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
  13 /**************************************************************************************************/
  14 Bayesian::Bayesian(string tfile, string tempFile, string method, int ksize, int cutoff, int i) :
  15 Classify(tfile, tempFile, method, ksize, 0.0, 0.0, 0.0, 0.0), kmerSize(ksize), confidenceThreshold(cutoff), iters(i)  {
  16         try {
  17
  18                 numKmers = database->getMaxKmer() + 1;
  19
  20                 //initialze probabilities
  21                 wordGenusProb.resize(numKmers);
  22
  23                 genusNodes = phyloTree->getGenusNodes();
  24
  25                 for (int j = 0; j < wordGenusProb.size(); j++) {        wordGenusProb[j].resize(genusNodes.size());             }
  26
  27                 //reset counts because we are on a new word
  28                 for (int j = 0; j < genusNodes.size(); j++) {
  29                         TaxNode temp = phyloTree->get(genusNodes[j]);
  30                         genusTotals.push_back(temp.accessions.size());
  31                 }
  32
  33
  34                 /************calculate the probablity that each word will be in a specific taxonomy*************/
  35                 ofstream out;
  36                 string probFileName = tempFile.substr(0,tempFile.find_last_of(".")+1) + char('0'+ kmerSize) + "mer.prob";
  37                 ifstream probFileTest(probFileName.c_str());
  38
  39                 ofstream out2;
  40                 string probFileName2 = tempFile.substr(0,tempFile.find_last_of(".")+1) + char('0'+ kmerSize) + "mer.numNonZero";
  41                 ifstream probFileTest2(probFileName2.c_str());
  42
  43                 int start = time(NULL);
  44
  45                 if(probFileTest && probFileTest2){
  46                         mothurOut("Reading template probabilities...     "); cout.flush();
  47                         readProbFile(probFileTest, probFileTest2);
  48                 }else{
  49                         mothurOut("Calculating template probabilities...     "); cout.flush();
  50
  51                         ofstream out;
  52                         openOutputFile(probFileName, out);
  53
  54                         ofstream out2;
  55                         openOutputFile(probFileName2, out2);
  56
  57                         //for each word
  58                         for (int i = 0; i < numKmers; i++) {
  59
  60                                 out << i << '\t';
  61
  62                                 vector<int> seqsWithWordi = database->getSequencesWithKmer(i);
  63
  64                                 map<int, int> count;
  65                                 for (int k = 0; k < genusNodes.size(); k++) {  count[genusNodes[k]] = 0;  }
  66
  67                                 //for each sequence with that word
  68                                 for (int j = 0; j < seqsWithWordi.size(); j++) {
  69                                         int temp = phyloTree->getIndex(names[seqsWithWordi[j]]);
  70                                         count[temp]++;  //increment count of seq in this genus who have this word
  71                                 }
  72
  73                                 //probabilityInTemplate = (# of seqs with that word in template + 0.05) / (total number of seqs in template + 1);
  74                                 float probabilityInTemplate = (seqsWithWordi.size() + 0.50) / (float) (names.size() + 1);
  75
  76                                 int numNotZero = 0;
  77                                 for (int k = 0; k < genusNodes.size(); k++) {
  78                                         //probabilityInThisTaxonomy = (# of seqs with that word in this taxonomy + probabilityInTemplate) / (total number of seqs in this taxonomy + 1);
  79                                         wordGenusProb[i][k] = log((count[genusNodes[k]] + probabilityInTemplate) / (float) (genusTotals[k] + 1));
  80                                         if (count[genusNodes[k]] != 0) {  out << k << '\t' << wordGenusProb[i][k] << '\t';  numNotZero++;  }
  81                                 }
  82                                 out << endl;
  83                                 out2 << probabilityInTemplate << '\t' << numNotZero << endl;
  84                         }
  85
  86                         out.close();
  87                         out2.close();
  88                 }
  89
  90
  91                 mothurOut("DONE."); mothurOutEndLine();
  92                 mothurOut("It took " + toString(time(NULL) - start) + " seconds get probabilities. "); mothurOutEndLine();
  93         }
  94         catch(exception& e) {
  95                 errorOut(e, "Bayesian", "getTaxonomy");
  96                 exit(1);
  97         }
  98 }
  99 /**************************************************************************************************/
 100 string Bayesian::getTaxonomy(Sequence* seq) {
 101         try {
 102                 string tax = "";
 103                 Kmer kmer(kmerSize);
 104
 105                 //get words contained in query
 106                 //getKmerString returns a string where the index in the string is hte kmer number
 107                 //and the character at that index can be converted to be the number of times that kmer was seen
 108                 string queryKmerString = kmer.getKmerString(seq->getUnaligned());
 109                 vector<int> queryKmers;
 110                 for (int i = 0; i < queryKmerString.length(); i++) {
 111                         if (queryKmerString[i] != '!') { //this kmer is in the query
 112                                 queryKmers.push_back(i);
 113                         }
 114                 }
 115
 116                 int index = getMostProbableTaxonomy(queryKmers);
 117
 118                 //bootstrap - to set confidenceScore
 119                 int numToSelect = queryKmers.size() / 8;
 120                 tax = bootstrapResults(queryKmers, index, numToSelect);
 121
 122                 return tax;
 123         }
 124         catch(exception& e) {
 125                 errorOut(e, "Bayesian", "getTaxonomy");
 126                 exit(1);
 127         }
 128 }
 129 /**************************************************************************************************/
 130 string Bayesian::bootstrapResults(vector<int> kmers, int tax, int numToSelect) {
 131         try {
 132
 133                 //taxConfidenceScore.clear(); //clear out previous seqs scores
 134
 135                 vector< map<string, int> > confidenceScores; //you need the added vector level of confusion to account for the level that name is seen since they can be the same
 136                                                                                 //map of classification to confidence for all areas seen
 137                                                                            //ie. Bacteria;Alphaproteobacteria;Rhizobiales;Azorhizobium_et_rel.;Methylobacterium_et_rel.;Bosea;
 138                                                                            //ie. Bacteria -> 100, Alphaproteobacteria -> 100, Rhizobiales -> 87, Azorhizobium_et_rel. -> 78, Methylobacterium_et_rel. -> 70, Bosea -> 50
 139                 confidenceScores.resize(100);  //if you have more than 100 levels of classification...
 140
 141                 map<string, int>::iterator itBoot;
 142                 map<string, int>::iterator itBoot2;
 143                 map<int, int>::iterator itConvert;
 144
 145                 for (int i = 0; i < iters; i++) {
 146                         vector<int> temp;
 147
 148                         for (int j = 0; j < numToSelect; j++) {
 149                                 int index = int(rand() % kmers.size());
 150
 151                                 //add word to temp
 152                                 temp.push_back(kmers[index]);
 153                         }
 154
 155                         //get taxonomy
 156                         int newTax = getMostProbableTaxonomy(temp);
 157                         TaxNode taxonomy = phyloTree->get(newTax);
 158
 159                         //add to confidence results
 160                         while (taxonomy.level != 0) { //while you are not at the root
 161
 162                                 itBoot2 = confidenceScores[taxonomy.level].find(taxonomy.name); //is this a classification we already have a count on
 163
 164                                 if (itBoot2 == confidenceScores[taxonomy.level].end()) { //not already in confidence scores
 165                                         confidenceScores[taxonomy.level][taxonomy.name] = 1;
 166                                 }else{
 167                                         confidenceScores[taxonomy.level][taxonomy.name]++;
 168                                 }
 169
 170                                 taxonomy = phyloTree->get(taxonomy.parent);
 171                         }
 172                 }
 173
 174                 string confidenceTax = "";
 175                 simpleTax = "";
 176                 TaxNode seqTax = phyloTree->get(tax);
 177
 178                 while (seqTax.level != 0) { //while you are not at the root
 179
 180                                 itBoot2 = confidenceScores[seqTax.level].find(seqTax.name); //is this a classification we already have a count on
 181
 182                                 int confidence = 0;
 183                                 if (itBoot2 != confidenceScores[seqTax.level].end()) { //not already in confidence scores
 184                                         confidence = confidenceScores[seqTax.level][seqTax.name];
 185                                 }
 186
 187                                 if (confidence >= confidenceThreshold) {
 188                                         confidenceTax = seqTax.name + "(" + toString(((confidence/(float)iters) * 100)) + ");" + confidenceTax;
 189                                         simpleTax = seqTax.name + ";" + simpleTax;
 190                                 }
 191
 192                                 seqTax = phyloTree->get(seqTax.parent);
 193                 }
 194
 195                 return confidenceTax;
 196
 197         }
 198         catch(exception& e) {
 199                 errorOut(e, "Bayesian", "bootstrapResults");
 200                 exit(1);
 201         }
 202 }
 203 /**************************************************************************************************/
 204 int Bayesian::getMostProbableTaxonomy(vector<int> queryKmer) {
 205         try {
 206                 int indexofGenus;
 207
 208                 double maxProbability = -1000000.0;
 209                 //find taxonomy with highest probability that this sequence is from it
 210                 for (int k = 0; k < genusNodes.size(); k++) {
 211
 212                         //for each taxonomy calc its probability
 213                         double prob = 1.0;
 214                         for (int i = 0; i < queryKmer.size(); i++) {
 215                                 prob += wordGenusProb[queryKmer[i]][k];
 216                         }
 217
 218                         //is this the taxonomy with the greatest probability?
 219                         if (prob > maxProbability) {
 220                                 indexofGenus = genusNodes[k];
 221                                 maxProbability = prob;
 222                         }
 223                 }
 224
 225                 return indexofGenus;
 226         }
 227         catch(exception& e) {
 228                 errorOut(e, "Bayesian", "getMostProbableTaxonomy");
 229                 exit(1);
 230         }
 231 }
 232 /*************************************************************************************************
 233 map<string, int> Bayesian::parseTaxMap(string newTax) {
 234         try{
 235
 236                 map<string, int> parsed;
 237
 238                 newTax = newTax.substr(0, newTax.length()-1);  //get rid of last ';'
 239
 240                 //parse taxonomy
 241                 string individual;
 242                 while (newTax.find_first_of(';') != -1) {
 243                         individual = newTax.substr(0,newTax.find_first_of(';'));
 244                         newTax = newTax.substr(newTax.find_first_of(';')+1, newTax.length());
 245                         parsed[individual] = 1;
 246                 }
 247
 248                 //get last one
 249                 parsed[newTax] = 1;
 250
 251                 return parsed;
 252
 253         }
 254         catch(exception& e) {
 255                 errorOut(e, "Bayesian", "parseTax");
 256                 exit(1);
 257         }
 258 }
 259 /**************************************************************************************************/
 260 void Bayesian::readProbFile(ifstream& in, ifstream& inNum) {
 261         try{
 262
 263                 int kmer, name, count;  count = 0;
 264                 vector<int> num; num.resize(numKmers);
 265                 float prob;
 266                 vector<float> zeroCountProb; zeroCountProb.resize(numKmers);
 267
 268                 while (inNum) {
 269                         inNum >> zeroCountProb[count] >> num[count];
 270                         count++;
 271                         gobble(inNum);
 272                 }
 273                 inNum.close();
 274
 275                 while(in) {
 276                         in >> kmer;
 277
 278                         //set them all to zero value
 279                         for (int i = 0; i < genusNodes.size(); i++) {
 280                                 wordGenusProb[kmer][i] = log(zeroCountProb[kmer] / (float) (genusTotals[i]+1));
 281                         }
 282
 283                         //get probs for nonzero values
 284                         for (int i = 0; i < num[kmer]; i++) {
 285                                 in >> name >> prob;
 286                                 wordGenusProb[kmer][name] = prob;
 287                         }
 288
 289                         gobble(in);
 290                 }
 291                 in.close();
 292         }
 293         catch(exception& e) {
 294                 errorOut(e, "Bayesian", "readProbFile");
 295                 exit(1);
 296         }
 297 }
 298 /**************************************************************************************************/
 299
 300
 301
 302
 303
 304