bool visit(StringGraph*, Vertex* pVertex)
{
if(pVertex->getColor() == GC_BLACK)
return false;
if((int)pVertex->getSeqLen() >= m_minLength)
{
std::stringstream idss;
idss << "unplaced-" << m_numUnplaced++;
writeFastaRecord(m_pWriter, idss.str(), pVertex->getSeq().toString());
}
return false;
}
// write the unplaced sequences of length at least minLength using pWriter
void MapSequenceCollection::writeUnplaced(std::ostream* pWriter, int minLength)
{
int numUnplaced = 0;
for(SMPMap::iterator iter = m_map.begin(); iter != m_map.end(); ++iter)
{
if(!iter->second.isPlaced && (int)iter->second.sequence.size() >= minLength)
{
std::stringstream idss;
idss << "unplaced-" << numUnplaced++;
writeFastaRecord(pWriter, idss.str(), iter->second.sequence);
}
}
}
bool SGSmoothingVisitor::visit(StringGraph* pGraph, Vertex* pVertex)
{
(void)pGraph;
if(pVertex->getColor() == GC_RED)
return false;
bool found = false;
for(size_t idx = 0; idx < ED_COUNT; idx++)
{
EdgeDir dir = EDGE_DIRECTIONS[idx];
EdgePtrVec edges = pVertex->getEdges(dir);
if(edges.size() <= 1)
continue;
for(size_t i = 0; i < edges.size(); ++i)
{
if(edges[i]->getEnd()->getColor() == GC_RED)
return false;
}
//std::cout << "Smoothing " << pVertex->getID() << "\n";
const int MAX_WALKS = 10;
const int MAX_DISTANCE = 5000;
bool bIsDegenerate = false;
bool bFailGapCheck = false;
bool bFailDivergenceCheck = false;
bool bFailIndelSizeCheck = false;
SGWalkVector variantWalks;
SGSearch::findVariantWalks(pVertex, dir, MAX_DISTANCE, MAX_WALKS, variantWalks);
if(variantWalks.size() > 0)
{
found = true;
size_t selectedIdx = -1;
size_t selectedCoverage = 0;
// Calculate the minimum amount overlapped on the start/end vertex.
// This is used to properly extract the sequences from walks that represent the variation.
int minOverlapX = std::numeric_limits<int>::max();
int minOverlapY = std::numeric_limits<int>::max();
for(size_t i = 0; i < variantWalks.size(); ++i)
{
if(variantWalks[i].getNumEdges() <= 1)
bIsDegenerate = true;
// Calculate the walk coverage using the internal vertices of the walk.
// The walk with the highest coverage will be retained
size_t walkCoverage = 0;
for(size_t j = 1; j < variantWalks[i].getNumVertices() - 1; ++j)
walkCoverage += variantWalks[i].getVertex(j)->getCoverage();
if(walkCoverage > selectedCoverage || selectedCoverage == 0)
{
selectedIdx = i;
selectedCoverage = walkCoverage;
}
Edge* pFirstEdge = variantWalks[i].getFirstEdge();
Edge* pLastEdge = variantWalks[i].getLastEdge();
if((int)pFirstEdge->getMatchLength() < minOverlapX)
minOverlapX = pFirstEdge->getMatchLength();
if((int)pLastEdge->getTwin()->getMatchLength() < minOverlapY)
minOverlapY = pLastEdge->getTwin()->getMatchLength();
}
// Calculate the strings for each walk that represent the region of variation
StringVector walkStrings;
for(size_t i = 0; i < variantWalks.size(); ++i)
{
Vertex* pStartVertex = variantWalks[i].getStartVertex();
Vertex* pLastVertex = variantWalks[i].getLastVertex();
assert(pStartVertex != NULL && pLastVertex != NULL);
std::string full = variantWalks[i].getString(SGWT_START_TO_END);
int posStart = 0;
int posEnd = 0;
if(dir == ED_ANTISENSE)
{
// pLast -----------
// pStart ------------
// full --------------------
// out ----
posStart = pLastVertex->getSeqLen() - minOverlapY;
posEnd = full.size() - (pStartVertex->getSeqLen() - minOverlapX);
}
else
{
// pStart --------------
// pLast -----------
// full ---------------------
// out ----
posStart = pStartVertex->getSeqLen() - minOverlapX; // match start position
posEnd = full.size() - (pLastVertex->getSeqLen() - minOverlapY); // match end position
}
std::string out;
if(posEnd > posStart)
out = full.substr(posStart, posEnd - posStart);
walkStrings.push_back(out);
}
assert(selectedIdx != (size_t)-1);
SGWalk& selectedWalk = variantWalks[selectedIdx];
assert(selectedWalk.isIndexed());
// Check the divergence of the other walks to this walk
StringVector cigarStrings;
std::vector<int> maxIndel;
std::vector<double> gapPercent; // percentage of matching that is gaps
std::vector<double> totalPercent; // percent of total alignment that is mismatch or gap
cigarStrings.resize(variantWalks.size());
gapPercent.resize(variantWalks.size());
totalPercent.resize(variantWalks.size());
maxIndel.resize(variantWalks.size());
for(size_t i = 0; i < variantWalks.size(); ++i)
{
if(i == selectedIdx)
continue;
// We want to compute the total gap length, total mismatches and percent
// divergence between the two paths.
int matchLen = 0;
int totalDiff = 0;
int gapLength = 0;
int maxGapLength = 0;
// We have to handle the degenerate case where one internal string has zero length
// this can happen when there is an isolated insertion/deletion and the walks are like:
// x -> y -> z
// x -> z
if(walkStrings[selectedIdx].empty() || walkStrings[i].empty())
{
matchLen = std::max(walkStrings[selectedIdx].size(), walkStrings[i].size());
totalDiff = matchLen;
gapLength = matchLen;
}
else
{
AlnAln *aln_global;
aln_global = aln_stdaln(walkStrings[selectedIdx].c_str(), walkStrings[i].c_str(), &aln_param_blast, 1, 1);
// Calculate the alignment parameters
while(aln_global->outm[matchLen] != '\0')
{
if(aln_global->outm[matchLen] == ' ')
totalDiff += 1;
matchLen += 1;
}
std::stringstream cigarSS;
for (int j = 0; j != aln_global->n_cigar; ++j)
{
char cigarOp = "MID"[aln_global->cigar32[j]&0xf];
int cigarLen = aln_global->cigar32[j]>>4;
if(cigarOp == 'I' || cigarOp == 'D')
{
gapLength += cigarLen;
if(gapLength > maxGapLength)
maxGapLength = gapLength;
}
cigarSS << cigarLen;
cigarSS << cigarOp;
}
cigarStrings[i] = cigarSS.str();
aln_free_AlnAln(aln_global);
}
double percentDiff = (double)totalDiff / matchLen;
double percentGap = (double)gapLength / matchLen;
if(percentDiff > m_maxTotalDivergence)
bFailDivergenceCheck = true;
if(percentGap > m_maxGapDivergence)
bFailGapCheck = true;
if(maxGapLength > m_maxIndelLength)
bFailIndelSizeCheck = true;
gapPercent[i] = percentGap;
totalPercent[i] = percentDiff;
maxIndel[i] = maxGapLength;
}
if(bIsDegenerate || bFailGapCheck || bFailDivergenceCheck || bFailIndelSizeCheck)
continue;
// Write the selected path to the variants file as variant 0
int variantIdx = 0;
std::string selectedSequence = selectedWalk.getString(SGWT_START_TO_END);
std::stringstream ss;
ss << "variant-" << m_numRemovedTotal << "/" << variantIdx++;
writeFastaRecord(&m_outFile, ss.str(), selectedSequence);
// The vertex set for each walk is not necessarily disjoint,
// the selected walk may contain vertices that are part
// of other paths. We handle this be initially marking all
// vertices of the
for(size_t i = 0; i < variantWalks.size(); ++i)
{
if(i == selectedIdx)
continue;
SGWalk& currWalk = variantWalks[i];
for(size_t j = 0; j < currWalk.getNumEdges() - 1; ++j)
{
Edge* currEdge = currWalk.getEdge(j);
// If the vertex is also on the selected path, do not mark it
Vertex* currVertex = currEdge->getEnd();
if(!selectedWalk.containsVertex(currVertex->getID()))
{
currEdge->getEnd()->setColor(GC_RED);
}
}
// Write the variant to a file
std::string variantSequence = currWalk.getString(SGWT_START_TO_END);
std::stringstream variantID;
std::stringstream ss;
ss << "variant-" << m_numRemovedTotal << "/" << variantIdx++;
ss << " IGD:" << (double)gapPercent[i] << " ITD:" << totalPercent[i] << " MID: " << maxIndel[i] << " InternalCigar:" << cigarStrings[i];
writeFastaRecord(&m_outFile, ss.str(), variantSequence);
}
if(variantWalks.size() == 2)
m_simpleBubblesRemoved += 1;
else
m_complexBubblesRemoved += 1;
++m_numRemovedTotal;
}
}
