// Build the graph by expanding nodes on the frontier
void StringGraphGenerator::buildGraph(FrontierQueue& queue)
{
while(!queue.empty())
{
if(queue.size() > 200)
break;
GraphFrontier node = queue.front();
queue.pop();
if(node.pVertex->getColor() == EXPLORED_COLOR)
continue; // node has been visited already
// Search the FM-index for the current vertex
SeqRecord record;
record.id = node.pVertex->getID();
record.seq = node.pVertex->getSeq().toString();
OverlapBlockList blockList;
assert(blockList.empty());
m_pOverlapper->overlapRead(record, m_minOverlap, &blockList);
// Update the graph and the frontier queue with newly found vertices
updateGraphAndQueue(node, queue, blockList);
node.pVertex->setColor(EXPLORED_COLOR);
}
m_pGraph->setColors(GC_WHITE);
}
// iterate through obList and determine the overlaps that are irreducible. This function is recursive.
// The final overlap blocks corresponding to irreducible overlaps are written to pOBFinal.
// Invariant: the blocks are ordered in descending order of the overlap size so that the longest overlap is first.
// Invariant: each block corresponds to the same extension of the root sequence w.
void OverlapAlgorithm::_processIrreducibleBlocksExactIterative(const BWT* pBWT, const BWT* pRevBWT,
OverlapBlockList& inList,
OverlapBlockList* pOBFinal) const
{
if(inList.empty())
return;
// We store the overlap blocks in groups of blocks that have the same right-extension.
// When a branch is found, the groups are split based on the extension
typedef std::list<OverlapBlockList> BlockGroups;
BlockGroups blockGroups;
blockGroups.push_back(inList);
int numExtensions = 0;
int numBranches = 0;
while(!blockGroups.empty())
{
// Perform one extenion round for each group.
// If the top-level block has ended, push the result
// to the final list and remove the group from processing
BlockGroups::iterator groupIter = blockGroups.begin();
BlockGroups incomingGroups; // Branched blocks are placed here
while(groupIter != blockGroups.end())
{
OverlapBlockList& currList = *groupIter;
bool bEraseGroup = false;
// Count the extensions in the top level (longest) blocks first
int topLen = currList.front().overlapLen;
AlphaCount64 ext_count;
OBLIter blockIter = currList.begin();
while(blockIter != currList.end() && blockIter->overlapLen == topLen)
{
ext_count += blockIter->getCanonicalExtCount(pBWT, pRevBWT);
++blockIter;
}
// Three cases:
// 1) The top level block has ended as it contains the extension $. Output TLB and end.
// 2) There is a singular unique extension base for all the blocks. Update the blocks and continue.
// 3) There are multiple extension bases, split the block group and continue.
// If some block other than the TLB ended, it must be contained within the TLB and it is not output
// or considered further.
// Likewise if multiple distinct strings in the TLB ended, we only output the top one. The rest
// must have the same sequence as the top one and are hence considered to be contained with the top element.
if(ext_count.get('$') > 0)
{
// An irreducible overlap has been found. It is possible that there are two top level blocks
// (one in the forward and reverse direction). Since we can't decide which one
// contains the other at this point, we output hits to both. Under a fixed
// length string assumption one will be contained within the other and removed later.
OBLIter tlbIter = currList.begin();
while(tlbIter != currList.end() && tlbIter->overlapLen == topLen)
{
// Ensure the tlb is actually terminal and not a substring block
AlphaCount64 test_count = tlbIter->getCanonicalExtCount(pBWT, pRevBWT);
if(test_count.get('$') == 0)
{
std::cerr << "Error: substring read found during overlap computation.\n";
std::cerr << "Please run sga rmdup before sga overlap\n";
exit(EXIT_FAILURE);
}
// Perform the final right-update to make the block terminal
OverlapBlock branched = *tlbIter;
BWTAlgorithms::updateBothR(branched.ranges, '$', branched.getExtensionBWT(pBWT, pRevBWT));
pOBFinal->push_back(branched);
#ifdef DEBUGOVERLAP
std::cout << "[IE] TLB of length " << branched.overlapLen << " has ended\n";
std::cout << "[IE]\tBlock data: " << branched << "\n";
#endif
++tlbIter;
}
// Set the flag to erase this group, it is finished
bEraseGroup = true;
}
else
{
// Count the extension for the rest of the blocks
while(blockIter != currList.end())
{
ext_count += blockIter->getCanonicalExtCount(pBWT, pRevBWT);
++blockIter;
}
if(ext_count.hasUniqueDNAChar())
{
// Update all the blocks using the unique extension character
// This character is in the canonical representation wrt to the query
char b = ext_count.getUniqueDNAChar();
updateOverlapBlockRangesRight(pBWT, pRevBWT, currList, b);
numExtensions++;
bEraseGroup = false;
}
else
{
for(size_t idx = 0; idx < DNA_ALPHABET_SIZE; ++idx)
{
char b = ALPHABET[idx];
if(ext_count.get(b) > 0)
{
numBranches++;
OverlapBlockList branched = currList;
updateOverlapBlockRangesRight(pBWT, pRevBWT, branched, b);
incomingGroups.push_back(branched);
bEraseGroup = true;
}
}
}
}
if(bEraseGroup)
groupIter = blockGroups.erase(groupIter);
else
++groupIter;
}
// Splice in the newly branched blocks, if any
blockGroups.splice(blockGroups.end(), incomingGroups);
}
}
// Classify the blocks in obList as irreducible, transitive or substrings. The irreducible blocks are
// put into pOBFinal. The remaining are discarded.
// Invariant: the blocks are ordered in descending order of the overlap size so that the longest overlap is first.
void OverlapAlgorithm::_processIrreducibleBlocksInexact(const BWT* pBWT, const BWT* pRevBWT,
OverlapBlockList& activeList,
OverlapBlockList* pOBFinal) const
{
if(activeList.empty())
return;
// The activeList contains all the blocks that are not yet right terminal
// Count the extensions in the top level (longest) blocks first
bool all_eliminated = false;
while(!activeList.empty() && !all_eliminated)
{
// The terminalBlock list contains all the blocks that became right-terminal
// in the current extension round.
OverlapBlockList terminalList;
OverlapBlockList potentialContainedList;
// Perform a single round of extension, any terminal blocks
// are moved to the terminated list
extendActiveBlocksRight(pBWT, pRevBWT, activeList, terminalList, potentialContainedList);
// Compare the blocks in the contained list against the other terminal and active blocks
// If they are a substring match to any of these, discard them
OverlapBlockList::iterator containedIter = potentialContainedList.begin();
for(; containedIter != potentialContainedList.end(); ++containedIter)
{
if(!isBlockSubstring(*containedIter, terminalList, m_errorRate) &&
!isBlockSubstring(*containedIter, activeList, m_errorRate))
{
// Not a substring, move to terminal list
terminalList.push_back(*containedIter);
//std::cout << "Contained block kept: " << containedIter->overlapLen << "\n";
}
else
{
//std::cout << "Contained block found and removed: " << containedIter->overlapLen << "\n";
}
}
// Using the terminated blocks, mark as eliminated any active blocks
// that form a valid overlap to the terminal block. These are transitive edges
// We do not compare two terminal blocks, we don't consider these overlaps to be
// transitive
OverlapBlockList::iterator terminalIter = terminalList.begin();
for(; terminalIter != terminalList.end(); ++terminalIter)
{
#ifdef DEBUGOVERLAP
std::cout << "[II] ***TLB of length " << terminalIter->overlapLen << " has ended\n";
#endif
all_eliminated = true;
OverlapBlockList::iterator activeIter = activeList.begin();
for(; activeIter != activeList.end(); ++activeIter)
{
if(activeIter->isEliminated)
continue; // skip previously marked blocks
// Two conditions must be met for a block to be transitive wrt terminal:
// 1) It must have a strictly shorter overlap than the terminal block
// 2) The error rate between the block and terminal must be less than the threshold
double inferredErrorRate = calculateBlockErrorRate(*terminalIter, *activeIter);
if(activeIter->overlapLen < terminalIter->overlapLen &&
isErrorRateAcceptable(inferredErrorRate, m_errorRate))
{
#ifdef DEBUGOVERLAP_2
std::cout << "Marking block of length " << activeIter->overlapLen << " as eliminated\n";
#endif
activeIter->isEliminated = true;
}
else
{
all_eliminated = false;
}
}
// Move this block to the final list if it has not been previously marked eliminated
if(!terminalIter->isEliminated)
{
#ifdef DEBUGOVERLAP
std::cout << "[II] Adding block " << *terminalIter << " to final list\n";
//std::cout << " extension: " << terminalIter->forwardHistory << "\n";
#endif
pOBFinal->push_back(*terminalIter);
}
}
}
activeList.clear();
}
OverlapResult OverlapAlgorithm::overlapReadInexact(const SeqRecord& read, int minOverlap, OverlapBlockList* pOBOut) const
{
OverlapResult result;
OverlapBlockList obWorkingList;
std::string seq = read.seq.toString();
#ifdef DEBUGOVERLAP
std::cout << "\n\n***Overlapping read " << read.id << " suffix\n";
#endif
// Match the suffix of seq to prefixes
// findInexact returns false is the maximum search time was exceeded. In this
// case we dont run any of the subsequent commands and return no overlaps.
bool valid = true;
valid = findOverlapBlocksInexact(seq, m_pBWT, m_pRevBWT, sufPreAF,
minOverlap, &obWorkingList, pOBOut, result);
if(valid)
valid = findOverlapBlocksInexact(complement(seq), m_pRevBWT, m_pBWT, prePreAF,
minOverlap, &obWorkingList, pOBOut, result);
if(valid)
{
if(m_bIrreducible)
{
computeIrreducibleBlocks(m_pBWT, m_pRevBWT, &obWorkingList, pOBOut);
obWorkingList.clear();
}
else
{
pOBOut->splice(pOBOut->end(), obWorkingList);
assert(obWorkingList.empty());
}
}
#ifdef DEBUGOVERLAP
std::cout << "\n\n***Overlapping read " << read.id << " prefix\n";
#endif
// Match the prefix of seq to suffixes
if(valid)
valid = findOverlapBlocksInexact(reverseComplement(seq), m_pBWT, m_pRevBWT, sufSufAF, minOverlap, &obWorkingList, pOBOut, result);
if(valid)
valid = findOverlapBlocksInexact(reverse(seq), m_pRevBWT, m_pBWT, preSufAF, minOverlap, &obWorkingList, pOBOut, result);
if(valid)
{
if(m_bIrreducible)
{
computeIrreducibleBlocks(m_pBWT, m_pRevBWT, &obWorkingList, pOBOut);
obWorkingList.clear();
}
else
{
pOBOut->splice(pOBOut->end(), obWorkingList);
assert(obWorkingList.empty());
}
}
if(!valid)
{
pOBOut->clear();
result.isSubstring = false;
result.searchAborted = true;
return result;
}
return result;
}