void TargetsManager::TrimAmpliseqPrimers(Alignment *rai, int unmerged_target_hint) const
{
// set these before any trimming
rai->align_start = rai->alignment.Position;
rai->align_end = rai->alignment.GetEndPosition(false, true);
rai->old_cigar = rai->alignment.CigarData;
if (not trim_ampliseq_primers)
return;
// Step 1: Find the first potential target region
int target_idx = unmerged_target_hint;
while (target_idx and (rai->alignment.RefID < unmerged[target_idx].chr or
(rai->alignment.RefID == unmerged[target_idx].chr and rai->alignment.Position < unmerged[target_idx].end)))
--target_idx;
while (target_idx < (int)unmerged.size() and (rai->alignment.RefID > unmerged[target_idx].chr or
(rai->alignment.RefID == unmerged[target_idx].chr and rai->alignment.Position >= unmerged[target_idx].end)))
++target_idx;
// Step 2: Iterate over potential target regions, evaluate fit, pick the best fit
int best_target_idx = -1;
int best_fit_penalty = 100;
int best_overlap = 0;
while (target_idx < (int)unmerged.size() and rai->alignment.RefID == unmerged[target_idx].chr and rai->end >= unmerged[target_idx].begin) {
int read_start = rai->alignment.Position;
int read_end = rai->end;
int read_prefix_size = unmerged[target_idx].begin - read_start;
int read_postfix_size = read_end - unmerged[target_idx].end;
int overlap = min(unmerged[target_idx].end, read_end) - max(unmerged[target_idx].begin, read_start);
int fit_penalty = 100;
if (not rai->alignment.IsReverseStrand()) {
if (read_prefix_size > 0)
fit_penalty = min(read_prefix_size,50) + max(0,50-overlap);
else
fit_penalty = min(-3*read_prefix_size,50) + max(0,50-overlap);
} else {
if (read_postfix_size > 0)
fit_penalty = min(read_postfix_size,50) + max(0,50-overlap);
else
fit_penalty = min(-3*read_postfix_size,50) + max(0,50-overlap);
}
if (read_prefix_size > 0 and read_postfix_size > 0)
fit_penalty -= 10;
if ((best_fit_penalty > fit_penalty and overlap > 0) or (best_fit_penalty == fit_penalty and overlap > best_overlap)) {
best_fit_penalty = fit_penalty;
best_target_idx = target_idx;
best_overlap = overlap;
}
++target_idx;
}
if (best_target_idx == -1) {
rai->filtered = true;
return;
}
// Step 3: Do the actual primer trimming.
//
// For now, only adjust Position and Cigar.
// Later, also adjust MD tag.
// Even later, ensure the reads stay sorted, so no extra sorting is required outside of tvc
vector<CigarOp>& old_cigar = rai->alignment.CigarData;
vector<CigarOp> new_cigar;
new_cigar.reserve(old_cigar.size() + 2);
vector<CigarOp>::iterator old_op = old_cigar.begin();
int ref_pos = rai->alignment.Position;
// 3A: Cigar ops left of the target
int begin = unmerged[best_target_idx].begin + unmerged[best_target_idx].trim_left;
if (begin > unmerged[best_target_idx].end)
begin = unmerged[best_target_idx].end;
int end = unmerged[best_target_idx].end - unmerged[best_target_idx].trim_right;
if (end <= begin)
end = begin;
while (old_op != old_cigar.end() and ref_pos <= begin) {
if (old_op->Type == 'H') {
++old_op;
continue;
}
if (old_op->Type == 'S' or old_op->Type == 'I') {
if (new_cigar.empty())
new_cigar.push_back(CigarOp('S'));
new_cigar.back().Length += old_op->Length;
++old_op;
continue;
}
unsigned int gap = begin - ref_pos;
if (gap == 0)
break;
if (old_op->Type == 'M' or old_op->Type == 'N') {
if (new_cigar.empty())
new_cigar.push_back(CigarOp('S'));
if (old_op->Length > gap) {
new_cigar.back().Length += gap;
old_op->Length -= gap;
ref_pos += gap;
break;
} else {
new_cigar.back().Length += old_op->Length;
ref_pos += old_op->Length;
++old_op;
continue;
}
}
if (old_op->Type == 'D') {
if (old_op->Length > gap) {
old_op->Length -= gap;
ref_pos += gap;
break;
} else {
ref_pos += old_op->Length;
++old_op;
continue;
}
}
}
// 3B: Cigar ops in the middle of the target
rai->alignment.Position = ref_pos;
while (old_op != old_cigar.end() and ref_pos < end) {
if (old_op->Type == 'H') {
++old_op;
continue;
}
unsigned int gap = end - ref_pos;
if (old_op->Type == 'S' or old_op->Type == 'I') {
new_cigar.push_back(*old_op);
++old_op;
continue;
}
if (old_op->Type == 'M' or old_op->Type == 'N') {
new_cigar.push_back(CigarOp(old_op->Type));
if (old_op->Length > gap) {
new_cigar.back().Length = gap;
old_op->Length -= gap;
ref_pos += gap;
break;
} else {
new_cigar.back().Length = old_op->Length;
ref_pos += old_op->Length;
++old_op;
continue;
}
}
if (old_op->Type == 'D') {
new_cigar.push_back(CigarOp('D'));
if (old_op->Length > gap) {
new_cigar.back().Length = gap;
old_op->Length -= gap;
ref_pos += gap;
break;
} else {
new_cigar.back().Length = old_op->Length;
ref_pos += old_op->Length;
++old_op;
continue;
}
}
}
// 3C: Cigar ops to the right of the target
for (; old_op != old_cigar.end(); ++old_op) {
if (old_op->Type == 'H' or old_op->Type == 'D')
continue;
if (new_cigar.empty() or new_cigar.back().Type != 'S')
new_cigar.push_back(CigarOp('S'));
new_cigar.back().Length += old_op->Length;
}
rai->alignment.CigarData.swap(new_cigar);
// Debugging info
stringstream ZL;
ZL << unmerged[best_target_idx].name << ":" << best_fit_penalty << ":" << best_overlap;
rai->alignment.AddTag("ZL", "Z", ZL.str());
}
// Attempts to left-realign all the indels represented by the alignment cigar.
//
// This is done by shifting all indels as far left as they can go without
// mismatch, then merging neighboring indels of the same class. leftAlign
// updates the alignment cigar with changes, and returns true if realignment
// changed the alignment cigar.
//
// To left-align, we move multi-base indels left by their own length as long as
// the preceding bases match the inserted or deleted sequence. After this
// step, we handle multi-base homopolymer indels by shifting them one base to
// the left until they mismatch the reference.
//
// To merge neighboring indels, we iterate through the set of left-stabilized
// indels. For each indel we add a new cigar element to the new cigar. If a
// deletion follows a deletion, or an insertion occurs at the same place as
// another insertion, we merge the events by extending the previous cigar
// element.
//
// In practice, we must call this function until the alignment is stabilized.
//
bool leftAlign(BamAlignment& alignment, string& referenceSequence) {
int arsOffset = 0; // pointer to insertion point in aligned reference sequence
string alignedReferenceSequence = referenceSequence;
int aabOffset = 0;
string alignmentAlignedBases = alignment.QueryBases;
// store information about the indels
vector<FBIndelAllele> indels;
int rp = 0; // read position, 0-based relative to read
int sp = 0; // sequence position
string softBegin;
string softEnd;
stringstream cigar_before, cigar_after;
for (vector<CigarOp>::const_iterator c = alignment.CigarData.begin();
c != alignment.CigarData.end(); ++c) {
unsigned int l = c->Length;
char t = c->Type;
cigar_before << l << t;
if (t == 'M') { // match or mismatch
sp += l;
rp += l;
} else if (t == 'D') { // deletion
indels.push_back(FBIndelAllele(false, l, sp, rp, referenceSequence.substr(sp, l)));
alignmentAlignedBases.insert(rp + aabOffset, string(l, '-'));
aabOffset += l;
sp += l; // update reference sequence position
} else if (t == 'I') { // insertion
indels.push_back(FBIndelAllele(true, l, sp, rp, alignment.QueryBases.substr(rp, l)));
alignedReferenceSequence.insert(sp + softBegin.size() + arsOffset, string(l, '-'));
arsOffset += l;
rp += l;
} else if (t == 'S') { // soft clip, clipped sequence present in the read not matching the reference
// remove these bases from the refseq and read seq, but don't modify the alignment sequence
if (rp == 0) {
alignedReferenceSequence = string(l, '*') + alignedReferenceSequence;
softBegin = alignmentAlignedBases.substr(0, l);
} else {
alignedReferenceSequence = alignedReferenceSequence + string(l, '*');
softEnd = alignmentAlignedBases.substr(alignmentAlignedBases.size() - l, l);
}
rp += l;
} else if (t == 'H') { // hard clip on the read, clipped sequence is not present in the read
} else if (t == 'N') { // skipped region in the reference not present in read, aka splice
sp += l;
}
}
int alignedLength = sp;
// if no indels, return the alignment
if (indels.empty()) { return false; }
// for each indel, from left to right
// while the indel sequence repeated to the left and we're not matched up with the left-previous indel
// move the indel left
vector<FBIndelAllele>::iterator previous = indels.begin();
for (vector<FBIndelAllele>::iterator id = indels.begin(); id != indels.end(); ++id) {
// left shift by repeats
//
// from 1 base to the length of the indel, attempt to shift left
// if the move would cause no change in alignment optimality (no
// introduction of mismatches, and by definition no change in gap
// length), move to the new position.
// in practice this moves the indel left when we reach the size of
// the repeat unit.
//
int steppos, readsteppos;
FBIndelAllele& indel = *id;
int i = 1;
while (i <= indel.length) {
int steppos = indel.position - i;
int readsteppos = indel.readPosition - i;
while (steppos >= 0 && readsteppos >= 0
&& indel.sequence == referenceSequence.substr(steppos, indel.length)
&& indel.sequence == alignment.QueryBases.substr(readsteppos, indel.length)
&& (id == indels.begin()
|| (previous->insertion && steppos >= previous->position)
|| (!previous->insertion && steppos >= previous->position + previous->length))) {
indel.position -= i;
indel.readPosition -= i;
steppos = indel.position - i;
readsteppos = indel.readPosition - i;
}
do {
++i;
} while (i <= indel.length && indel.length % i != 0);
}
// left shift indels with exchangeable flanking sequence
//
// for example:
//
// GTTACGTT GTTACGTT
// GT-----T ----> G-----TT
//
// GTGTGACGTGT GTGTGACGTGT
// GTGTG-----T ----> GTG-----TGT
//
// GTGTG-----T GTG-----TGT
// GTGTGACGTGT ----> GTGTGACGTGT
//
//
steppos = indel.position - 1;
readsteppos = indel.readPosition - 1;
while (steppos >= 0 && readsteppos >= 0
&& alignment.QueryBases.at(readsteppos) == referenceSequence.at(steppos)
&& alignment.QueryBases.at(readsteppos) == indel.sequence.at(indel.sequence.size() - 1)
&& (id == indels.begin()
|| (previous->insertion && indel.position - 1 >= previous->position)
|| (!previous->insertion && indel.position - 1 >= previous->position + previous->length))) {
indel.sequence = indel.sequence.at(indel.sequence.size() - 1) + indel.sequence.substr(0, indel.sequence.size() - 1);
indel.position -= 1;
indel.readPosition -= 1;
steppos = indel.position - 1;
readsteppos = indel.readPosition - 1;
}
// tracks previous indel, so we don't run into it with the next shift
previous = id;
}
// bring together floating indels
// from left to right
// check if we could merge with the next indel
// if so, adjust so that we will merge in the next step
if (indels.size() > 1) {
previous = indels.begin();
for (vector<FBIndelAllele>::iterator id = (indels.begin() + 1); id != indels.end(); ++id) {
FBIndelAllele& indel = *id;
// parsimony: could we shift right and merge with the previous indel?
// if so, do it
int prev_end_ref = previous->insertion ? previous->position : previous->position + previous->length;
int prev_end_read = !previous->insertion ? previous->readPosition : previous->readPosition + previous->length;
if (previous->insertion == indel.insertion
&& ((previous->insertion
&& (previous->position < indel.position
&& previous->readPosition + previous->readPosition < indel.readPosition))
||
(!previous->insertion
&& (previous->position + previous->length < indel.position)
&& (previous->readPosition < indel.readPosition)
))) {
if (previous->homopolymer()) {
string seq = referenceSequence.substr(prev_end_ref, indel.position - prev_end_ref);
string readseq = alignment.QueryBases.substr(prev_end_read, indel.position - prev_end_ref);
if (previous->sequence.at(0) == seq.at(0)
&& FBhomopolymer(seq)
&& FBhomopolymer(readseq)) {
previous->position = indel.insertion ? indel.position : indel.position - previous->length;
}
}
else {
int pos = previous->position;
while (pos < (int) referenceSequence.length() &&
((previous->insertion && pos + previous->length <= indel.position)
||
(!previous->insertion && pos + previous->length < indel.position))
&& previous->sequence
== referenceSequence.substr(pos + previous->length, previous->length)) {
pos += previous->length;
}
if (pos < previous->position &&
((previous->insertion && pos + previous->length == indel.position)
||
(!previous->insertion && pos == indel.position - previous->length))
) {
previous->position = pos;
}
}
}
previous = id;
}
}
// for each indel
// if ( we're matched up to the previous insertion (or deletion)
// and it's also an insertion or deletion )
// merge the indels
//
// and simultaneously reconstruct the cigar
vector<CigarOp> newCigar;
if (!softBegin.empty()) {
newCigar.push_back(CigarOp('S', softBegin.size()));
}
vector<FBIndelAllele>::iterator id = indels.begin();
FBIndelAllele last = *id++;
if (last.position > 0) {
newCigar.push_back(CigarOp('M', last.position));
newCigar.push_back(CigarOp((last.insertion ? 'I' : 'D'), last.length));
} else {
newCigar.push_back(CigarOp((last.insertion ? 'I' : 'D'), last.length));
}
int lastend = last.insertion ? last.position : (last.position + last.length);
for (; id != indels.end(); ++id) {
FBIndelAllele& indel = *id;
if (indel.position < lastend) {
cerr << "impossibility?: indel realigned left of another indel" << endl << alignment.Name
<< " " << alignment.Position << endl << alignment.QueryBases << endl;
exit(1);
} else if (indel.position == lastend && indel.insertion == last.insertion) {
CigarOp& op = newCigar.back();
op.Length += indel.length;
} else if (indel.position >= lastend) { // also catches differential indels, but with the same position
newCigar.push_back(CigarOp('M', indel.position - lastend));
newCigar.push_back(CigarOp((indel.insertion ? 'I' : 'D'), indel.length));
}
last = *id;
lastend = last.insertion ? last.position : (last.position + last.length);
}
if (lastend < alignedLength) {
newCigar.push_back(CigarOp('M', alignedLength - lastend));
}
if (!softEnd.empty()) {
newCigar.push_back(CigarOp('S', softEnd.size()));
}
alignment.CigarData = newCigar;
for (vector<CigarOp>::const_iterator c = alignment.CigarData.begin();
c != alignment.CigarData.end(); ++c) {
unsigned int l = c->Length;
char t = c->Type;
cigar_after << l << t;
}
// check if we're realigned
if (cigar_after.str() == cigar_before.str()) {
return false;
} else {
return true;
}
}
