1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
|
// $Id: bestTamura92param.cpp 962 2006-11-07 15:13:34Z privmane $
#include "bestTamura92param.h"
#include <iostream>
using namespace std;
#include "bblEM.h"
#include "bblEMProportionalEB.h"
#include "bblLSProportionalEB.h"
#include "numRec.h"
#include "logFile.h"
#include "bestAlpha.h"
bestTamura92ParamFixedTree::bestTamura92ParamFixedTree(const tree& et, // find best TrTv and theta
const sequenceContainer& sc,
stochasticProcess& sp,
const Vdouble * weights,
const int maxTotalIterations,
const MDOUBLE epsilonLikelihoodImprovment,
const MDOUBLE epsilonLoglikelihoodForTrTvOptimization,
const MDOUBLE epsilonLoglikelihoodForThetaOptimization,
const MDOUBLE upperBoundOnTrTv) {
LOG(5,<<"Starting bestTamura92ParamFixedTree: find Best TrTv and theta"<<endl);
MDOUBLE oldL = VERYSMALL;
MDOUBLE newL = VERYSMALL;
// first guess for the parameters
MDOUBLE prevTrTv = upperBoundOnTrTv*0.3;
MDOUBLE prevTheta = 0.5;
for (int i=0; i < maxTotalIterations; ++i) {
// optimize TrTv
newL = -brent(0.0, prevTrTv, upperBoundOnTrTv,
C_evalTrTvParam(et,sc,sp,weights),
epsilonLoglikelihoodForTrTvOptimization,
&_bestTrTv);
// optimize Theta
newL = -brent(0.0, prevTheta, 1.0,
C_evalTheta(et,sc,sp,weights),
epsilonLoglikelihoodForThetaOptimization,
&_bestTheta);
// check for improvement in the likelihood
if (newL > oldL+epsilonLikelihoodImprovment) {
prevTrTv = _bestTrTv;
prevTheta = _bestTheta;
oldL = newL;
_bestL = newL;
} else {
if (newL>oldL) {
_bestL = newL;
} else {
_bestL = oldL;
_bestTrTv = prevTrTv;
_bestTheta = prevTheta;
}
break;
}
}
}
bestTamura92ParamAndBBL::bestTamura92ParamAndBBL(tree& et, //find best TrTv, theta and best BBL
const sequenceContainer& sc,
stochasticProcess& sp,
const Vdouble * weights,
const int maxTotalIterations,
const MDOUBLE epsilonLikelihoodImprovment,
const MDOUBLE epsilonLoglikelihoodForTrTvOptimization,
const MDOUBLE epsilonLoglikelihoodForThetaOptimization,
const MDOUBLE epsilonLoglikelihoodForBBL,
const MDOUBLE upperBoundOnTrTv,
const int maxBBLIterations){
LOG(5,<<"Starting bestTamura92ParamAndBBL: find best TrTv, theta and BBL"<<endl);
MDOUBLE oldL = VERYSMALL;
MDOUBLE newL = VERYSMALL;
// first guess for the parameters
MDOUBLE prevTrTv = upperBoundOnTrTv*0.3;
MDOUBLE prevTheta = 0.5;
tree prevTree;
for (int i=0; i < maxTotalIterations; ++i) {
// optimize TrTv
newL = -brent(0.0, prevTrTv, upperBoundOnTrTv,
C_evalTrTvParam(et,sc,sp,weights),
epsilonLoglikelihoodForTrTvOptimization,
&_bestTrTv);
(static_cast<tamura92*>(sp.getPijAccelerator()->getReplacementModel()))->changeTrTv(_bestTrTv);
// optimize Theta
newL = -brent(0.0, prevTheta, 1.0,
C_evalTheta(et,sc,sp,weights),
epsilonLoglikelihoodForThetaOptimization,
&_bestTheta);
(static_cast<tamura92*>(sp.getPijAccelerator()->getReplacementModel()))->changeTheta(_bestTheta);
// optimize branch lengths
bblEM bblEM1(et,sc,sp,NULL,maxBBLIterations,epsilonLoglikelihoodForBBL);//maxIterations=1000
newL =bblEM1.getTreeLikelihood();
// check for improvement in the likelihood
if (newL > oldL+epsilonLikelihoodImprovment) {
prevTrTv = _bestTrTv;
prevTheta = _bestTheta;
oldL = newL;
_bestL = newL;
prevTree = et;
} else {
if (newL>oldL) {
_bestL = newL;
} else {
_bestL = oldL;
_bestTrTv = prevTrTv;
_bestTheta = prevTheta;
et = prevTree;
}
break;
}
}
}
bestTamura92ParamAlphaAndBBL::bestTamura92ParamAlphaAndBBL( //find best TrTv, theta, Alpha and best branch lengths
tree& et,
const sequenceContainer& sc,
stochasticProcess& sp,
const Vdouble * weights,
const int maxTotalIterations,
const MDOUBLE epsilonLikelihoodImprovment,
const MDOUBLE epsilonLoglikelihoodForTrTvOptimization,
const MDOUBLE epsilonLoglikelihoodForThetaOptimization,
const MDOUBLE epsilonLoglikelihoodForAlphaOptimization,
const MDOUBLE epsilonLoglikelihoodForBBL,
const MDOUBLE upperBoundOnTrTv,
const int maxBBLIterations,
const MDOUBLE initAlpha,
const MDOUBLE upperBoundOnAlpha)
{
MDOUBLE oldL = VERYSMALL;
MDOUBLE newL = VERYSMALL;
// first guess for the parameters
MDOUBLE prevTrTv = static_cast<tamura92*>(sp.getPijAccelerator()->getReplacementModel())->getTrTv();
MDOUBLE prevTheta = static_cast<tamura92*>(sp.getPijAccelerator()->getReplacementModel())->getTheta();
MDOUBLE prevAlpha = initAlpha;
tree prevTree;
for (int i=0; i < maxTotalIterations; ++i) {
// optimize TrTv
newL = -brent(0.0, prevTrTv, upperBoundOnTrTv,
C_evalTrTvParam(et,sc,sp,weights),
epsilonLoglikelihoodForTrTvOptimization,
&_bestTrTv);
(static_cast<tamura92*>(sp.getPijAccelerator()->getReplacementModel()))->changeTrTv(_bestTrTv);
// optimize Theta
newL = -brent(0.0, prevTheta, 1.0,
C_evalTheta(et,sc,sp,weights),
epsilonLoglikelihoodForThetaOptimization,
&_bestTheta);
(static_cast<tamura92*>(sp.getPijAccelerator()->getReplacementModel()))->changeTheta(_bestTheta);
// optimize Alpha
newL = -brent(0.0, prevAlpha, upperBoundOnAlpha,
C_evalAlpha(et,sc,sp,weights),
epsilonLoglikelihoodForAlphaOptimization,
&_bestAlpha);
(static_cast<gammaDistribution*>(sp.distr()))->setAlpha(_bestAlpha);
LOG(5,<<"# bestTamura92ParamAlphaAndBBL::bestTamura92ParamAlphaAndBBL iteration " << i << ": after param optimization:" <<endl
<<"# old L = " << oldL << "\t"
<<"# new L = " << newL << endl
<<"# new Alpha = " << _bestAlpha << endl);
// optimize branch lengths
bblEM bblEM1(et,sc,sp,NULL,maxBBLIterations,epsilonLoglikelihoodForBBL);//maxIterations=1000
newL =bblEM1.getTreeLikelihood();
LOG(5,<<"# bestTamura92ParamAlphaAndBBL::bestTamura92ParamAlphaAndBBL iteration " << i << ": after branch lengths optimization:" <<endl
<<"# After BBL new L = "<<newL<<" old L = "<<oldL<<endl
<<"# The tree:" );
LOGDO(5,et.output(myLog::LogFile()));
// check for improvement in the likelihood
if (newL > oldL+epsilonLikelihoodImprovment) {
oldL = newL;
_bestL = newL;
prevTrTv = _bestTrTv;
prevTheta = _bestTheta;
prevAlpha = _bestAlpha;
prevTree = et;
} else {
if (newL>oldL) {
_bestL = newL;
} else {
_bestL = oldL;
_bestTrTv = prevTrTv;
_bestTheta = prevTheta;
et = prevTree;
}
break;
}
}
}
bestTamura92ParamAlphaAndBBLProportional::bestTamura92ParamAlphaAndBBLProportional( //find best TrTv, theta, loca Alpha for each gene, global Alpha and best branch lengths
tree& et,
vector<sequenceContainer>& sc,
multipleStochasticProcess* msp,
gammaDistribution* pProportionDist,
Vdouble initLocalAlphas,
Vdouble initLocalKappas,
Vdouble initLocalThetas,
const MDOUBLE upperBoundOnLocalAlpha,
const MDOUBLE initGlobalAlpha,
const MDOUBLE upperBoundOnGlobalAlpha,
const MDOUBLE upperBoundOnTrTv,
const int maxTotalIterations,
const int maxBBLIterations,
const bool optimizeSelectedBranches,
const bool optimizeTree,
const string branchLengthOptimizationMethod,
const bool optimizeLocalParams,
const bool optimizeGlobalAlpha,
const Vdouble * weights,
const MDOUBLE epsilonLikelihoodImprovment,
const MDOUBLE epsilonLoglikelihoodForLocalTrTvOptimization,
const MDOUBLE epsilonLoglikelihoodForLocalThetaOptimization,
const MDOUBLE epsilonLoglikelihoodForLocalAlphaOptimization,
const MDOUBLE epsilonLoglikelihoodForGlobalAlphaOptimization,
const MDOUBLE epsilonLoglikelihoodForBBL)
{
LOG(5,<<"Starting bestTamura92ParamAlphaAndBBLProportional"<<endl);
Vdouble currentTrTvVec,currentThetaVec,currentLocalAlphaVec;
MDOUBLE currentGlobalAlpha = initGlobalAlpha;
currentTrTvVec = initLocalKappas;
currentThetaVec = initLocalThetas;
currentLocalAlphaVec = initLocalAlphas;
Vdouble newLvec;
newLvec.resize(msp->getSPVecSize());
//doubleRep oldL(VERYSMALL);//DR
//doubleRep newL;//DR
MDOUBLE oldL = VERYSMALL;
MDOUBLE newL;
//doubleRep epsilonLoglikelihoodForGlobalAlphaOptimizationDR(epsilonLoglikelihoodForGlobalAlphaOptimization);//DR
_bestLvec.resize(msp->getSPVecSize(),0.0);
_bestLocalAlphaVec = initLocalAlphas;
_bestGlobalAlpha = initGlobalAlpha;
int spIndex;
_bestTrTvVec = currentTrTvVec;
_bestThetaVec = currentThetaVec;
pProportionDist->setAlpha(_bestGlobalAlpha);
for(spIndex = 0;spIndex < msp->getSPVecSize();++spIndex){
(static_cast<tamura92*>(msp->getSp(spIndex)->getPijAccelerator()->getReplacementModel()))->changeTheta(_bestThetaVec[spIndex]);//safety
(static_cast<tamura92*>(msp->getSp(spIndex)->getPijAccelerator()->getReplacementModel()))->changeTrTv(_bestTrTvVec[spIndex]);
(static_cast<gammaDistribution*>(msp->getSp(spIndex)->distr()))->setAlpha(_bestLocalAlphaVec[spIndex]);
}
//first compute the likelihood;
_bestLvec = likelihoodComputation::getTreeLikelihoodProportionalAllPosAlphTheSame(et,sc,msp,pProportionDist,weights);
MDOUBLE ax_local = 0.0;
MDOUBLE c_TrTv_x = upperBoundOnTrTv;
MDOUBLE c_theta_x = 1.0;
MDOUBLE c_localAlpha_x = upperBoundOnLocalAlpha;
for (int i=0; i < maxTotalIterations; ++i) {
if(optimizeLocalParams){
for(spIndex = 0;spIndex < msp->getSPVecSize();++spIndex){
//optimize Theta
MDOUBLE theta_x(_bestThetaVec[spIndex]);
newLvec[spIndex] = -brent(ax_local,theta_x,c_theta_x,
C_evalLocalTheta(et,sc[spIndex],*msp->getSp(spIndex),pProportionDist,weights),
epsilonLoglikelihoodForLocalThetaOptimization,
¤tThetaVec[spIndex]);
if (newLvec[spIndex] >= _bestLvec[spIndex])
{
_bestLvec[spIndex] = newLvec[spIndex];
_bestThetaVec[spIndex] = currentThetaVec[spIndex];
}
else
{//likelihood went down!
LOG(2,<<"likelihood went down in optimizing TrTv param"<<endl<<"oldL = "<<sumVdouble(_bestLvec));
}
(static_cast<tamura92*>(msp->getSp(spIndex)->getPijAccelerator()->getReplacementModel()))->changeTheta(_bestThetaVec[spIndex]);//safety
//optimize TrTv
MDOUBLE TrTv_x(_bestTrTvVec[spIndex]);
newLvec[spIndex] = -brent(ax_local,TrTv_x,c_TrTv_x,
C_evalLocalTrTvParam(et,sc[spIndex],*msp->getSp(spIndex),pProportionDist,weights),
epsilonLoglikelihoodForLocalTrTvOptimization,
¤tTrTvVec[spIndex]);
if (newLvec[spIndex] >= _bestLvec[spIndex])
{
_bestLvec[spIndex] = newLvec[spIndex];
_bestTrTvVec[spIndex] = currentTrTvVec[spIndex];
}
else
{//likelihood went down!
LOG(2,<<"likelihood went down in optimizing TrTv param"<<endl<<"oldL = "<<sumVdouble(_bestLvec));
}
(static_cast<tamura92*>(msp->getSp(spIndex)->getPijAccelerator()->getReplacementModel()))->changeTrTv(_bestTrTvVec[spIndex]);//safety
//optimize local alpha
MDOUBLE localAlpha_x(_bestLocalAlphaVec[spIndex]);
newLvec[spIndex] = -brent(ax_local,localAlpha_x, c_localAlpha_x,
C_evalLocalAlpha(et,sc[spIndex],*msp->getSp(spIndex),pProportionDist,weights),
epsilonLoglikelihoodForLocalAlphaOptimization,
¤tLocalAlphaVec[spIndex]);
if (newLvec[spIndex] >= _bestLvec[spIndex])
{
_bestLvec[spIndex] = newLvec[spIndex];
_bestLocalAlphaVec[spIndex] = currentLocalAlphaVec[spIndex];
}
else
{//likelihood went down!
LOG(2,<<"likelihood went down in optimizing local alpha"<<endl<<"oldL = "<<sumVdouble(_bestLvec));
}
(static_cast<gammaDistribution*>(msp->getSp(spIndex)->distr()))->setAlpha(_bestLocalAlphaVec[spIndex]); //safety
}
LOGnOUT(2,<<"Done with Tamura92 local params optimization. LL: "<<sumVdouble(_bestLvec)<<endl);
LOGnOUT(2,<<"Local Params:"<<endl);
LOGnOUT(2,<<"TrTv:");
for(spIndex = 0;spIndex < _bestTrTvVec.size();++spIndex){
LOGnOUT(2,<<_bestTrTvVec[spIndex]<<",";);
}
LOGnOUT(2,<<endl);
LOGnOUT(2,<<"Theta:");
for(spIndex = 0;spIndex < _bestThetaVec.size();++spIndex){
LOGnOUT(2,<<_bestThetaVec[spIndex]<<",";);
}
LOGnOUT(2,<<endl);
LOGnOUT(2,<<"local alpha:");
for(spIndex = 0;spIndex < _bestLocalAlphaVec.size();++spIndex){
LOGnOUT(2,<<_bestLocalAlphaVec[spIndex]<<",";);
}
LOGnOUT(2,<<endl);
}
if(optimizeGlobalAlpha){
//doubleRep ax_global(0.0);//DR
//doubleRep c_globalAlpha_x(upperBoundOnGlobalAlpha);//DR
//doubleRep minusOne(-1.0);//DR
MDOUBLE ax_global = 0.0;
MDOUBLE c_globalAlpha_x = upperBoundOnGlobalAlpha;
//optimize global alpha
//doubleRep globalAlpha_x(prevGlobalAlpha);//DR
MDOUBLE globalAlpha_x = _bestGlobalAlpha;
//newL = minusOne*brentDoubleRep(ax_global,globalAlpha_x,c_globalAlpha_x,
// C_evalGlobalAlpha(et,sc,msp,pProportionDist,weights),
// epsilonLoglikelihoodForGlobalAlphaOptimizationDR,
// &_bestGlobalAlpha);//DR
newL = -brent(ax_global,globalAlpha_x,c_globalAlpha_x,
C_evalGlobalAlpha(et,sc,msp,pProportionDist,weights),
epsilonLoglikelihoodForGlobalAlphaOptimization,
¤tGlobalAlpha);
if (newL >= sumVdouble(_bestLvec))
{
_bestGlobalAlpha = currentGlobalAlpha;
}
else
{//likelihood went down!
LOG(2,<<"likelihood went down in optimizing global alpha"<<endl<<"oldL = "<<sumVdouble(_bestLvec));
}
pProportionDist->setAlpha(_bestGlobalAlpha); //safety
//whether or not likelihood has improved we need to update _bestLvec
_bestLvec = likelihoodComputation::getTreeLikelihoodProportionalAllPosAlphTheSame(et,sc,msp,pProportionDist,weights);
LOGnOUT(2,<<"Done with global alpha optimization"<<endl<<"LL:"<<sumVdouble(_bestLvec)<<endl);
LOGnOUT(2,<<"Global Alpha:"<<_bestGlobalAlpha<<endl);
}
if(optimizeTree)
{
if(branchLengthOptimizationMethod == "bblLS"){
bblLSProportionalEB bblLSPEB1(et,sc,msp,pProportionDist,_bestLvec,optimizeSelectedBranches,maxBBLIterations,epsilonLoglikelihoodForBBL);
_bestLvec = bblLSPEB1.getTreeLikelihoodVec();
LOGnOUT(2,<<"Done with bblLS"<<endl<<"LL:"<<sumVdouble(_bestLvec)<<endl);
}
else if(branchLengthOptimizationMethod == "bblEM"){
bblEMProportionalEB bblEMPEB1(et,sc,msp,pProportionDist,optimizeSelectedBranches,NULL,maxBBLIterations,epsilonLoglikelihoodForBBL);
_bestLvec = bblEMPEB1.getTreeLikelihood();
LOGnOUT(2,<<"Done with bblEM. LL: "<<sumVdouble(_bestLvec)<<endl);
}
LOGnOUT(2,<<et.stringTreeInPhylipTreeFormat()<<endl);
}
// check for improvement in the likelihood
if (sumVdouble(_bestLvec) > oldL+epsilonLikelihoodImprovment) {
//all params have already been updated
oldL = sumVdouble(_bestLvec);
} else {
break;
}
LOGnOUT(4,<<"Done with optimization iteration "<<i<<". LL: "<<sumVdouble(_bestLvec)<<endl);
}
}
|
