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