void IRKE::compute_sequence_assemblies(KmerCounter& kcounter, float min_connectivity,
unsigned int MIN_ASSEMBLY_LENGTH, unsigned int MIN_ASSEMBLY_COVERAGE,
bool WRITE_COVERAGE, string COVERAGE_OUTPUT_FILENAME) {
if (! got_sorted_kmers_flag) {
stringstream error;
error << stacktrace() << " Error, must populate_sorted_kmers_list() before computing sequence assemblies" << endl;
throw(error.str());
}
unsigned int kmer_length = kcounter.get_kmer_length();
ofstream coverage_writer;
if (WRITE_COVERAGE) {
coverage_writer.open(COVERAGE_OUTPUT_FILENAME.c_str());
}
vector<Kmer_counter_map_iterator>& kmers = sorted_kmers; //kcounter.get_kmers_sort_descending_counts();
unsigned long init_size = kcounter.size();
// string s = "before.kmers";
// kcounter.dump_kmers_to_file(s);
for (unsigned int i = 0; i < kmers.size(); i++) {
// cerr << "round: " << i << endl;
unsigned long kmer_counter_size = kcounter.size();
if (kmer_counter_size > init_size) {
// string s = "after.kmers";
// kcounter.dump_kmers_to_file(s);
stringstream error;
error << stacktrace() << "Error, Kcounter size has grown from " << init_size
<< " to " << kmer_counter_size << endl;
throw (error.str());
}
kmer_int_type_t kmer = kmers[i]->first;
unsigned int kmer_count = kmers[i]->second;
if (kmer_count == 0) {
continue;
}
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "SEED kmer: " << kcounter.get_kmer_string(kmer) << ", count: " << kmer_count << endl;
}
if (kmer == revcomp_val(kmer, kmer_length)) {
// palindromic kmer, avoid palindromes as seeds
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "SEED kmer: " << kcounter.get_kmer_string(kmer) << " is palidnromic. Skipping. " << endl;
}
continue;
}
if (kmer_count < MIN_SEED_COVERAGE) {
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "-seed has insufficient coverage, skipping" << endl;
}
continue;
}
float entropy = compute_entropy(kmer, kmer_length);
if (entropy < MIN_SEED_ENTROPY) {
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "-skipping seed due to low entropy: " << entropy << endl;
}
continue;
}
/* Extend to the right */
Kmer_visitor visitor(kmer_length, DOUBLE_STRANDED_MODE);
Path_n_count_pair selected_path_n_pair_forward = inchworm(kcounter, 'F', kmer, visitor, min_connectivity);
visitor.clear();
// add selected path to visitor
vector<kmer_int_type_t>& forward_path = selected_path_n_pair_forward.first;
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "Forward path contains: " << forward_path.size() << " kmers. " << endl;
}
for (unsigned int i = 0; i < forward_path.size(); i++) {
kmer_int_type_t kmer = forward_path[i];
visitor.add(kmer);
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "\tForward path kmer: " << kcounter.get_kmer_string(kmer) << endl;
}
}
/* Extend to the left */
visitor.erase(kmer); // reset the seed
Path_n_count_pair selected_path_n_pair_reverse = inchworm(kcounter, 'R', kmer, visitor, min_connectivity);
if (IRKE_COMMON::MONITOR >= 2) {
vector<kmer_int_type_t>& reverse_path = selected_path_n_pair_reverse.first;
cerr << "Reverse path contains: " << reverse_path.size() << " kmers. " << endl;
for (unsigned int i = 0; i < reverse_path.size(); i++) {
cerr << "\tReverse path kmer: " << kcounter.get_kmer_string(reverse_path[i]) << endl;
}
}
unsigned int total_counts = selected_path_n_pair_forward.second + selected_path_n_pair_reverse.second + kcounter.get_kmer_count(kmer);
vector<kmer_int_type_t>& reverse_path = selected_path_n_pair_reverse.first;
vector<kmer_int_type_t> joined_path = _join_forward_n_reverse_paths(reverse_path, kmer, forward_path);
// report sequence reconstructed from path.
vector<unsigned int> assembly_base_coverage;
string sequence = reconstruct_path_sequence(kcounter, joined_path, assembly_base_coverage);
unsigned int avg_cov = static_cast<unsigned int> ( (float)total_counts/(sequence.length()-kcounter.get_kmer_length() +1) + 0.5);
/*
cout << "Inchworm-reconstructed sequence, length: " << sequence.length()
<< ", avgCov: " << avg_cov
<< " " << sequence << endl;
*/
if (sequence.length() >= MIN_ASSEMBLY_LENGTH && avg_cov >= MIN_ASSEMBLY_COVERAGE) {
INCHWORM_ASSEMBLY_COUNTER++;
stringstream headerstream;
headerstream << ">a" << INCHWORM_ASSEMBLY_COUNTER << ";" << avg_cov
<< " K: " << kmer_length
<< " length: " << sequence.length();
string header = headerstream.str();
sequence = add_fasta_seq_line_breaks(sequence, 60);
cout << header << endl << sequence << endl;
if (WRITE_COVERAGE) {
coverage_writer << header << endl;
for (unsigned int i = 0; i < assembly_base_coverage.size(); i++) {
coverage_writer << assembly_base_coverage[i];
if ( (i+1) % 30 == 0) {
coverage_writer << endl;
}
else {
coverage_writer << " ";
}
}
coverage_writer << endl;
}
}
// remove path
for (unsigned int i = 0; i < joined_path.size(); i++) {
kmer_int_type_t kmer = joined_path[i];
/*
if (DEBUG) {
cout << "\tpruning kmer: " << kmer << endl;
}
*/
kcounter.clear_kmer(kmer);
}
/*
if (DEBUG) {
cout << "done pruning kmers." << endl;
}
*/
}
if (IRKE_COMMON::MONITOR) {
cerr << endl;
}
if (WRITE_COVERAGE) {
coverage_writer.close();
}
// drop sorted kmer list as part of cleanup
clear_sorted_kmers_list();
return; // end of runIRKE
}
void IRKE::compute_sequence_assemblies(KmerCounter &kcounter, float min_connectivity,
unsigned int MIN_ASSEMBLY_LENGTH, unsigned int MIN_ASSEMBLY_COVERAGE,
bool PARALLEL_IWORM, bool WRITE_COVERAGE, string COVERAGE_OUTPUT_FILENAME)
{
if (!got_sorted_kmers_flag) {
stringstream error;
error << stacktrace() << " Error, must populate_sorted_kmers_list() before computing sequence assemblies"
<< endl;
throw (error.str());
}
//vector<Kmer_counter_map_iterator>& kmers = sorted_kmers;
vector<Kmer_Occurence_Pair> &kmers = sorted_kmers; // note, these are not actually sorted if PARALLEL_IWORM mode.
unsigned long init_size = kcounter.size();
cerr << "Total kcounter hash size: " << init_size << " vs. sorted list size: " << kmers.size() << endl;
unsigned int kmer_length = kcounter.get_kmer_length();
ofstream coverage_writer;
if (WRITE_COVERAGE) {
coverage_writer.open(COVERAGE_OUTPUT_FILENAME.c_str());
}
// string s = "before.kmers";
// kcounter.dump_kmers_to_file(s);
/* -----------------------------------------------------------
Two reconstruction modes. (bhaas, Jan-4-2014)
1. Original (not setting PARALLEL_IWORM): kmer list is sorted descendingly by abundance.
Seed kmer is chosen as most abundant, and contig construction extends from this seed.
2. PARALLEL_IWORM: the kmer list is not sorted. A random kmer (just ordered by hash iterator, nothing special, probably not so random)
is used as a seed for draft contig reconstruction.
The most abundant kmer in the draft contig is chosen as a proper seed.
An inchworm contig is then constructed using this new seed, and reported.
So, in this case, the draft contig in the first phase is just used to find a better seed, from which the inchworm contig is then
properly reconstructed.
--------------------------------------------------------------- */
// try building an inchworm contig from each seed kmer
int myTid;
if (PARALLEL_IWORM) {
omp_set_num_threads(IRKE_COMMON::NUM_THREADS);
}
else {
omp_set_num_threads(1); // turn off multithreading for the contig building.
}
//-----------------------------------------------------------------
// Prep writing to separate inchworm output files for each thread
//-----------------------------------------------------------------
vector<iworm_tmp_file> tmpfiles;
int num_threads = omp_get_max_threads();
cerr << "num threads set to: " << num_threads << endl;
for (int i = 0; i < num_threads; i++) {
iworm_tmp_file tmpfile_struct;
tmpfiles.push_back(tmpfile_struct);
iworm_tmp_file &itmp = tmpfiles[i];
stringstream filename_constructor;
filename_constructor << "tmp.iworm.fa.pid_" << getpid() << ".thread_" << i;
itmp.tmp_filename = new char[100];
strcpy(itmp.tmp_filename, filename_constructor.str().c_str());
itmp.tmp_filename[filename_constructor.str().length()] = '\0';
itmp.fh = new ofstream();
itmp.fh->open(itmp.tmp_filename);
cerr << "Done opening file. " << itmp.tmp_filename << endl;
}
//-------------------
// Build contigs.
//-------------------
#pragma omp parallel for private (myTid) schedule (dynamic, 1000)
for (unsigned int i = 0; i < kmers.size(); i++) {
// cerr << "round: " << i << endl;
myTid = omp_get_thread_num();
unsigned long kmer_counter_size = kcounter.size();
if (kmer_counter_size > init_size) {
// string s = "after.kmers";
// kcounter.dump_kmers_to_file(s);
stringstream error;
error << stacktrace() << "Error, Kcounter size has grown from " << init_size
<< " to " << kmer_counter_size << endl;
throw (error.str());
}
//kmer_int_type_t kmer = kmers[i]->first;
//unsigned int kmer_count = kmers[i]->second;
kmer_int_type_t kmer = kmers[i].first;
// unsigned int kmer_count = kmers[i].second; // NO!!! Use for sorting, but likely zeroed out in the hashtable after contig construction
unsigned int kmer_count = kcounter.get_kmer_count(kmer);
if (!is_good_seed_kmer(kmer, kmer_count, kmer_length, min_connectivity)) {
continue;
}
// cout << "SEED kmer: " << kcounter.get_kmer_string(kmer) << ", count: " << kmer_count << endl;
if (IRKE_COMMON::MONITOR >= 2) {
cerr << "SEED kmer: " << kcounter.get_kmer_string(kmer) << ", count: " << kmer_count << endl;
}
if (IRKE_COMMON::MONITOR >= 2) {
#pragma omp critical
cerr << "Seed for thread: " << myTid << " is " << kcounter.get_kmer_string(kmer) << " with count: "
<< kmer_count << endl;
}
unsigned int total_counts;
vector<kmer_int_type_t> joined_path =
build_inchworm_contig_from_seed(kmer, kcounter, min_connectivity, total_counts, PARALLEL_IWORM);
if (PARALLEL_IWORM && TWO_PHASE) {
// get a new seed based on the draft contig
// choose the 'good' kmer with highest abundance
kmer_int_type_t new_seed = extract_best_seed(joined_path, kcounter, min_connectivity);
if (kcounter.get_kmer_count(new_seed) == 0) {
continue; // must have been zapped by another thread
}
joined_path =
build_inchworm_contig_from_seed(new_seed, kcounter, min_connectivity, total_counts, PARALLEL_IWORM);
}
// report sequence reconstructed from path.
vector<unsigned int> assembly_base_coverage;
string sequence = reconstruct_path_sequence(kcounter, joined_path, assembly_base_coverage);
unsigned int avg_cov =
static_cast<unsigned int> ((float) total_counts / (sequence.length() - kcounter.get_kmer_length() + 1)
+ 0.5);
/*
cout << "Inchworm-reconstructed sequence, length: " << sequence.length()
<< ", avgCov: " << avg_cov
<< " " << sequence << endl;
*/
size_t contig_length = sequence.length();
if (contig_length >= MIN_ASSEMBLY_LENGTH && avg_cov >= MIN_ASSEMBLY_COVERAGE) {
*(tmpfiles[myTid].fh) << total_counts << endl
<< avg_cov << endl
<< kmer_count << endl
<< sequence << endl;
}
// remove path
if (IRKE_COMMON::__DEVEL_zero_kmer_on_use) {
// dont forget the seed. The forward/reverse path kmers already cleared.
kcounter.clear_kmer(kmer);
} else {
for (unsigned int i = 0; i < joined_path.size(); i++) {
kmer_int_type_t kmer = joined_path[i];
kcounter.clear_kmer(kmer);
}
}
}
if (IRKE_COMMON::MONITOR) {
cerr << endl;
}
if (WRITE_COVERAGE) {
coverage_writer.close();
}
// drop sorted kmer list as part of cleanup
clear_sorted_kmers_list();
//------------------------------------------------------------------------------
// examine the contigs generated by the individual threads, remove redundancies.
//------------------------------------------------------------------------------
map<unsigned long long, bool> seen_contig_already;
for (unsigned int i = 0; i < tmpfiles.size(); i++) {
tmpfiles[i].fh->close();
ifstream tmpreader(tmpfiles[i].tmp_filename);
while (!tmpreader.eof()) {
unsigned int total_counts;
unsigned int avg_cov;
unsigned int kmer_count;
string sequence;
tmpreader >> total_counts
>> avg_cov
>> kmer_count
>> sequence;
if (tmpreader.eof()) // apparently only happens on the read after the last line is read.
break;
//cerr << "Read sequence: " << sequence << endl;
unsigned int contig_hash = generateHash(sequence);
if (!seen_contig_already[contig_hash]
) {
seen_contig_already[contig_hash] = true;
INCHWORM_ASSEMBLY_COUNTER++;
stringstream headerstream;
headerstream << ">a" << INCHWORM_ASSEMBLY_COUNTER << ";" << avg_cov
<< " total_counts: " << total_counts
//<< " Fpath: " << selected_path_n_pair_forward.second << " Rpath: " << selected_path_n_pair_reverse.second
<< " Seed: " << kmer_count
<< " K: " << kmer_length
<< " length: " << sequence.length();
string header = headerstream.str();
sequence = add_fasta_seq_line_breaks(sequence, 60);
cout << header << endl << sequence << endl;
}
}
// cleanup from earlier dynamic allocation
delete (tmpfiles[i].fh);
if (!IRKE_COMMON::KEEP_TMP_FILES) {
remove(tmpfiles[i].tmp_filename);
}
delete (tmpfiles[i].tmp_filename);
}
/*
if (WRITE_COVERAGE) {
coverage_writer << header << endl;
for (unsigned int i = 0; i < assembly_base_coverage.size(); i++) {
coverage_writer << assembly_base_coverage[i];
if ( (i+1) % 30 == 0) {
coverage_writer << endl;
}
else {
coverage_writer << " ";
}
}
coverage_writer << endl;
}
}
*/
return; // end of runIRKE
}
