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