bayesian.cpp

   1 /*
   2  *  bayesian.cpp
   3  *  Mothur
   4  *
   5  *  Created by westcott on 11/3/09.
   6  *  Copyright 2009 Schloss Lab. All rights reserved.
   7  *
   8  */
   9
  10 #include "bayesian.h"
  11 #include "kmer.hpp"
  12 #include "phylosummary.h"
  13
  14 /**************************************************************************************************/
  15 Bayesian::Bayesian(string tfile, string tempFile, string method, int ksize, int cutoff, int i) :
  16 Classify(), kmerSize(ksize), confidenceThreshold(cutoff), iters(i)  {
  17         try {
  18
  19                 /************calculate the probablity that each word will be in a specific taxonomy*************/
  20                 string tfileroot = tfile.substr(0,tfile.find_last_of(".")+1);
  21                 string tempfileroot = getRootName(getSimpleName(tempFile));
  22                 string phyloTreeName = tfileroot + "tree.train";
  23                 string probFileName = tfileroot + tempfileroot + char('0'+ kmerSize) + "mer.prob";
  24                 string probFileName2 = tfileroot + tempfileroot + char('0'+ kmerSize) + "mer.numNonZero";
  25
  26                 ofstream out;
  27                 ofstream out2;
  28
  29                 ifstream phyloTreeTest(phyloTreeName.c_str());
  30                 ifstream probFileTest2(probFileName2.c_str());
  31                 ifstream probFileTest(probFileName.c_str());
  32
  33                 int start = time(NULL);
  34
  35                 if(probFileTest && probFileTest2 && phyloTreeTest){
  36                         m->mothurOut("Reading template taxonomy...     "); cout.flush();
  37
  38                         phyloTree = new PhyloTree(phyloTreeTest, phyloTreeName);
  39
  40                         m->mothurOut("DONE."); m->mothurOutEndLine();
  41
  42                         genusNodes = phyloTree->getGenusNodes();
  43                         genusTotals = phyloTree->getGenusTotals();
  44
  45                         m->mothurOut("Reading template probabilities...     "); cout.flush();
  46                         readProbFile(probFileTest, probFileTest2, probFileName, probFileName2);
  47
  48                 }else{
  49
  50                         //create search database and names vector
  51                         generateDatabaseAndNames(tfile, tempFile, method, ksize, 0.0, 0.0, 0.0, 0.0);
  52
  53                         //prevents errors caused by creating shortcut files if you had an error in the sanity check.
  54                         if (m->control_pressed) {  remove(phyloTreeName.c_str());  remove(probFileName.c_str()); remove(probFileName2.c_str()); }
  55                         else{
  56                                 genusNodes = phyloTree->getGenusNodes();
  57                                 genusTotals = phyloTree->getGenusTotals();
  58
  59                                 m->mothurOut("Calculating template taxonomy tree...     "); cout.flush();
  60
  61                                 phyloTree->printTreeNodes(phyloTreeName);
  62
  63                                 m->mothurOut("DONE."); m->mothurOutEndLine();
  64
  65                                 m->mothurOut("Calculating template probabilities...     "); cout.flush();
  66
  67                                 numKmers = database->getMaxKmer() + 1;
  68
  69                                 //initialze probabilities
  70                                 wordGenusProb.resize(numKmers);
  71
  72                                 for (int j = 0; j < wordGenusProb.size(); j++) {        wordGenusProb[j].resize(genusNodes.size());             }
  73
  74                                 ofstream out;
  75                                 ofstream out2;
  76
  77                                 #ifdef USE_MPI
  78                                         int pid;
  79                                         MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
  80
  81                                         if (pid == 0) {
  82                                 #endif
  83
  84
  85                                 openOutputFile(probFileName, out);
  86
  87                                 out << numKmers << endl;
  88
  89                                 openOutputFile(probFileName2, out2);
  90
  91                                 #ifdef USE_MPI
  92                                         }
  93                                 #endif
  94
  95
  96                                 //for each word
  97                                 for (int i = 0; i < numKmers; i++) {
  98                                         if (m->control_pressed) {  break; }
  99
 100                                         #ifdef USE_MPI
 101                                                 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
 102
 103                                                 if (pid == 0) {
 104                                         #endif
 105
 106                                         out << i << '\t';
 107
 108                                         #ifdef USE_MPI
 109                                                 }
 110                                         #endif
 111
 112                                         vector<int> seqsWithWordi = database->getSequencesWithKmer(i);
 113
 114                                         map<int, int> count;
 115                                         for (int k = 0; k < genusNodes.size(); k++) {  count[genusNodes[k]] = 0;  }
 116
 117                                         //for each sequence with that word
 118                                         for (int j = 0; j < seqsWithWordi.size(); j++) {
 119                                                 int temp = phyloTree->getIndex(names[seqsWithWordi[j]]);
 120                                                 count[temp]++;  //increment count of seq in this genus who have this word
 121                                         }
 122
 123                                         //probabilityInTemplate = (# of seqs with that word in template + 0.05) / (total number of seqs in template + 1);
 124                                         float probabilityInTemplate = (seqsWithWordi.size() + 0.50) / (float) (names.size() + 1);
 125
 126                                         int numNotZero = 0;
 127                                         for (int k = 0; k < genusNodes.size(); k++) {
 128                                                 //probabilityInThisTaxonomy = (# of seqs with that word in this taxonomy + probabilityInTemplate) / (total number of seqs in this taxonomy + 1);
 129                                                 wordGenusProb[i][k] = log((count[genusNodes[k]] + probabilityInTemplate) / (float) (genusTotals[k] + 1));
 130                                                 if (count[genusNodes[k]] != 0) {
 131                                                         #ifdef USE_MPI
 132                                                                 int pid;
 133                                                                 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
 134
 135                                                                 if (pid == 0) {
 136                                                         #endif
 137
 138                                                         out << k << '\t' << wordGenusProb[i][k] << '\t';
 139
 140                                                         #ifdef USE_MPI
 141                                                                 }
 142                                                         #endif
 143
 144                                                         numNotZero++;
 145                                                 }
 146                                         }
 147
 148                                         #ifdef USE_MPI
 149                                                 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
 150
 151                                                 if (pid == 0) {
 152                                         #endif
 153
 154                                         out << endl;
 155                                         out2 << probabilityInTemplate << '\t' << numNotZero << endl;
 156
 157                                         #ifdef USE_MPI
 158                                                 }
 159                                         #endif
 160                                 }
 161
 162                                 #ifdef USE_MPI
 163                                         MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
 164
 165                                         if (pid == 0) {
 166                                 #endif
 167
 168                                 out.close();
 169                                 out2.close();
 170
 171                                 #ifdef USE_MPI
 172                                         }
 173                                 #endif
 174
 175                                 //read in new phylotree with less info. - its faster
 176                                 ifstream phyloTreeTest(phyloTreeName.c_str());
 177                                 delete phyloTree;
 178
 179                                 phyloTree = new PhyloTree(phyloTreeTest, phyloTreeName);
 180                         }
 181                 }
 182
 183                 m->mothurOut("DONE."); m->mothurOutEndLine();
 184                 m->mothurOut("It took " + toString(time(NULL) - start) + " seconds get probabilities. "); m->mothurOutEndLine();
 185         }
 186         catch(exception& e) {
 187                 m->errorOut(e, "Bayesian", "Bayesian");
 188                 exit(1);
 189         }
 190 }
 191 /**************************************************************************************************/
 192 Bayesian::~Bayesian() {
 193         try {
 194                  delete phyloTree;
 195                  if (database != NULL) {  delete database; }
 196         }
 197         catch(exception& e) {
 198                 m->errorOut(e, "Bayesian", "~Bayesian");
 199                 exit(1);
 200         }
 201 }
 202
 203 /**************************************************************************************************/
 204 string Bayesian::getTaxonomy(Sequence* seq) {
 205         try {
 206                 string tax = "";
 207                 Kmer kmer(kmerSize);
 208
 209                 //get words contained in query
 210                 //getKmerString returns a string where the index in the string is hte kmer number
 211                 //and the character at that index can be converted to be the number of times that kmer was seen
 212
 213                 string queryKmerString = kmer.getKmerString(seq->getUnaligned());
 214
 215                 vector<int> queryKmers;
 216                 for (int i = 0; i < queryKmerString.length(); i++) {
 217                         if (queryKmerString[i] != '!') { //this kmer is in the query
 218                                 queryKmers.push_back(i);
 219                         }
 220                 }
 221
 222                 if (queryKmers.size() == 0) {  m->mothurOut(seq->getName() + "is bad."); m->mothurOutEndLine(); return "bad seq"; }
 223
 224                 int index = getMostProbableTaxonomy(queryKmers);
 225
 226                 if (m->control_pressed) { return tax; }
 227 //cout << seq->getName() << '\t' << index << endl;
 228                 //bootstrap - to set confidenceScore
 229                 int numToSelect = queryKmers.size() / 8;
 230                 tax = bootstrapResults(queryKmers, index, numToSelect);
 231
 232                 return tax;
 233         }
 234         catch(exception& e) {
 235                 m->errorOut(e, "Bayesian", "getTaxonomy");
 236                 exit(1);
 237         }
 238 }
 239 /**************************************************************************************************/
 240 string Bayesian::bootstrapResults(vector<int> kmers, int tax, int numToSelect) {
 241         try {
 242
 243                 map<int, int> confidenceScores;
 244
 245                 map<int, int>::iterator itBoot;
 246                 map<int, int>::iterator itBoot2;
 247                 map<int, int>::iterator itConvert;
 248
 249                 for (int i = 0; i < iters; i++) {
 250                         if (m->control_pressed) { return "control"; }
 251
 252                         vector<int> temp;
 253
 254                         for (int j = 0; j < numToSelect; j++) {
 255                                 int index = int(rand() % kmers.size());
 256
 257                                 //add word to temp
 258                                 temp.push_back(kmers[index]);
 259                         }
 260
 261                         //get taxonomy
 262                         int newTax = getMostProbableTaxonomy(temp);
 263                         TaxNode taxonomyTemp = phyloTree->get(newTax);
 264
 265                         //add to confidence results
 266                         while (taxonomyTemp.level != 0) { //while you are not at the root
 267
 268                                 itBoot2 = confidenceScores.find(newTax); //is this a classification we already have a count on
 269
 270                                 if (itBoot2 == confidenceScores.end()) { //not already in confidence scores
 271                                         confidenceScores[newTax] = 1;
 272                                 }else{
 273                                         confidenceScores[newTax]++;
 274                                 }
 275
 276                                 newTax = taxonomyTemp.parent;
 277                                 taxonomyTemp = phyloTree->get(newTax);
 278                         }
 279
 280                 }
 281
 282                 string confidenceTax = "";
 283                 simpleTax = "";
 284
 285                 int seqTaxIndex = tax;
 286                 TaxNode seqTax = phyloTree->get(tax);
 287
 288                 while (seqTax.level != 0) { //while you are not at the root
 289
 290                                 itBoot2 = confidenceScores.find(seqTaxIndex); //is this a classification we already have a count on
 291
 292                                 int confidence = 0;
 293                                 if (itBoot2 != confidenceScores.end()) { //already in confidence scores
 294                                         confidence = confidenceScores[seqTaxIndex];
 295                                 }
 296
 297                                 if (((confidence/(float)iters) * 100) >= confidenceThreshold) {
 298                                         confidenceTax = seqTax.name + "(" + toString(((confidence/(float)iters) * 100)) + ");" + confidenceTax;
 299                                         simpleTax = seqTax.name + ";" + simpleTax;
 300                                 }
 301
 302                                 seqTaxIndex = seqTax.parent;
 303                                 seqTax = phyloTree->get(seqTax.parent);
 304                 }
 305
 306                 if (confidenceTax == "") { confidenceTax = "unclassified;"; simpleTax = "unclassified;"; }
 307                 return confidenceTax;
 308
 309         }
 310         catch(exception& e) {
 311                 m->errorOut(e, "Bayesian", "bootstrapResults");
 312                 exit(1);
 313         }
 314 }
 315 /**************************************************************************************************/
 316 int Bayesian::getMostProbableTaxonomy(vector<int> queryKmer) {
 317         try {
 318                 int indexofGenus = 0;
 319
 320                 double maxProbability = -1000000.0;
 321                 //find taxonomy with highest probability that this sequence is from it
 322                 for (int k = 0; k < genusNodes.size(); k++) {
 323                         //for each taxonomy calc its probability
 324                         double prob = 1.0;
 325                         for (int i = 0; i < queryKmer.size(); i++) {
 326                                 prob += wordGenusProb[queryKmer[i]][k];
 327                         }
 328
 329                         //is this the taxonomy with the greatest probability?
 330                         if (prob > maxProbability) {
 331                                 indexofGenus = genusNodes[k];
 332                                 maxProbability = prob;
 333                         }
 334                 }
 335
 336                 return indexofGenus;
 337         }
 338         catch(exception& e) {
 339                 m->errorOut(e, "Bayesian", "getMostProbableTaxonomy");
 340                 exit(1);
 341         }
 342 }
 343 /*************************************************************************************************
 344 map<string, int> Bayesian::parseTaxMap(string newTax) {
 345         try{
 346
 347                 map<string, int> parsed;
 348
 349                 newTax = newTax.substr(0, newTax.length()-1);  //get rid of last ';'
 350
 351                 //parse taxonomy
 352                 string individual;
 353                 while (newTax.find_first_of(';') != -1) {
 354                         individual = newTax.substr(0,newTax.find_first_of(';'));
 355                         newTax = newTax.substr(newTax.find_first_of(';')+1, newTax.length());
 356                         parsed[individual] = 1;
 357                 }
 358
 359                 //get last one
 360                 parsed[newTax] = 1;
 361
 362                 return parsed;
 363
 364         }
 365         catch(exception& e) {
 366                 m->errorOut(e, "Bayesian", "parseTax");
 367                 exit(1);
 368         }
 369 }
 370 /**************************************************************************************************/
 371 void Bayesian::readProbFile(ifstream& in, ifstream& inNum, string inName, string inNumName) {
 372         try{
 373
 374                 #ifdef USE_MPI
 375
 376                         int pid, num, num2, processors;
 377                         vector<unsigned long int> positions;
 378                         vector<unsigned long int> positions2;
 379
 380                         MPI_Status status;
 381                         MPI_File inMPI;
 382                         MPI_File inMPI2;
 383                         MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
 384                         MPI_Comm_size(MPI_COMM_WORLD, &processors);
 385                         int tag = 2001;
 386
 387                         char inFileName[1024];
 388                         strcpy(inFileName, inNumName.c_str());
 389
 390                         char inFileName2[1024];
 391                         strcpy(inFileName2, inName.c_str());
 392
 393                         MPI_File_open(MPI_COMM_WORLD, inFileName, MPI_MODE_RDONLY, MPI_INFO_NULL, &inMPI);  //comm, filename, mode, info, filepointer
 394                         MPI_File_open(MPI_COMM_WORLD, inFileName2, MPI_MODE_RDONLY, MPI_INFO_NULL, &inMPI2);  //comm, filename, mode, info, filepointer
 395
 396                         if (pid == 0) {
 397                                 positions = setFilePosEachLine(inNumName, num);
 398                                 positions2 = setFilePosEachLine(inName, num2);
 399
 400                                 for(int i = 1; i < processors; i++) {
 401                                         MPI_Send(&num, 1, MPI_INT, i, tag, MPI_COMM_WORLD);
 402                                         MPI_Send(&positions[0], (num+1), MPI_LONG, i, tag, MPI_COMM_WORLD);
 403
 404                                         MPI_Send(&num2, 1, MPI_INT, i, tag, MPI_COMM_WORLD);
 405                                         MPI_Send(&positions2[0], (num2+1), MPI_LONG, i, tag, MPI_COMM_WORLD);
 406                                 }
 407
 408                         }else{
 409                                 MPI_Recv(&num, 1, MPI_INT, 0, tag, MPI_COMM_WORLD, &status);
 410                                 positions.resize(num+1);
 411                                 MPI_Recv(&positions[0], (num+1), MPI_LONG, 0, tag, MPI_COMM_WORLD, &status);
 412
 413                                 MPI_Recv(&num2, 1, MPI_INT, 0, tag, MPI_COMM_WORLD, &status);
 414                                 positions2.resize(num2+1);
 415                                 MPI_Recv(&positions2[0], (num2+1), MPI_LONG, 0, tag, MPI_COMM_WORLD, &status);
 416                         }
 417
 418                         //read numKmers
 419                         int length = positions2[1] - positions2[0];
 420                         char* buf = new char[length];
 421
 422                         MPI_File_read_at(inMPI2, positions2[0], buf, length, MPI_CHAR, &status);
 423
 424                         string tempBuf = buf;
 425                         if (tempBuf.length() > length) { tempBuf = tempBuf.substr(0, length); }
 426                         delete buf;
 427
 428                         istringstream iss (tempBuf,istringstream::in);
 429                         iss >> numKmers;
 430
 431                         //initialze probabilities
 432                         wordGenusProb.resize(numKmers);
 433
 434                         for (int j = 0; j < wordGenusProb.size(); j++) {        wordGenusProb[j].resize(genusNodes.size());             }
 435
 436                         int kmer, name;
 437                         vector<int> numbers; numbers.resize(numKmers);
 438                         float prob;
 439                         vector<float> zeroCountProb; zeroCountProb.resize(numKmers);
 440
 441                         //read file
 442                         for(int i=0;i<num;i++){
 443                                 //read next sequence
 444                                 length = positions[i+1] - positions[i];
 445                                 char* buf4 = new char[length];
 446
 447                                 MPI_File_read_at(inMPI, positions[i], buf4, length, MPI_CHAR, &status);
 448
 449                                 tempBuf = buf4;
 450                                 if (tempBuf.length() > length) { tempBuf = tempBuf.substr(0, length); }
 451                                 delete buf4;
 452
 453                                 istringstream iss (tempBuf,istringstream::in);
 454                                 iss >> zeroCountProb[i] >> numbers[i];
 455                         }
 456
 457                         MPI_File_close(&inMPI);
 458
 459                         for(int i=1;i<num2;i++){
 460                                 //read next sequence
 461                                 length = positions2[i+1] - positions2[i];
 462                                 char* buf4 = new char[length];
 463
 464                                 MPI_File_read_at(inMPI2, positions2[i], buf4, length, MPI_CHAR, &status);
 465
 466                                 tempBuf = buf4;
 467                                 if (tempBuf.length() > length) { tempBuf = tempBuf.substr(0, length); }
 468                                 delete buf4;
 469
 470                                 istringstream iss (tempBuf,istringstream::in);
 471
 472                                 iss >> kmer;
 473
 474                                 //set them all to zero value
 475                                 for (int i = 0; i < genusNodes.size(); i++) {
 476                                         wordGenusProb[kmer][i] = log(zeroCountProb[kmer] / (float) (genusTotals[i]+1));
 477                                 }
 478
 479                                 //get probs for nonzero values
 480                                 for (int i = 0; i < numbers[kmer]; i++) {
 481                                         iss >> name >> prob;
 482                                         wordGenusProb[kmer][name] = prob;
 483                                 }
 484
 485                         }
 486                         MPI_File_close(&inMPI2);
 487                         MPI_Barrier(MPI_COMM_WORLD); //make everyone wait - just in case
 488                 #else
 489
 490                         in >> numKmers; gobble(in);
 491
 492                         //initialze probabilities
 493                         wordGenusProb.resize(numKmers);
 494
 495                         for (int j = 0; j < wordGenusProb.size(); j++) {        wordGenusProb[j].resize(genusNodes.size());             }
 496
 497                         int kmer, name, count;  count = 0;
 498                         vector<int> num; num.resize(numKmers);
 499                         float prob;
 500                         vector<float> zeroCountProb; zeroCountProb.resize(numKmers);
 501
 502                         while (inNum) {
 503                                 inNum >> zeroCountProb[count] >> num[count];
 504                                 count++;
 505                                 gobble(inNum);
 506                         }
 507                         inNum.close();
 508
 509                         while(in) {
 510                                 in >> kmer;
 511
 512                                 //set them all to zero value
 513                                 for (int i = 0; i < genusNodes.size(); i++) {
 514                                         wordGenusProb[kmer][i] = log(zeroCountProb[kmer] / (float) (genusTotals[i]+1));
 515                                 }
 516
 517                                 //get probs for nonzero values
 518                                 for (int i = 0; i < num[kmer]; i++) {
 519                                         in >> name >> prob;
 520                                         wordGenusProb[kmer][name] = prob;
 521                                 }
 522
 523                                 gobble(in);
 524                         }
 525                         in.close();
 526
 527                 #endif
 528         }
 529         catch(exception& e) {
 530                 m->errorOut(e, "Bayesian", "readProbFile");
 531                 exit(1);
 532         }
 533 }
 534 /**************************************************************************************************/
 535
 536
 537
 538
 539
 540