// Compute the initial BWTs for the input file split into blocks of records using the BCR algorithm
MergeVector computeInitialBCR(const BWTDiskParameters& parameters)
{
SeqReader* pReader = new SeqReader(parameters.inFile);
SeqRecord record;
int groupID = 0;
size_t numReadTotal = 0;
MergeVector mergeVector;
MergeItem mergeItem;
mergeItem.start_index = 0;
// Phase 1: Compute the initial BWTs
DNAEncodedStringVector readSequences;
bool done = false;
while(!done)
{
done = !pReader->get(record);
if(!done)
{
// the read is valid
SeqItem item = record.toSeqItem();
if(parameters.bBuildReverse)
item.seq.reverse();
readSequences.push_back(item.seq.toString());
++numReadTotal;
}
if(readSequences.size() >= parameters.numReadsPerBatch || (done && readSequences.size() > 0))
{
std::string bwt_temp_filename = makeTempName(parameters.outPrefix, groupID, parameters.bwtExtension);
std::string sai_temp_filename = makeTempName(parameters.outPrefix, groupID, parameters.saiExtension);
BWTCA::runBauerCoxRosone(&readSequences, bwt_temp_filename, sai_temp_filename);
// Push the merge info
mergeItem.end_index = numReadTotal - 1; // inclusive
mergeItem.reads_filename = parameters.inFile;
mergeItem.bwt_filename = bwt_temp_filename;
mergeItem.sai_filename = sai_temp_filename;
mergeVector.push_back(mergeItem);
// Start the new group
mergeItem.start_index = numReadTotal;
++groupID;
readSequences.clear();
}
}
delete pReader;
return mergeVector;
}
//
// Main
//
int overlapLongMain(int argc, char** argv)
{
parseOverlapLongOptions(argc, argv);
// Open output file
std::ostream* pASQGWriter = createWriter(opt::outFile);
// Build and write the ASQG header
ASQG::HeaderRecord headerRecord;
headerRecord.setOverlapTag(opt::minOverlap);
headerRecord.setErrorRateTag(opt::errorRate);
headerRecord.setInputFileTag(opt::readsFile);
headerRecord.setTransitiveTag(true);
headerRecord.write(*pASQGWriter);
// Determine which index files to use. If a target file was provided,
// use the index of the target reads
std::string indexPrefix;
if(!opt::targetFile.empty())
indexPrefix = stripFilename(opt::targetFile);
else
indexPrefix = stripFilename(opt::readsFile);
BWT* pBWT = new BWT(indexPrefix + BWT_EXT, opt::sampleRate);
SampledSuffixArray* pSSA = new SampledSuffixArray(indexPrefix + SAI_EXT, SSA_FT_SAI);
Timer* pTimer = new Timer(PROGRAM_IDENT);
pBWT->printInfo();
// Read the sequence file and write vertex records for each
// Also store the read names in a vector of strings
ReadTable reads;
SeqReader* pReader = new SeqReader(opt::readsFile, SRF_NO_VALIDATION);
SeqRecord record;
while(pReader->get(record))
{
reads.addRead(record.toSeqItem());
ASQG::VertexRecord vr(record.id, record.seq.toString());
vr.write(*pASQGWriter);
if(reads.getCount() % 100000 == 0)
printf("Read %zu sequences\n", reads.getCount());
}
delete pReader;
pReader = NULL;
BWTIndexSet index;
index.pBWT = pBWT;
index.pSSA = pSSA;
index.pReadTable = &reads;
// Make a prefix for the temporary hits files
size_t n_reads = reads.getCount();
omp_set_num_threads(opt::numThreads);
#pragma omp parallel for
for(size_t read_idx = 0; read_idx < n_reads; ++read_idx)
{
const SeqItem& curr_read = reads.getRead(read_idx);
printf("read %s %zubp\n", curr_read.id.c_str(), curr_read.seq.length());
SequenceOverlapPairVector sopv =
KmerOverlaps::retrieveMatches(curr_read.seq.toString(),
opt::seedLength,
opt::minOverlap,
1 - opt::errorRate,
100,
index);
printf("Found %zu matches\n", sopv.size());
for(size_t i = 0; i < sopv.size(); ++i)
{
std::string match_id = reads.getRead(sopv[i].match_idx).id;
// We only want to output each edge once so skip this overlap
// if the matched read has a lexicographically lower ID
if(curr_read.id > match_id)
continue;
std::string ao = ascii_overlap(sopv[i].sequence[0], sopv[i].sequence[1], sopv[i].overlap, 50);
printf("\t%s\t[%d %d] ID=%s OL=%d PI:%.2lf C=%s\n", ao.c_str(),
sopv[i].overlap.match[0].start,
sopv[i].overlap.match[0].end,
match_id.c_str(),
sopv[i].overlap.getOverlapLength(),
sopv[i].overlap.getPercentIdentity(),
sopv[i].overlap.cigar.c_str());
// Convert to ASQG
SeqCoord sc1(sopv[i].overlap.match[0].start, sopv[i].overlap.match[0].end, sopv[i].overlap.length[0]);
SeqCoord sc2(sopv[i].overlap.match[1].start, sopv[i].overlap.match[1].end, sopv[i].overlap.length[1]);
// KmerOverlaps returns the coordinates of the overlap after flipping the reads
// to ensure the strand matches. The ASQG file wants the coordinate of the original
// sequencing strand. Flip here if necessary
if(sopv[i].is_reversed)
sc2.flip();
// Convert the SequenceOverlap the ASQG's overlap format
Overlap ovr(curr_read.id, sc1, match_id, sc2, sopv[i].is_reversed, -1);
ASQG::EdgeRecord er(ovr);
er.setCigarTag(sopv[i].overlap.cigar);
er.setPercentIdentityTag(sopv[i].overlap.getPercentIdentity());
#pragma omp critical
{
er.write(*pASQGWriter);
}
}
}
// Cleanup
delete pReader;
delete pBWT;
delete pSSA;
delete pASQGWriter;
delete pTimer;
if(opt::numThreads > 1)
pthread_exit(NULL);
return 0;
}
// The algorithm is as follows. We create M BWTs for subsets of
// the input reads. These are created independently and written
// to disk. They are then merged either sequentially or pairwise
// to create the final BWT
void buildBWTDisk(const BWTDiskParameters& parameters)
{
// Build the initial bwts for subsets of the data
MergeVector mergeVector;
if(parameters.bUseBCR)
mergeVector = computeInitialBCR(parameters);
else
mergeVector = computeInitialSAIS(parameters);
// Phase 2: Pairwise merge the BWTs
int groupID = mergeVector.size(); // Initial the name of the next intermediate bwt
int round = 1;
MergeVector nextMergeRound;
while(mergeVector.size() > 1)
{
std::cout << "Starting round " << round << "\n";
SeqReader* pReader = new SeqReader(parameters.inFile);
SeqRecord record;
for(size_t i = 0; i < mergeVector.size(); i+=2)
{
if(i + 1 != mergeVector.size())
{
std::string bwt_merged_name = makeTempName(parameters.outPrefix, groupID, parameters.bwtExtension);
std::string sai_merged_name = makeTempName(parameters.outPrefix, groupID, parameters.saiExtension);
MergeItem item1 = mergeVector[i];
MergeItem item2 = mergeVector[i+1];
// Perform the actual merge
int64_t curr_idx = merge(pReader, item1, item2,
bwt_merged_name, sai_merged_name,
parameters.bBuildReverse, parameters.numThreads, parameters.storageLevel);
// pReader now points to the end of item1's block of
// reads. Skip item2's reads
assert(curr_idx == item2.start_index);
while(curr_idx <= item2.end_index)
{
bool eof = !pReader->get(record);
assert(!eof);
(void)eof;
++curr_idx;
}
// Create the merged mergeItem to use in the next round
MergeItem merged;
merged.start_index = item1.start_index;
merged.end_index = item2.end_index;
merged.bwt_filename = bwt_merged_name;
merged.sai_filename = sai_merged_name;
nextMergeRound.push_back(merged);
// Done with the temp files, remove them
unlink(item1.bwt_filename.c_str());
unlink(item2.bwt_filename.c_str());
unlink(item1.sai_filename.c_str());
unlink(item2.sai_filename.c_str());
++groupID;
}
else
{
// Singleton, pass through to the next round
nextMergeRound.push_back(mergeVector[i]);
}
}
delete pReader;
mergeVector.clear();
mergeVector.swap(nextMergeRound);
++round;
}
assert(mergeVector.size() == 1);
// Done, rename the files to their final name
std::stringstream bwt_ss;
bwt_ss << parameters.outPrefix << parameters.bwtExtension << (USE_GZ ? ".gz" : "");
std::string bwt_final_filename = bwt_ss.str();
rename(mergeVector.front().bwt_filename.c_str(), bwt_final_filename.c_str());
std::stringstream sai_ss;
sai_ss << parameters.outPrefix << parameters.saiExtension << (USE_GZ ? ".gz" : "");
std::string sai_final_filename = sai_ss.str();
rename(mergeVector.front().sai_filename.c_str(), sai_final_filename.c_str());
}
// Compute the initial BWTs for the input file split into blocks of records using the SAIS algorithm
MergeVector computeInitialSAIS(const BWTDiskParameters& parameters)
{
SeqReader* pReader = new SeqReader(parameters.inFile);
SeqRecord record;
int groupID = 0;
size_t numReadTotal = 0;
MergeVector mergeVector;
MergeItem mergeItem;
mergeItem.start_index = 0;
// Phase 1: Compute the initial BWTs
ReadTable* pCurrRT = new ReadTable;
bool done = false;
while(!done)
{
done = !pReader->get(record);
if(!done)
{
// the read is valid
SeqItem item = record.toSeqItem();
if(parameters.bBuildReverse)
item.seq.reverse();
pCurrRT->addRead(item);
++numReadTotal;
}
if(pCurrRT->getCount() >= parameters.numReadsPerBatch || (done && pCurrRT->getCount() > 0))
{
// Compute the SA and BWT for this group
SuffixArray* pSA = new SuffixArray(pCurrRT, 1);
// Write the BWT to disk
std::string bwt_temp_filename = makeTempName(parameters.outPrefix, groupID, parameters.bwtExtension);
pSA->writeBWT(bwt_temp_filename, pCurrRT);
std::string sai_temp_filename = makeTempName(parameters.outPrefix, groupID, parameters.saiExtension);
pSA->writeIndex(sai_temp_filename);
// Push the merge info
mergeItem.end_index = numReadTotal - 1; // inclusive
mergeItem.reads_filename = parameters.inFile;
mergeItem.bwt_filename = bwt_temp_filename;
mergeItem.sai_filename = sai_temp_filename;
mergeVector.push_back(mergeItem);
// Cleanup
delete pSA;
// Start the new group
mergeItem.start_index = numReadTotal;
++groupID;
pCurrRT->clear();
}
}
delete pCurrRT;
delete pReader;
return mergeVector;
}
// The algorithm is as follows. We create M BWTs for subsets of
// the input reads. These are created independently and written
// to disk. They are then merged either sequentially or pairwise
// to create the final BWT
void buildBWTDisk(const std::string& in_filename, const std::string& out_prefix,
const std::string& bwt_extension, const std::string& sai_extension,
bool doReverse, int numThreads, int numReadsPerBatch, int storageLevel)
{
size_t MAX_READS_PER_GROUP = numReadsPerBatch;
SeqReader* pReader = new SeqReader(in_filename);
SeqRecord record;
int groupID = 0;
size_t numReadTotal = 0;
MergeVector mergeVector;
MergeItem mergeItem;
mergeItem.start_index = 0;
// Phase 1: Compute the initial BWTs
ReadTable* pCurrRT = new ReadTable;
bool done = false;
while(!done)
{
done = !pReader->get(record);
if(!done)
{
// the read is valid
SeqItem item = record.toSeqItem();
if(doReverse)
item.seq.reverse();
pCurrRT->addRead(item);
++numReadTotal;
}
if(pCurrRT->getCount() >= MAX_READS_PER_GROUP || (done && pCurrRT->getCount() > 0))
{
// Compute the SA and BWT for this group
SuffixArray* pSA = new SuffixArray(pCurrRT, numThreads);
// Write the BWT to disk
std::string bwt_temp_filename = makeTempName(out_prefix, groupID, bwt_extension);
pSA->writeBWT(bwt_temp_filename, pCurrRT);
std::string sai_temp_filename = makeTempName(out_prefix, groupID, sai_extension);
pSA->writeIndex(sai_temp_filename);
// Push the merge info
mergeItem.end_index = numReadTotal - 1; // inclusive
mergeItem.reads_filename = in_filename;
mergeItem.bwt_filename = bwt_temp_filename;
mergeItem.sai_filename = sai_temp_filename;
mergeVector.push_back(mergeItem);
// Cleanup
delete pSA;
// Start the new group
mergeItem.start_index = numReadTotal;
++groupID;
pCurrRT->clear();
}
}
delete pCurrRT;
delete pReader;
// Phase 2: Pairwise merge the BWTs
int round = 1;
MergeVector nextMergeRound;
while(mergeVector.size() > 1)
{
std::cout << "Starting round " << round << "\n";
pReader = new SeqReader(in_filename);
for(size_t i = 0; i < mergeVector.size(); i+=2)
{
if(i + 1 != mergeVector.size())
{
std::string bwt_merged_name = makeTempName(out_prefix, groupID, bwt_extension);
std::string sai_merged_name = makeTempName(out_prefix, groupID, sai_extension);
MergeItem item1 = mergeVector[i];
MergeItem item2 = mergeVector[i+1];
// Perform the actual merge
int64_t curr_idx = merge(pReader, item1, item2,
bwt_merged_name, sai_merged_name,
doReverse, numThreads, storageLevel);
// pReader now points to the end of item1's block of
// reads. Skip item2's reads
assert(curr_idx == item2.start_index);
while(curr_idx <= item2.end_index)
{
bool eof = !pReader->get(record);
assert(!eof);
(void)eof;
++curr_idx;
}
// Create the merged mergeItem to use in the next round
MergeItem merged;
merged.start_index = item1.start_index;
merged.end_index = item2.end_index;
merged.bwt_filename = bwt_merged_name;
merged.sai_filename = sai_merged_name;
nextMergeRound.push_back(merged);
// Done with the temp files, remove them
unlink(item1.bwt_filename.c_str());
unlink(item2.bwt_filename.c_str());
unlink(item1.sai_filename.c_str());
unlink(item2.sai_filename.c_str());
++groupID;
}
else
{
// Singleton, pass through to the next round
nextMergeRound.push_back(mergeVector[i]);
}
}
delete pReader;
mergeVector.clear();
mergeVector.swap(nextMergeRound);
++round;
}
assert(mergeVector.size() == 1);
// Done, rename the files to their final name
std::stringstream bwt_ss;
bwt_ss << out_prefix << bwt_extension << (USE_GZ ? ".gz" : "");
std::string bwt_final_filename = bwt_ss.str();
rename(mergeVector.front().bwt_filename.c_str(), bwt_final_filename.c_str());
std::stringstream sai_ss;
sai_ss << out_prefix << sai_extension << (USE_GZ ? ".gz" : "");
std::string sai_final_filename = sai_ss.str();
rename(mergeVector.front().sai_filename.c_str(), sai_final_filename.c_str());
}
