void setMateInfo( BamAlignment & rec1, BamAlignment & rec2, SamHeader & header) {
const int NO_ALIGNMENT_REFERENCE_INDEX = -1;
const int NO_ALIGNMENT_START = -1;
// If neither read is unmapped just set their mate info
if (rec1.IsMapped() && rec2.IsMapped()) {
rec1.MateRefID = rec2.MateRefID;
rec1.MatePosition = rec2.Position;
rec1.SetIsReverseStrand(rec2.IsReverseStrand());
rec1.SetIsMapped(true);
rec1.AddTag("MQ", "i", rec2.MapQuality);
rec2.MateRefID = rec1.RefID;
rec2.MatePosition = rec1.Position;
rec2.SetIsReverseStrand( rec1.IsReverseStrand() );
rec2.SetIsMapped(true);
rec2.AddTag("MQ", "i", rec1.MapQuality);
}
// Else if they're both unmapped set that straight
else if (!rec1.IsMapped() && !rec2.IsMapped()) {
rec1.RefID = NO_ALIGNMENT_REFERENCE_INDEX;
rec1.Position = NO_ALIGNMENT_START;
rec1.MateRefID = NO_ALIGNMENT_REFERENCE_INDEX;
rec1.MatePosition = NO_ALIGNMENT_START;
rec1.SetIsReverseStrand(rec2.IsReverseStrand());
rec1.SetIsMapped(false);
rec2.RemoveTag("MQ");
rec1.Length = 0;
rec2.RefID = NO_ALIGNMENT_REFERENCE_INDEX;
rec2.Position = NO_ALIGNMENT_START;
rec2.MateRefID = NO_ALIGNMENT_REFERENCE_INDEX;
rec2.MatePosition = NO_ALIGNMENT_START;
rec2.SetIsReverseStrand(rec1.IsReverseStrand());
rec2.SetIsMapped(false);
rec2.RemoveTag("MQ");
rec2.Length = 0;
}
// And if only one is mapped copy it's coordinate information to the mate
else {
BamAlignment & mapped = rec1.IsMapped() ? rec1 : rec2;
BamAlignment & unmapped = rec1.IsMapped() ? rec2 : rec1;
unmapped.RefID = mapped.RefID;
unmapped.Position = mapped.Position;
mapped.MateRefID = unmapped.RefID;
mapped.MatePosition = unmapped.Position;
mapped.SetIsMateReverseStrand(unmapped.IsReverseStrand());
mapped.SetIsMateMapped(false);
mapped.Length = 0;
unmapped.MateRefID = mapped.RefID;
unmapped.MatePosition = mapped.Position;
unmapped.SetIsMateReverseStrand(mapped.IsReverseStrand());
unmapped.SetIsMateMapped(true);
unmapped.Length = 0;
}
const int insertSize = computeInsertSize(rec1, rec2);
rec1.Length = insertSize;
rec2.Length = -insertSize;
}
void realign_bam(Parameters& params) {
FastaReference reference;
reference.open(params.fasta_reference);
bool suppress_output = false;
int dag_window_size = params.dag_window_size;
// open BAM file
BamReader reader;
if (!reader.Open("stdin")) {
cerr << "could not open stdin for reading" << endl;
exit(1);
}
BamWriter writer;
if (!params.dry_run && !writer.Open("stdout", reader.GetHeaderText(), reader.GetReferenceData())) {
cerr << "could not open stdout for writing" << endl;
exit(1);
}
// store the names of all the reference sequences in the BAM file
map<int, string> referenceIDToName;
vector<RefData> referenceSequences = reader.GetReferenceData();
int i = 0;
for (RefVector::iterator r = referenceSequences.begin(); r != referenceSequences.end(); ++r) {
referenceIDToName[i] = r->RefName;
++i;
}
vcf::VariantCallFile vcffile;
if (!params.vcf_file.empty()) {
if (!vcffile.open(params.vcf_file)) {
cerr << "could not open VCF file " << params.vcf_file << endl;
exit(1);
}
} else {
cerr << "realignment requires VCF file" << endl;
exit(1);
}
vcf::Variant var(vcffile);
BamAlignment alignment;
map<long int, vector<BamAlignment> > alignmentSortQueue;
// get alignment
// assemble DAG in region around alignment
// loop for each alignment in BAM:
// update DAG when current alignment gets close to edge of assembled DAG
// attempt to realign if read has a certain number of mismatches + gaps or softclips, weighted by basequal
// if alignment to DAG has fewer mismatches and gaps than original alignment, use it
// flatten read into reference space (for now just output alleles from VCF un-spanned insertions)
// write read to queue for streaming re-sorting (some positional change will occur)
long int dag_start_position = 0;
string currentSeqname;
string ref;
//vector<Cigar> cigars; // contains the Cigar strings of nodes in the graph
//vector<long int> refpositions; // contains the reference start coords of nodes in the graph
ReferenceMappings ref_map;
gssw_graph* graph = gssw_graph_create(0);
int8_t* nt_table = gssw_create_nt_table();
int8_t* mat = gssw_create_score_matrix(params.match, params.mism);
int total_reads = 0;
int total_realigned = 0;
int total_improved = 0;
bool emptyDAG = false; // if the dag is constructed over empty sequence
// such as when realigning reads mapped to all-N sequence
if (params.debug) {
cerr << "about to start processing alignments" << endl;
}
while (reader.GetNextAlignment(alignment)) {
string& seqname = referenceIDToName[alignment.RefID];
if (params.debug) {
cerr << "--------------------------------------------" << endl
<< "processing alignment " << alignment.Name << " at "
<< seqname << ":" << alignment.Position << endl;
}
/*
if (!alignment.IsMapped() && graph->size == 0) {
if (params.debug) {
cerr << "unable to build DAG using unmapped read "
<< alignment.Name << " @ "
<< seqname << ":" << alignment.Position
<< " no previous mapped read found and DAG currently empty" << endl;
}
alignmentSortQueue[dag_start_position+dag_window_size].push_back(alignment);
continue;
}
*/
++total_reads;
BamAlignment originalAlignment = alignment;
long unsigned int initialAlignmentPosition = alignment.Position;
//if (dag_start_position == 1) {
// dag_start_position = max(1, (int)initialAlignmentPosition - dag_window_size/2);
//}
// should we construct a new DAG? do so when 3/4 of the way through the current one
// center on current position + 1/2 dag window
// TODO check this scheme using some scribbles on paper
// alignment.IsMapped()
if ((seqname != currentSeqname
|| ((alignment.Position + (alignment.QueryBases.size()/2)
> (3*dag_window_size/4) + dag_start_position)))
&& alignment.Position < reference.sequenceLength(seqname)) {
if (seqname != currentSeqname) {
if (params.debug) {
cerr << "switched ref seqs" << endl;
}
dag_start_position = max((long int) 0,
(long int) (alignment.GetEndPosition() - dag_window_size/2));
// recenter DAG
} else if (!ref_map.empty()) {
dag_start_position = dag_start_position + dag_window_size/2;
dag_start_position = max(dag_start_position,
(long int) (alignment.GetEndPosition() - dag_window_size/2));
} else {
dag_start_position = alignment.Position - dag_window_size/2;
}
dag_start_position = max((long int)0, dag_start_position);
// TODO get sequence length and use to bound noted window size (edge case)
//cerr << "getting ref " << seqname << " " << max((long int) 0, dag_start_position) << " " << dag_window_size << endl;
// get variants for new DAG
vector<vcf::Variant> variants;
if (!vcffile.setRegion(seqname,
dag_start_position + 1,
dag_start_position + dag_window_size)) {
// this is not necessarily an error; there should be a better way to check for VCF file validity
/*
cerr << "could not set region on VCF file to " << currentSeqname << ":"
<< dag_start_position << "-" << dag_start_position + ref.size()
<< endl;
*/
//exit(1);
} else {
// check first variant
if (vcffile.getNextVariant(var)) {
while (var.position <= dag_start_position + 1) {
//cerr << "var position == dag_start_position " << endl;
dag_start_position -= 1;
vcffile.setRegion(seqname,
dag_start_position + 1,
dag_start_position + dag_window_size);
if (!vcffile.getNextVariant(var)) { break; }
}
}
vcffile.setRegion(seqname,
dag_start_position + 1,
dag_start_position + dag_window_size);
while (vcffile.getNextVariant(var)) {
if (params.debug) cerr << "getting variant at " << var.sequenceName << ":" << var.position << endl;
//cerr << var.position << " + " << var.ref.length() << " <= " << dag_start_position << " + " << dag_window_size << endl;
//cerr << var.position << " >= " << dag_start_position << endl;
if (var.position + var.ref.length() <= dag_start_position + dag_window_size
&& var.position >= dag_start_position) {
variants.push_back(var);
}
}
}
//cerr << "dag_start_position " << dag_start_position << endl;
ref = reference.getSubSequence(seqname,
max((long int) 0, dag_start_position),
dag_window_size); // 0/1 conversion
// clear graph and metadata
ref_map.clear();
//cigars.clear();
//refpositions.clear();
gssw_graph_destroy(graph);
if (params.debug) { cerr << "constructing DAG" << endl; }
// and build the DAG
graph = gssw_graph_create(0);
constructDAGProgressive(graph,
ref_map,
ref,
seqname,
variants,
dag_start_position,
nt_table,
mat,
params.flat_input_vcf);
if (params.debug) {
cerr << "graph has " << graph->size << " nodes" << endl;
cerr << "DAG generated from input variants over "
<< seqname << ":" << dag_start_position << "-" << dag_start_position + dag_window_size
<< endl;
}
if (params.display_dag) {
gssw_graph_print(graph);
/*
for (Backbone::iterator b = backbone.begin(); b != backbone.end(); ++b) {
cout << b->first << " "
<< b->first->id << " "
<< b->second.ref_position << " "
<< b->second.cigar << endl
<< b->first->seq << endl;
}
*/
}
if (graph->size == 1 && allN(ref) || graph->size == 0) {
if (params.debug) {
cerr << "DAG is empty (1 node, all N). Alignment is irrelevant." << endl;
}
emptyDAG = true;
} else {
emptyDAG = false;
}
}
AlignmentStats stats_before;
bool was_mapped = alignment.IsMapped();
bool has_realigned = false;
if (was_mapped) {
if (dag_start_position + dag_window_size < alignment.GetEndPosition()) {
ref = reference.getSubSequence(seqname,
max((long int) 0, dag_start_position),
alignment.GetEndPosition() - dag_start_position); // 0/1 conversion
}
}
if (params.debug) {
if (emptyDAG) {
cerr << "cannot realign against empty (all-N single node) graph" << endl;
}
}
if (!emptyDAG && shouldRealign(alignment, ref, dag_start_position, params, stats_before)) {
++total_realigned;
if (params.debug) {
cerr << "realigning: " << alignment.Name
<< " " << alignment.QueryBases << endl
<< " aligned @ " << alignment.Position
<< " to variant graph over "
<< seqname
<< ":" << dag_start_position
<< "-" << dag_start_position + dag_window_size << endl;
}
//{
try {
Cigar flat_cigar;
string read = alignment.QueryBases;
string qualities = alignment.Qualities;
int score;
long int position;
string strand;
gssw_graph_mapping* gm =
gswalign(graph,
ref_map,
read,
qualities,
params,
position,
score,
flat_cigar,
strand,
nt_table,
mat);
//
gssw_graph_mapping_destroy(gm);
if (params.dry_run) {
if (strand == "-" && !alignment.IsMapped()) {
read = reverseComplement(read);
}
cout << read << endl;
cout << graph_mapping_to_string(gm) << endl;
cout << score << " " << strand << " "
<< position << " "
<< flat_cigar << endl;
} else {
/*
if (strand == "-") {
read = reverseComplement(trace_report.read);
}
*/
// TODO the qualities are not on the right side of the read
if (strand == "-" && alignment.IsMapped()) {
// if we're realigning, this is always true unless we swapped strands
alignment.SetIsReverseStrand(true);
//reverse(alignment.Qualities.begin(), alignment.Qualities.end()); // reverse qualities
}
//alignment.QueryBases = reverseComplement(trace_report.read);
alignment.QueryBases = read;
alignment.Qualities = qualities;
alignment.Position = position;// + 1;// + 1;//(trace_report.node->position - 1) + trace_report.x;
alignment.SetIsMapped(true);
if (!alignment.MapQuality) {
alignment.MapQuality = 20; // horrible hack... at least approximate with alignment mismatches against graph
}
// check if somehow we've ended up with an indel at the ends
// if so, grab the reference sequence right beyond it and add
// a single match to the cigar, allowing variant detection methods
// to run on the results without internal modification
Cigar& cigar = flat_cigar;
//cerr << flat_cigar << " " << flat_cigar.readLen() << " " << flat_cigar.refLen() << endl;
int flankSize = params.flatten_flank;
if (cigar.front().isIndel() ||
(cigar.front().isSoftclip() && cigar.at(1).isIndel())) {
alignment.Position -= flankSize;
string refBase = reference.getSubSequence(seqname, alignment.Position, flankSize);
if (cigar.front().isSoftclip()) {
alignment.QueryBases.erase(alignment.QueryBases.begin(),
alignment.QueryBases.begin()+cigar.front().length);
alignment.Qualities.erase(alignment.Qualities.begin(),
alignment.Qualities.begin()+cigar.front().length);
cigar.erase(cigar.begin());
}
alignment.QueryBases.insert(0, refBase);
alignment.Qualities.insert(0, string(flankSize, shortInt2QualityChar(30)));
Cigar newCigar; newCigar.push_back(CigarElement(flankSize, 'M'));
newCigar.append(flat_cigar);
flat_cigar = newCigar;
}
if (cigar.back().isIndel() ||
(cigar.back().isSoftclip() && cigar.at(cigar.size()-2).isIndel())) {
string refBase = reference.getSubSequence(seqname,
alignment.Position
+ flat_cigar.refLen(),
flankSize);
if (cigar.back().isSoftclip()) {
alignment.QueryBases.erase(alignment.QueryBases.end()-cigar.back().length,
alignment.QueryBases.end());
alignment.Qualities.erase(alignment.Qualities.end()-cigar.back().length,
alignment.Qualities.end());
cigar.pop_back();
}
Cigar newCigar; newCigar.push_back(CigarElement(flankSize, 'M'));
flat_cigar.append(newCigar);
//flat_cigar.append(newCigar);
alignment.QueryBases.append(refBase);
alignment.Qualities.append(string(flankSize, shortInt2QualityChar(30)));
}
flat_cigar.toCigarData(alignment.CigarData);
//cerr << flat_cigar << " " << flat_cigar.readLen() << " " << flat_cigar.refLen() << endl;
if (dag_start_position + dag_window_size < alignment.GetEndPosition()) {
ref = reference.getSubSequence(seqname,
max((long int) 0, dag_start_position),
alignment.GetEndPosition() - dag_start_position); // 0/1 conversion
}
AlignmentStats stats_after;
countMismatchesAndGaps(alignment, flat_cigar, ref, dag_start_position, stats_after, params.debug);
/*
if ((!was_mapped || (stats_before.softclip_qsum >= stats_after.softclip_qsum
&& stats_before.mismatch_qsum >= stats_after.mismatch_qsum))
&& acceptRealignment(alignment, ref, dag_start_position, params, stats_after)) {
*/
/*
if ((!was_mapped || (stats_before.softclip_qsum + stats_before.mismatch_qsum
>= stats_after.softclip_qsum + stats_after.mismatch_qsum))
&& acceptRealignment(alignment, ref, dag_start_position, params, stats_after)) {
*/
// we accept the new alignment if...
if (!was_mapped // it wasn't mapped previously
// or if we have removed soft clips or mismatches (per quality) from the alignment
//|| ((stats_before.softclip_qsum >= stats_after.softclip_qsum
// && stats_before.mismatch_qsum >= stats_after.mismatch_qsum)
|| ((stats_before.softclip_qsum + stats_before.mismatch_qsum
>= stats_after.softclip_qsum + stats_after.mismatch_qsum)
// and if we have added gaps, we have added them to remove mismatches or softclips
&& (stats_before.gaps >= stats_after.gaps // accept any time we reduce gaps while not increasing softclips/mismatches
|| (stats_before.gaps < stats_after.gaps // and allow gap increases when they improve the alignment
&& (stats_before.softclip_qsum
+ stats_before.mismatch_qsum
>
stats_after.softclip_qsum
+ stats_after.mismatch_qsum))))
// and the alignment must not have more than the acceptable number of gaps, softclips, or mismatches
// as provided in input parameters
&& acceptRealignment(alignment, ref, dag_start_position, params, stats_after)) {
// keep the alignment
// TODO require threshold of softclips to keep alignment (or count of gaps, mismatches,...)
if (params.debug) {
cerr << "realigned " << alignment.Name << " to graph, which it maps to with "
<< stats_after.mismatch_qsum << "q in mismatches and "
<< stats_after.softclip_qsum << "q in soft clips" << endl;
}
++total_improved;
has_realigned = true;
} else {
// reset to old version of alignment
if (params.debug) {
cerr << "failed realignment of " << alignment.Name << " to graph, which it maps to with: "
<< stats_after.mismatch_qsum << "q in mismatches " << "(vs " << stats_before.mismatch_qsum << "q before), and "
<< stats_after.softclip_qsum << "q in soft clips " << "(vs " << stats_before.softclip_qsum << "q before) " << endl;
}
has_realigned = false;
alignment = originalAlignment;
}
}
//} // try block
} catch (...) {
cerr << "exception when realigning " << alignment.Name
<< " at position " << referenceIDToName[alignment.RefID]
<< ":" << alignment.Position
<< " " << alignment.QueryBases << endl;
// reset to original alignment
has_realigned = false;
alignment = originalAlignment;
}
}
// ensure correct order if alignments move
long int maxOutputPos = initialAlignmentPosition - dag_window_size;
// if we switched sequences we need to flush out all the reads from the previous one
string lastSeqname = currentSeqname;
if (seqname != currentSeqname) {
// so the max output position is set past the end of the last chromosome
if (!currentSeqname.empty()) {
maxOutputPos = reference.sequenceLength(currentSeqname) + dag_window_size;
}
currentSeqname = seqname;
}
if (!params.dry_run) {
map<long int, vector<BamAlignment> >::iterator p = alignmentSortQueue.begin();
for ( ; p != alignmentSortQueue.end(); ++p) {
// except if we are running in unsorted mode, stop when we are at the window size
if (!params.unsorted_output && p->first > maxOutputPos) {
break; // no more to do
} else {
for (vector<BamAlignment>::iterator a = p->second.begin(); a != p->second.end(); ++a) {
writer.SaveAlignment(*a);
}
}
}
if (p != alignmentSortQueue.begin()) {
alignmentSortQueue.erase(alignmentSortQueue.begin(), p);
}
if (!params.only_realigned || has_realigned) {
alignmentSortQueue[alignment.Position].push_back(alignment);
}
}
} // end GetNextAlignment loop
if (!params.dry_run) {
map<long int, vector<BamAlignment> >::iterator p = alignmentSortQueue.begin();
for ( ; p != alignmentSortQueue.end(); ++p) {
for (vector<BamAlignment>::iterator a = p->second.begin(); a != p->second.end(); ++a)
writer.SaveAlignment(*a);
}
}
gssw_graph_destroy(graph);
free(nt_table);
free(mat);
reader.Close();
writer.Close();
if (params.debug) {
cerr << "total reads:\t" << total_reads << endl;
cerr << "realigned:\t" << total_realigned << endl;
cerr << "improved:\t" << total_improved << endl;
}
}