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