5 * Created by westcott on 11/3/09.
6 * Copyright 2009 Schloss Lab. All rights reserved.
12 #include "phylosummary.h"
13 #include "referencedb.h"
14 /**************************************************************************************************/
15 Bayesian::Bayesian(string tfile, string tempFile, string method, int ksize, int cutoff, int i, int tid, bool f, bool sh) :
16 Classify(), kmerSize(ksize), confidenceThreshold(cutoff), iters(i) {
18 ReferenceDB* rdb = ReferenceDB::getInstance();
23 string baseName = tempFile;
25 if (baseName == "saved") { baseName = rdb->getSavedReference(); }
27 string baseTName = tfile;
28 if (baseTName == "saved") { baseTName = rdb->getSavedTaxonomy(); }
30 /************calculate the probablity that each word will be in a specific taxonomy*************/
31 string tfileroot = m->getFullPathName(baseTName.substr(0,baseTName.find_last_of(".")+1));
32 string tempfileroot = m->getRootName(m->getSimpleName(baseName));
33 string phyloTreeName = tfileroot + "tree.train";
34 string phyloTreeSumName = tfileroot + "tree.sum";
35 string probFileName = tfileroot + tempfileroot + char('0'+ kmerSize) + "mer.prob";
36 string probFileName2 = tfileroot + tempfileroot + char('0'+ kmerSize) + "mer.numNonZero";
41 ifstream phyloTreeTest(phyloTreeName.c_str());
42 ifstream probFileTest2(probFileName2.c_str());
43 ifstream probFileTest(probFileName.c_str());
44 ifstream probFileTest3(phyloTreeSumName.c_str());
46 int start = time(NULL);
48 //if they are there make sure they were created after this release date
49 bool FilesGood = false;
50 if(probFileTest && probFileTest2 && phyloTreeTest && probFileTest3){
51 FilesGood = checkReleaseDate(probFileTest, probFileTest2, phyloTreeTest, probFileTest3);
54 //if you want to save, but you dont need to calculate then just read
55 if (rdb->save && probFileTest && probFileTest2 && phyloTreeTest && probFileTest3 && FilesGood && (tempFile != "saved")) {
57 m->openInputFile(tempFile, saveIn);
59 while (!saveIn.eof()) {
60 Sequence temp(saveIn);
63 rdb->referenceSeqs.push_back(temp);
68 if(probFileTest && probFileTest2 && phyloTreeTest && probFileTest3 && FilesGood){
69 if (tempFile == "saved") { m->mothurOutEndLine(); m->mothurOut("Using sequences from " + rdb->getSavedReference() + " that are saved in memory."); m->mothurOutEndLine(); }
71 m->mothurOut("Reading template taxonomy... "); cout.flush();
73 phyloTree = new PhyloTree(phyloTreeTest, phyloTreeName);
75 m->mothurOut("DONE."); m->mothurOutEndLine();
77 genusNodes = phyloTree->getGenusNodes();
78 genusTotals = phyloTree->getGenusTotals();
80 if (tfile == "saved") {
81 m->mothurOutEndLine(); m->mothurOut("Using probabilties from " + rdb->getSavedTaxonomy() + " that are saved in memory... "); cout.flush();;
82 wordGenusProb = rdb->wordGenusProb;
83 WordPairDiffArr = rdb->WordPairDiffArr;
85 m->mothurOut("Reading template probabilities... "); cout.flush();
86 readProbFile(probFileTest, probFileTest2, probFileName, probFileName2);
90 if (rdb->save) { rdb->wordGenusProb = wordGenusProb; rdb->WordPairDiffArr = WordPairDiffArr; }
93 //create search database and names vector
94 generateDatabaseAndNames(tfile, tempFile, method, ksize, 0.0, 0.0, 0.0, 0.0);
96 //prevents errors caused by creating shortcut files if you had an error in the sanity check.
97 if (m->control_pressed) { m->mothurRemove(phyloTreeName); m->mothurRemove(probFileName); m->mothurRemove(probFileName2); }
99 genusNodes = phyloTree->getGenusNodes();
100 genusTotals = phyloTree->getGenusTotals();
102 m->mothurOut("Calculating template taxonomy tree... "); cout.flush();
104 phyloTree->printTreeNodes(phyloTreeName);
106 m->mothurOut("DONE."); m->mothurOutEndLine();
108 m->mothurOut("Calculating template probabilities... "); cout.flush();
110 numKmers = database->getMaxKmer() + 1;
112 //initialze probabilities
113 wordGenusProb.resize(numKmers);
114 WordPairDiffArr.resize(numKmers);
116 for (int j = 0; j < wordGenusProb.size(); j++) { wordGenusProb[j].resize(genusNodes.size()); }
122 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
129 m->openOutputFile(probFileName, out);
131 //output mothur version
132 out << "#" << m->getVersion() << endl;
134 out << numKmers << endl;
136 m->openOutputFile(probFileName2, out2);
138 //output mothur version
139 out2 << "#" << m->getVersion() << endl;
148 for (int i = 0; i < numKmers; i++) {
149 if (m->control_pressed) { break; }
152 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
157 if (shortcuts) { out << i << '\t'; }
163 vector<int> seqsWithWordi = database->getSequencesWithKmer(i);
166 for (int k = 0; k < genusNodes.size(); k++) { count[genusNodes[k]] = 0; }
168 //for each sequence with that word
169 for (int j = 0; j < seqsWithWordi.size(); j++) {
170 int temp = phyloTree->getIndex(names[seqsWithWordi[j]]);
171 count[temp]++; //increment count of seq in this genus who have this word
174 //probabilityInTemplate = (# of seqs with that word in template + 0.50) / (total number of seqs in template + 1);
175 float probabilityInTemplate = (seqsWithWordi.size() + 0.50) / (float) (names.size() + 1);
176 diffPair tempProb(log(probabilityInTemplate), 0.0);
177 WordPairDiffArr[i] = tempProb;
180 for (int k = 0; k < genusNodes.size(); k++) {
181 //probabilityInThisTaxonomy = (# of seqs with that word in this taxonomy + probabilityInTemplate) / (total number of seqs in this taxonomy + 1);
184 wordGenusProb[i][k] = log((count[genusNodes[k]] + probabilityInTemplate) / (float) (genusTotals[k] + 1));
186 if (count[genusNodes[k]] != 0) {
189 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
194 if (shortcuts) { out << k << '\t' << wordGenusProb[i][k] << '\t' ; }
205 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
212 out2 << probabilityInTemplate << '\t' << numNotZero << '\t' << log(probabilityInTemplate) << endl;
221 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
234 //read in new phylotree with less info. - its faster
235 ifstream phyloTreeTest(phyloTreeName.c_str());
238 phyloTree = new PhyloTree(phyloTreeTest, phyloTreeName);
241 if (rdb->save) { rdb->wordGenusProb = wordGenusProb; rdb->WordPairDiffArr = WordPairDiffArr; }
245 generateWordPairDiffArr();
248 if (rdb->save) { rdb->wordGenusProb = wordGenusProb; rdb->WordPairDiffArr = WordPairDiffArr; }
250 m->mothurOut("DONE."); m->mothurOutEndLine();
251 m->mothurOut("It took " + toString(time(NULL) - start) + " seconds get probabilities. "); m->mothurOutEndLine();
253 catch(exception& e) {
254 m->errorOut(e, "Bayesian", "Bayesian");
258 /**************************************************************************************************/
259 Bayesian::~Bayesian() {
263 if (database != NULL) { delete database; }
265 catch(exception& e) {
266 m->errorOut(e, "Bayesian", "~Bayesian");
271 /**************************************************************************************************/
272 string Bayesian::getTaxonomy(Sequence* seq) {
278 //get words contained in query
279 //getKmerString returns a string where the index in the string is hte kmer number
280 //and the character at that index can be converted to be the number of times that kmer was seen
281 string queryKmerString = kmer.getKmerString(seq->getUnaligned());
283 vector<int> queryKmers;
284 for (int i = 0; i < queryKmerString.length()-1; i++) { // the -1 is to ignore any kmer with an N in it
285 if (queryKmerString[i] != '!') { //this kmer is in the query
286 queryKmers.push_back(i);
290 //if user wants to test reverse compliment and its reversed use that instead
292 if (isReversed(queryKmers)) {
294 seq->reverseComplement();
295 queryKmerString = kmer.getKmerString(seq->getUnaligned());
297 for (int i = 0; i < queryKmerString.length()-1; i++) { // the -1 is to ignore any kmer with an N in it
298 if (queryKmerString[i] != '!') { //this kmer is in the query
299 queryKmers.push_back(i);
305 if (queryKmers.size() == 0) { m->mothurOut(seq->getName() + "is bad."); m->mothurOutEndLine(); simpleTax = "unknown;"; return "unknown;"; }
308 int index = getMostProbableTaxonomy(queryKmers);
310 if (m->control_pressed) { return tax; }
312 //bootstrap - to set confidenceScore
313 int numToSelect = queryKmers.size() / 8;
315 tax = bootstrapResults(queryKmers, index, numToSelect);
319 catch(exception& e) {
320 m->errorOut(e, "Bayesian", "getTaxonomy");
324 /**************************************************************************************************/
325 string Bayesian::bootstrapResults(vector<int> kmers, int tax, int numToSelect) {
328 map<int, int> confidenceScores;
330 //initialize confidences to 0
332 TaxNode seq = phyloTree->get(tax);
333 confidenceScores[tax] = 0;
335 while (seq.level != 0) { //while you are not at the root
336 seqIndex = seq.parent;
337 confidenceScores[seqIndex] = 0;
338 seq = phyloTree->get(seq.parent);
341 map<int, int>::iterator itBoot;
342 map<int, int>::iterator itBoot2;
343 map<int, int>::iterator itConvert;
345 for (int i = 0; i < iters; i++) {
346 if (m->control_pressed) { return "control"; }
349 for (int j = 0; j < numToSelect; j++) {
350 int index = int(rand() % kmers.size());
353 temp.push_back(kmers[index]);
357 int newTax = getMostProbableTaxonomy(temp);
359 TaxNode taxonomyTemp = phyloTree->get(newTax);
361 //add to confidence results
362 while (taxonomyTemp.level != 0) { //while you are not at the root
363 itBoot2 = confidenceScores.find(newTax); //is this a classification we already have a count on
365 if (itBoot2 != confidenceScores.end()) { //this is a classification we need a confidence for
369 newTax = taxonomyTemp.parent;
370 taxonomyTemp = phyloTree->get(newTax);
375 string confidenceTax = "";
378 int seqTaxIndex = tax;
379 TaxNode seqTax = phyloTree->get(tax);
381 while (seqTax.level != 0) { //while you are not at the root
383 itBoot2 = confidenceScores.find(seqTaxIndex); //is this a classification we already have a count on
386 if (itBoot2 != confidenceScores.end()) { //already in confidence scores
387 confidence = itBoot2->second;
390 if (((confidence/(float)iters) * 100) >= confidenceThreshold) {
391 confidenceTax = seqTax.name + "(" + toString(((confidence/(float)iters) * 100)) + ");" + confidenceTax;
392 simpleTax = seqTax.name + ";" + simpleTax;
395 seqTaxIndex = seqTax.parent;
396 seqTax = phyloTree->get(seqTax.parent);
399 if (confidenceTax == "") { confidenceTax = "unknown;"; simpleTax = "unknown;"; }
401 return confidenceTax;
404 catch(exception& e) {
405 m->errorOut(e, "Bayesian", "bootstrapResults");
409 /**************************************************************************************************/
410 int Bayesian::getMostProbableTaxonomy(vector<int> queryKmer) {
412 int indexofGenus = 0;
414 double maxProbability = -1000000.0;
415 //find taxonomy with highest probability that this sequence is from it
418 // cout << genusNodes.size() << endl;
421 for (int k = 0; k < genusNodes.size(); k++) {
422 //for each taxonomy calc its probability
424 double prob = 0.0000;
425 for (int i = 0; i < queryKmer.size(); i++) {
426 prob += wordGenusProb[queryKmer[i]][k];
429 // cout << phyloTree->get(genusNodes[k]).name << '\t' << prob << endl;
431 //is this the taxonomy with the greatest probability?
432 if (prob > maxProbability) {
433 indexofGenus = genusNodes[k];
434 maxProbability = prob;
441 catch(exception& e) {
442 m->errorOut(e, "Bayesian", "getMostProbableTaxonomy");
446 //********************************************************************************************************************
447 //if it is more probable that the reverse compliment kmers are in the template, then we assume the sequence is reversed.
448 bool Bayesian::isReversed(vector<int>& queryKmers){
450 bool reversed = false;
452 float reverseProb = 0;
454 for (int i = 0; i < queryKmers.size(); i++){
455 int kmer = queryKmers[i];
457 prob += WordPairDiffArr[kmer].prob;
458 reverseProb += WordPairDiffArr[kmer].reverseProb;
462 if (reverseProb > prob){ reversed = true; }
466 catch(exception& e) {
467 m->errorOut(e, "Bayesian", "isReversed");
471 //********************************************************************************************************************
472 int Bayesian::generateWordPairDiffArr(){
475 for (int i = 0; i < WordPairDiffArr.size(); i++) {
476 int reversedWord = kmer.getReverseKmerNumber(i);
477 WordPairDiffArr[i].reverseProb = WordPairDiffArr[reversedWord].prob;
481 }catch(exception& e) {
482 m->errorOut(e, "Bayesian", "generateWordPairDiffArr");
486 /*************************************************************************************************
487 map<string, int> Bayesian::parseTaxMap(string newTax) {
490 map<string, int> parsed;
492 newTax = newTax.substr(0, newTax.length()-1); //get rid of last ';'
496 while (newTax.find_first_of(';') != -1) {
497 individual = newTax.substr(0,newTax.find_first_of(';'));
498 newTax = newTax.substr(newTax.find_first_of(';')+1, newTax.length());
499 parsed[individual] = 1;
508 catch(exception& e) {
509 m->errorOut(e, "Bayesian", "parseTax");
513 **************************************************************************************************/
514 void Bayesian::readProbFile(ifstream& in, ifstream& inNum, string inName, string inNumName) {
519 int pid, num, num2, processors;
520 vector<unsigned long long> positions;
521 vector<unsigned long long> positions2;
526 MPI_Comm_rank(MPI_COMM_WORLD, &pid); //find out who we are
527 MPI_Comm_size(MPI_COMM_WORLD, &processors);
530 char inFileName[1024];
531 strcpy(inFileName, inNumName.c_str());
533 char inFileName2[1024];
534 strcpy(inFileName2, inName.c_str());
536 MPI_File_open(MPI_COMM_WORLD, inFileName, MPI_MODE_RDONLY, MPI_INFO_NULL, &inMPI); //comm, filename, mode, info, filepointer
537 MPI_File_open(MPI_COMM_WORLD, inFileName2, MPI_MODE_RDONLY, MPI_INFO_NULL, &inMPI2); //comm, filename, mode, info, filepointer
540 positions = m->setFilePosEachLine(inNumName, num);
541 positions2 = m->setFilePosEachLine(inName, num2);
543 for(int i = 1; i < processors; i++) {
544 MPI_Send(&num, 1, MPI_INT, i, tag, MPI_COMM_WORLD);
545 MPI_Send(&positions[0], (num+1), MPI_LONG, i, tag, MPI_COMM_WORLD);
547 MPI_Send(&num2, 1, MPI_INT, i, tag, MPI_COMM_WORLD);
548 MPI_Send(&positions2[0], (num2+1), MPI_LONG, i, tag, MPI_COMM_WORLD);
552 MPI_Recv(&num, 1, MPI_INT, 0, tag, MPI_COMM_WORLD, &status);
553 positions.resize(num+1);
554 MPI_Recv(&positions[0], (num+1), MPI_LONG, 0, tag, MPI_COMM_WORLD, &status);
556 MPI_Recv(&num2, 1, MPI_INT, 0, tag, MPI_COMM_WORLD, &status);
557 positions2.resize(num2+1);
558 MPI_Recv(&positions2[0], (num2+1), MPI_LONG, 0, tag, MPI_COMM_WORLD, &status);
562 int length = positions2[1] - positions2[0];
563 char* buf5 = new char[length];
565 MPI_File_read_at(inMPI2, positions2[0], buf5, length, MPI_CHAR, &status);
569 length = positions2[2] - positions2[1];
570 char* buf = new char[length];
572 MPI_File_read_at(inMPI2, positions2[1], buf, length, MPI_CHAR, &status);
574 string tempBuf = buf;
575 if (tempBuf.length() > length) { tempBuf = tempBuf.substr(0, length); }
578 istringstream iss (tempBuf,istringstream::in);
581 //initialze probabilities
582 wordGenusProb.resize(numKmers);
584 for (int j = 0; j < wordGenusProb.size(); j++) { wordGenusProb[j].resize(genusNodes.size()); }
587 vector<int> numbers; numbers.resize(numKmers);
589 vector<float> zeroCountProb; zeroCountProb.resize(numKmers);
590 WordPairDiffArr.resize(numKmers);
593 length = positions[1] - positions[0];
594 char* buf6 = new char[length];
596 MPI_File_read_at(inMPI2, positions[0], buf6, length, MPI_CHAR, &status);
600 for(int i=1;i<num;i++){
602 length = positions[i+1] - positions[i];
603 char* buf4 = new char[length];
605 MPI_File_read_at(inMPI, positions[i], buf4, length, MPI_CHAR, &status);
608 if (tempBuf.length() > length) { tempBuf = tempBuf.substr(0, length); }
611 istringstream iss (tempBuf,istringstream::in);
613 iss >> zeroCountProb[i] >> numbers[i] >> probTemp;
614 WordPairDiffArr[i].prob = probTemp;
618 MPI_File_close(&inMPI);
620 for(int i=2;i<num2;i++){
622 length = positions2[i+1] - positions2[i];
623 char* buf4 = new char[length];
625 MPI_File_read_at(inMPI2, positions2[i], buf4, length, MPI_CHAR, &status);
628 if (tempBuf.length() > length) { tempBuf = tempBuf.substr(0, length); }
631 istringstream iss (tempBuf,istringstream::in);
635 //set them all to zero value
636 for (int i = 0; i < genusNodes.size(); i++) {
637 wordGenusProb[kmer][i] = log(zeroCountProb[kmer] / (float) (genusTotals[i]+1));
640 //get probs for nonzero values
641 for (int i = 0; i < numbers[kmer]; i++) {
643 wordGenusProb[kmer][name] = prob;
647 MPI_File_close(&inMPI2);
648 MPI_Barrier(MPI_COMM_WORLD); //make everyone wait - just in case
651 string line = m->getline(in); m->gobble(in);
653 in >> numKmers; m->gobble(in);
654 //cout << threadID << '\t' << line << '\t' << numKmers << &in << '\t' << &inNum << '\t' << genusNodes.size() << endl;
655 //initialze probabilities
656 wordGenusProb.resize(numKmers);
658 for (int j = 0; j < wordGenusProb.size(); j++) { wordGenusProb[j].resize(genusNodes.size()); }
660 int kmer, name, count; count = 0;
661 vector<int> num; num.resize(numKmers);
663 vector<float> zeroCountProb; zeroCountProb.resize(numKmers);
664 WordPairDiffArr.resize(numKmers);
667 string line2 = m->getline(inNum); m->gobble(inNum);
669 //cout << threadID << '\t' << line2 << '\t' << this << endl;
671 inNum >> zeroCountProb[count] >> num[count] >> probTemp;
672 WordPairDiffArr[count].prob = probTemp;
675 //cout << threadID << '\t' << count << endl;
678 //cout << threadID << '\t' << "here1 " << &wordGenusProb << '\t' << &num << endl; //
679 //cout << threadID << '\t' << &genusTotals << '\t' << endl;
680 //cout << threadID << '\t' << genusNodes.size() << endl;
683 //cout << threadID << '\t' << kmer << endl;
684 //set them all to zero value
685 for (int i = 0; i < genusNodes.size(); i++) {
686 wordGenusProb[kmer][i] = log(zeroCountProb[kmer] / (float) (genusTotals[i]+1));
688 //cout << threadID << '\t' << num[kmer] << "here" << endl;
689 //get probs for nonzero values
690 for (int i = 0; i < num[kmer]; i++) {
692 wordGenusProb[kmer][name] = prob;
698 //cout << threadID << '\t' << "here" << endl;
701 catch(exception& e) {
702 m->errorOut(e, "Bayesian", "readProbFile");
706 /**************************************************************************************************/
707 bool Bayesian::checkReleaseDate(ifstream& file1, ifstream& file2, ifstream& file3, ifstream& file4) {
712 vector<string> lines;
713 lines.push_back(m->getline(file1));
714 lines.push_back(m->getline(file2));
715 lines.push_back(m->getline(file3));
716 lines.push_back(m->getline(file4));
718 //before we added this check
719 if ((lines[0][0] != '#') || (lines[1][0] != '#') || (lines[2][0] != '#') || (lines[3][0] != '#')) { good = false; }
722 for (int i = 0; i < lines.size(); i++) { lines[i] = lines[i].substr(1); }
724 //get mothurs current version
725 string version = m->getVersion();
727 vector<string> versionVector;
728 m->splitAtChar(version, versionVector, '.');
730 //check each files version
731 for (int i = 0; i < lines.size(); i++) {
732 vector<string> linesVector;
733 m->splitAtChar(lines[i], linesVector, '.');
735 if (versionVector.size() != linesVector.size()) { good = false; break; }
737 for (int j = 0; j < versionVector.size(); j++) {
739 convert(versionVector[j], num1);
740 convert(linesVector[j], num2);
742 //if mothurs version is newer than this files version, then we want to remake it
743 if (num1 > num2) { good = false; break; }
747 if (!good) { break; }
751 if (!good) { file1.close(); file2.close(); file3.close(); file4.close(); }
752 else { file1.seekg(0); file2.seekg(0); file3.seekg(0); file4.seekg(0); }
756 catch(exception& e) {
757 m->errorOut(e, "Bayesian", "checkReleaseDate");
761 /**************************************************************************************************/