bool
isMateInsertionEvidenceCandidate(
const bam_record& bamRead,
const unsigned minMapq)
{
if (! bamRead.is_paired()) return false;
if (bamRead.isNonStrictSupplement()) return false;
if (bamRead.is_unmapped() || bamRead.is_mate_unmapped()) return false;
if (bamRead.map_qual() < minMapq) return false;
if (bamRead.target_id() < 0) return false;
if (bamRead.mate_target_id() < 0) return false;
if (bamRead.target_id() != bamRead.mate_target_id()) return true;
/// TODO: better candidate definition based on fragment size distro:
static const int minSize(10000);
return (std::abs(bamRead.pos()-bamRead.mate_pos()) >= minSize);
}
FragmentSizeType::index_t
SVLocusScanner::
_getFragmentSizeType(
const bam_record& bamRead,
const unsigned defaultReadGroupIndex) const
{
using namespace FragmentSizeType;
if (bamRead.target_id() != bamRead.mate_target_id()) return DISTANT;
const int32_t fragmentSize(std::abs(bamRead.template_size()));
return classifySize(_stats[defaultReadGroupIndex], fragmentSize);
}
/// extract poorly aligned read ends (semi-aligned and/or soft-clipped)
/// to internal candidate format
static
void
getSVCandidatesFromSemiAligned(
const ReadScannerOptions& opt,
const ReadScannerDerivOptions& dopt,
const bam_record& bamRead,
const SimpleAlignment& bamAlign,
const FRAGSOURCE::index_t fragSource,
const reference_contig_segment& refSeq,
std::vector<SVObservation>& candidates)
{
unsigned leadingMismatchLen(0);
unsigned trailingMismatchLen(0);
pos_t leadingRefPos(0), trailingRefPos(0);
getSVBreakendCandidateSemiAligned(bamRead, bamAlign, refSeq,
dopt.isUseOverlappingPairs,
leadingMismatchLen, leadingRefPos,
trailingMismatchLen, trailingRefPos);
if ((leadingMismatchLen + trailingMismatchLen) >= bamRead.read_size()) return;
using namespace SVEvidenceType;
static const index_t svSource(SEMIALIGN);
// semi-aligned reads don't define a full hypothesis, so they're always evidence for a 'complex' ie. undefined, event
// in a fashion analogous to clipped reads
static const bool isComplex(true);
if (leadingMismatchLen >= opt.minSemiAlignedMismatchLen)
{
const pos_t pos(leadingRefPos);
candidates.push_back(GetSplitSVCandidate(dopt,bamRead.target_id(),pos,pos,svSource, fragSource,isComplex));
}
if (trailingMismatchLen >= opt.minSemiAlignedMismatchLen)
{
const pos_t pos(trailingRefPos);
candidates.push_back(GetSplitSVCandidate(dopt,bamRead.target_id(),pos,pos,svSource, fragSource,isComplex));
}
}
/// get SV candidates from shadow/singleton pairs
/// look for singletons, create candidateSV around conf. interval of shadow position
/// cache singletons? might be needed to remove poor quality shadows.
/// should be able to re-use code, follow soft-clipping example.
static
void
getSVCandidatesFromShadow(
const ReadScannerOptions& opt,
const SVLocusScanner::CachedReadGroupStats& rstats,
const bam_record& localRead,
const SimpleAlignment& localAlign,
const bam_record* remoteReadPtr,
TrackedCandidates& candidates)
{
using namespace SVEvidenceType;
static const index_t svSource(SHADOW);
static const bool isComplex(true);
pos_t singletonGenomePos(0);
int targetId(0);
if (NULL == remoteReadPtr)
{
if (!localRead.is_unmapped()) return;
// need to take care of this case
// need to rely on cached mapq and qname
return;
if (!isGoodShadow(localRead,lastMapq,lastQname,opt.minSingletonMapqGraph))
{
return;
}
singletonGenomePos = localAlign.pos;
targetId = localRead.target_id();
}
else
{
// have both reads, straightforward from here
const bam_record& remoteRead(*remoteReadPtr);
const SimpleAlignment remoteAlign(remoteRead);
if (localRead.is_mate_unmapped())
{
// remote read is shadow candidate
if (!isGoodShadow(remoteRead,localRead.map_qual(),localRead.qname(),opt.minSingletonMapqGraph))
{
return;
}
singletonGenomePos = localAlign.pos;
targetId = remoteRead.target_id();
}
else if (localRead.is_unmapped())
{
// local is shadow candidate
if (!isGoodShadow(localRead,remoteRead.map_qual(),remoteRead.qname(),opt.minSingletonMapqGraph))
{
return;
}
singletonGenomePos = remoteAlign.pos;
targetId = localRead.target_id();
}
else
{
// none unmapped, skip this one
return;
}
}
const pos_t properPairRangeOffset = static_cast<pos_t>(rstats.properPair.min + (rstats.properPair.max-rstats.properPair.min)/2);
const pos_t shadowGenomePos = singletonGenomePos + properPairRangeOffset;
candidates.push_back(GetSplitSVCandidate(opt,targetId,shadowGenomePos,shadowGenomePos, svSource, isComplex));
}