|
|
1027 if ( fscanf( fp, "%lf", &r ) == EOF )
|
|
1028 break ;
|
|
1029 cov[ irBlockCnt ] = r ;
|
|
1030 ++irBlockCnt ;
|
|
1031 }
|
|
1032 fclose( fp ) ;*/
|
|
1033 RatioAndCovEM( covRatio, cov, irBlockCnt, piRatio, kRatio, thetaRatio, piCov, kCov, thetaCov ) ;
|
|
1034
|
|
1035 for ( i = 0 ; i < irBlockCnt ; ++i )
|
|
1036 {
|
|
1037 //double lf0 = -k[0] * log( theta[0] ) + ( k[0] - 1 ) * log( cov[i]) - cov[i] / theta[0] - lgamma( k[0] ) ;
|
|
1038 //double lf1 = -k[1] * log( theta[1] ) + ( k[1] - 1 ) * log( cov[i]) - cov[i] / theta[1] - lgamma( k[1] ) ;
|
|
1039
|
|
1040 double p1, p2, p ;
|
|
1041
|
|
1042 p1 = MixtureGammaAssignmentAdjust( covRatio[i], piRatio, kRatio, thetaRatio ) ;
|
|
1043 p2 = MixtureGammaAssignmentAdjust( cov[i], piCov, kCov, thetaCov ) ;
|
|
1044
|
|
1045 /*p1 = GetPValue( covRatio[i], kRatio, thetaRatio ) ; //1 - gammad( covRatio[i] / thetaRatio[0], kRatio[0], &fault ) ;
|
|
1046 if ( piRatio != 0 )
|
|
1047 p2 = GetPValue( cov[i], kCov, thetaCov ) ;//1 - gammad( cov[i] / thetaCov[0], kCov[0], &fault ) ;
|
|
1048 else
|
|
1049 p2 = p1 ;*/
|
|
1050 //printf( "%lf %lf: %lf %lf\n", covRatio[i], cov[i], p1, p2 ) ;
|
|
1051 p = p1 > p2 ? p1 : p2 ;
|
|
1052
|
|
1053
|
|
1054 //printf( "%d %d: avg: %lf ratio: %lf p: %lf\n", irBlocks[i].start, irBlocks[i].end, irBlocks[i].depthSum / (double)( irBlocks[i].end - irBlocks[i].start + 1 ), covRatio[i],
|
|
1055 // p ) ;
|
|
1056 leftClassifier[ irBlocks[i].contigId ] = p ;
|
|
1057 rightClassifier[ irBlocks[i].contigId ] = p ;
|
|
1058 }
|
|
1059
|
|
1060 // Process the classifier for overhang subexons and the subexons to see whether we need soft boundary
|
|
1061 int overhangBlockCnt = overhangBlocks.size() ;
|
|
1062 delete []cov ;
|
|
1063 delete []covRatio ;
|
|
1064
|
|
1065 cov = new double[ overhangBlockCnt ] ;
|
|
1066 covRatio = new double[overhangBlockCnt] ;
|
|
1067 double overhangPiRatio, overhangKRatio[2], overhangThetaRatio[2] ;
|
|
1068 double overhangPiCov, overhangKCov[2], overhangThetaCov[2] ;
|
|
1069
|
|
1070 for ( i = 0 ; i < overhangBlockCnt ; ++i )
|
|
1071 {
|
|
1072 covRatio[i] = overhangBlocks[i].ratio ;
|
|
1073 cov[i] = TransformCov( regions.GetAvgDepth( overhangBlocks[i] ) ) ;
|
|
1074 }
|
|
1075 RatioAndCovEM( covRatio, cov, overhangBlockCnt, overhangPiRatio, overhangKRatio, overhangThetaRatio, overhangPiCov, overhangKCov, overhangThetaCov ) ;
|
|
1076
|
|
1077 for ( i = 0 ; i < overhangBlockCnt ; ++i )
|
|
1078 {
|
|
1079 //double lf0 = -k[0] * log( theta[0] ) + ( k[0] - 1 ) * log( cov[i]) - cov[i] / theta[0] - lgamma( k[0] ) ;
|
|
1080 //double lf1 = -k[1] * log( theta[1] ) + ( k[1] - 1 ) * log( cov[i]) - cov[i] / theta[1] - lgamma( k[1] ) ;
|
|
1081
|
|
1082 double p1, p2, p ;
|
|
1083 p1 = MixtureGammaAssignmentAdjust( covRatio[i], overhangPiRatio, overhangKRatio, overhangThetaRatio ) ;
|
|
1084 p2 = MixtureGammaAssignmentAdjust( cov[i], overhangPiCov, overhangKCov, overhangThetaCov ) ;
|
|
1085
|
|
1086 /*p1 = GetPValue( covRatio[i], overhangKRatio, overhangThetaRatio ) ; //1 - gammad( covRatio[i] / thetaRatio[0], kRatio[0], &fault ) ;
|
|
1087 if ( overhangPiRatio != 0)
|
|
1088 p2 = GetPValue( cov[i], overhangKCov, overhangThetaCov ) ;//1 - gammad( cov[i] / thetaCov[0], kCov[0], &fault ) ;
|
|
1089 else
|
|
1090 p2 = p1 ;*/
|
|
1091
|
|
1092 //p = p1 < p2 ? p1 : p2 ;
|
|
1093 p = sqrt( p1 * p2 ) ;
|
|
1094
|
|
1095 int idx = overhangBlocks[i].contigId ;
|
|
1096 if ( regions.exonBlocks[idx].rightType == 0 )
|
|
1097 leftClassifier[ idx ] = rightClassifier[ idx ] = p ;
|
|
1098 else
|
|
1099 leftClassifier[ idx ] = rightClassifier[ idx ] = p ;
|
|
1100 }
|
|
1101
|
|
1102 for ( i = 0 ; i < blockCnt ; ++i )
|
|
1103 {
|
|
1104 struct _block &e = regions.exonBlocks[i] ;
|
|
1105 //if ( ( e.leftType == 0 && e.rightType == 1 ) ||
|
|
1106 // variance-stabailizing transformation of poisson distribution. But we are more conservative here.
|
|
1107 // The multiply 2 before that is because we ignore the region below 0, so we need to somehow renormalize the distribution.
|
|
1108 if ( e.leftType == 1 )
|
|
1109 {
|
|
1110 if ( e.leftRatio >= 0 )
|
|
1111 leftClassifier[i] = 2 * alnorm( e.leftRatio * 2.0 , true ) ;
|
|
1112 else
|
|
1113 leftClassifier[i] = 1 ;
|
|
1114 }
|
|
1115 /*else if ( e.leftType == 0 )
|
|
1116 {
|
|
1117 // If this region is a start of a gene, the other sample might introduce
|
|
1118 // a new intron before it. So we want to test whether this region can still
|
|
1119 // be a start of a gene even there is an intron before it.
|
|
1120 for ( j = i + 1 ; j < blockCnt ; ++j )
|
|
1121 {
|
|
1122 if ( regions.exonBlocks[j].contigId != regions.exonBlocks[i].contigId )
|
|
1123 break ;
|
|
1124 }
|
|
1125
|
|
1126 for ( k = i ; k < j ; ++k )
|
|
1127 if ( regions.exonBlocks[j].leftType == 1 )
|
|
1128 break ;
|
|
1129 if ( k >= j )
|
|
1130 {
|
|
1131 leftClassifier[i] = alnorm( )
|
|
1132 }
|
|
1133 }*/
|
|
1134
|
|
1135 //if ( ( e.rightType == 0 && e.leftType == 2 ) ||
|
|
1136 if ( e.rightType == 2 )
|
|
1137 {
|
|
1138 if ( e.rightRatio >= 0 )
|
|
1139 rightClassifier[i] = 2 * alnorm( e.rightRatio * 2.0, true ) ;
|
|
1140 else
|
|
1141 rightClassifier[i] = 1 ;
|
|
1142 }
|
|
1143 }
|
|
1144
|
|
1145 // Process the result for subexons seems like single-exon transcript (...)
|
|
1146 int islandBlockCnt = islandBlocks.size() ;
|
|
1147 //std::sort( islandBlocks.begin(), islandBlocks.end(), CompBlocksByAvgDepth ) ;
|
|
1148 for ( i = 0, j = 0 ; i < islandBlockCnt ; ++i )
|
|
1149 {
|
|
1150 /*if ( regions.GetAvgDepth( islandBlocks[i] ) != regions.GetAvgDepth( islandBlocks[j] ) )
|
|
1151 j = i ;
|
|
1152 leftClassifier[ islandBlocks[i].contigId ] = 1 - (j + 1) / (double)( islandBlockCnt + 1 ) ;
|
|
1153 rightClassifier[ islandBlocks[i].contigId ] = 1 - (j + 1) / (double)( islandBlockCnt + 1 ) ;*/
|
|
1154 double p = GetPValue( TransformCov( regions.GetAvgDepth( islandBlocks[i] ) ), kCov, thetaCov ) ;
|
|
1155 leftClassifier[ islandBlocks[i].contigId ] = p ;
|
|
1156 rightClassifier[ islandBlocks[i].contigId ] = p ;
|
|
1157 }
|
|
1158
|
|
1159
|
|
1160 // Output the result.
|
|
1161 if ( realpath( argv[1], buffer ) == NULL )
|
|
1162 {
|
|
1163 strcpy( buffer, argv[1] ) ;
|
|
1164 }
|
|
1165 printf( "#%s\n", buffer ) ;
|
|
1166 // TODO: higher precision.
|
|
1167 printf( "#fitted_ir_parameter_ratio: pi: %lf k0: %lf theta0: %lf k1: %lf theta1: %lf\n", piRatio, kRatio[0], thetaRatio[0], kRatio[1], thetaRatio[1] ) ;
|
|
1168 printf( "#fitted_ir_parameter_cov: pi: %lf k0: %lf theta0: %lf k1: %lf theta1: %lf\n", piCov, kCov[0], thetaCov[0], kCov[1], thetaCov[1] ) ;
|
|
1169
|
|
1170 printf( "#fitted_overhang_parameter_ratio: pi: %lf k0: %lf theta0: %lf k1: %lf theta1: %lf\n", overhangPiRatio, overhangKRatio[0], overhangThetaRatio[0], overhangKRatio[1], overhangThetaRatio[1] ) ;
|
|
1171 printf( "#fitted_overhang_parameter_cov: pi: %lf k0: %lf theta0: %lf k1: %lf theta1: %lf\n", overhangPiCov, overhangKCov[0], overhangThetaCov[0], overhangKCov[1], overhangThetaCov[1] ) ;
|
|
1172
|
|
1173
|
|
1174 for ( int i = 0 ; i < blockCnt ; ++i )
|
|
1175 {
|
|
1176 struct _block &e = regions.exonBlocks[i] ;
|
|
1177 double avgDepth = regions.GetAvgDepth( e ) ;
|
|
1178 printf( "%s %" PRId64 " %" PRId64 " %d %d %c %c %lf %lf %lf %lf %lf ", alignments.GetChromName( e.chrId ), e.start + 1, e.end + 1, e.leftType, e.rightType,
|
|
1179 e.leftStrand, e.rightStrand, avgDepth,
|
|
1180 e.leftRatio, e.rightRatio, leftClassifier[i], rightClassifier[i] ) ;
|
|
1181 int prevCnt = e.prevCnt ;
|
|
1182 if ( i > 0 && e.start == regions.exonBlocks[i - 1].end + 1 )
|
|
1183 //&& e.leftType == regions.exonBlocks[i - 1].rightType )
|
|
1184 {
|
|
1185 printf( "%d ", prevCnt + 1 ) ;
|
|
1186 for ( j = 0 ; j < prevCnt ; ++j )
|
|
1187 printf( "%" PRId64 " ", regions.exonBlocks[ e.prev[j] ].end + 1 ) ;
|
|
1188 printf( "%" PRId64 " ", regions.exonBlocks[i - 1].end + 1 ) ;
|
|
1189 }
|
|
1190 else
|
|
1191 {
|
|
1192 printf( "%d ", prevCnt ) ;
|
|
1193 for ( j = 0 ; j < prevCnt ; ++j )
|
|
1194 printf( "%" PRId64 " ", regions.exonBlocks[ e.prev[j] ].end + 1 ) ;
|
|
1195 }
|
|
1196
|
|
1197 int nextCnt = e.nextCnt ;
|
|
1198 if ( i < blockCnt - 1 && e.end == regions.exonBlocks[i + 1].start - 1 )
|
|
1199 //&& e.rightType == regions.exonBlocks[i + 1].leftType )
|
|
1200 {
|
|
1201 printf( "%d %" PRId64 " ", nextCnt + 1, regions.exonBlocks[i + 1].start + 1 ) ;
|
|
1202 }
|
|
1203 else
|
|
1204 printf( "%d ", nextCnt ) ;
|
|
1205 for ( j = 0 ; j < nextCnt ; ++j )
|
|
1206 printf( "%" PRId64 " ", regions.exonBlocks[ e.next[j] ].start + 1 ) ;
|
|
1207 printf( "\n" ) ;
|
|
1208 }
|
|
1209
|
|
1210 delete[] cov ;
|
|
1211 delete[] covRatio ;
|
|
1212 delete[] leftClassifier ;
|
|
1213 delete[] rightClassifier ;
|
|
1214 }
|