// Examine the first (few?) fragments of a unitig, evaluate if they indicate a join should be made.
static
bool
joinUnitigs_examineEnd(UnitigVector &unitigs,
Unitig *fr,
uint32 idx,
bool frFirstEnd,
vector<joinEntry> &joins) {
uint32 frgIdx = (frFirstEnd) ? (idx) : (fr->ufpath.size() - 1 - idx);
ufNode *frg = &fr->ufpath[frgIdx];
bool frgRev = (frg->position.end < frg->position.bgn);
// Grab the best edge for this end frag. The last arg requests the 3' end if true.
//
// If we're looking at the first read, we want to get:
// 5' - if the frag is forward
// 3' - if the frag is reverse (frgRev == true)
//
// If we're looking at the lat read, we want to get:
// 5' - if the frag is reverse
// 3' - if the frag is forward (frgRev == false)
//
BestEdgeOverlap *bestEdge = OG->getBestEdgeOverlap(frg->ident, (frgRev == frFirstEnd));
uint32 tgtId = bestEdge->fragId();
bool tgt3p = bestEdge->frag3p();
if (tgtId == 0)
// No best edge? Skip it.
return(false);
// Grab the unitig for that best edge.
uint32 toID = fr->fragIn(tgtId);
Unitig *to = unitigs[toID];
if (to->ufpath.size() == 1)
// Joining to something teeny? Don't bother checking further.
return(false);
if (to->id() == fr->id())
// Join to myself? Nope.
return(false);
// Grab the read we have an edge to, an compute the overlapping length and left over length.
ufNode *tgt = &to->ufpath[to->pathPosition(tgtId)];
bool tgtRev = (tgt->position.end < tgt->position.bgn);
// If tgt3p (we overlap to the 3' end) is the same as tgtRev (read is reverse) then the unitig is oriented
// correctly. Otherwise, positions need to be reverse-complemented.
bool toFlip = false;
if ((frFirstEnd == true) && (tgt3p == false) && (tgtRev == false))
// source read is at the start, overlap to 5' and the read is forward, need to flip the target unitig
toFlip = true;
if ((frFirstEnd == true) && (tgt3p == true) && (tgtRev == true))
// source read is at the start, overlap to 3' and the read is reverse, need to flip the target unitig
toFlip = true;
if ((frFirstEnd == false) && (tgt3p == false) && (tgtRev == true))
// source read is at the end, overlap to 5' and the read is reverse, need to flip the target unitig
toFlip = true;
if ((frFirstEnd == false) && (tgt3p == true) && (tgtRev == false))
// source read is at the end, overlap to 3' and the read is forward, need to flip the target unitig
toFlip = true;
uint32 toMin = MIN(tgt->position.bgn, tgt->position.end);
uint32 toMax = MAX(tgt->position.bgn, tgt->position.end);
uint32 toLen = to->getLength();
uint32 frLen = fr->getLength();
if (toFlip) {
toMin = toLen - MAX(tgt->position.bgn, tgt->position.end);
toMax = toLen - MIN(tgt->position.bgn, tgt->position.end);
}
assert(toMin < toMax);
// Our two unitigs are of length frLen and toLen. We are appending some portion of 'to' onto
// 'fr', and 'discarding' the rest. If the 'discarded' piece is larger than the 'fr' unitig, we
// don't want to do the join.
//
// We err on the side of the discarded piece.
uint32 joinLen = 0;
uint32 discLen = 0;
if (frFirstEnd == true) {
joinLen = toMin + frLen; // Prepend the start of 'to' onto 'fr'.
discLen = toLen - toMin;
} else {
joinLen = frLen + toLen - toMax; // Append the end of 'to' onto 'fr'.
discLen = toMax;
}
// If the discard is bigger than us, we do damage by joining.
if (discLen > frLen)
return(false);
// The joined should be much larger and the discarded much smaller.
uint32 maxLen = MAX(frLen, toLen);
uint32 minLen = MIN(frLen, toLen);
double joinChange = (double)joinLen / maxLen;
double discChange = (double)discLen / minLen;
bool isBad = false;
if ((joinChange < 1.10) ||
(0.75 < discChange))
// Bad if we didn't really change sizes.
isBad = true;
if ((1.0 < joinChange) &&
(discChange < 0.5))
// But good if discard is tiny. This occurs if we merge a small with a big. The join change
// is somewhat small (1.05 say) yet most of the smaller unitig is used.
isBad = false;
if (isBad) {
writeLog("joinUnitigs_examineEnd()-- join unitig %6u (%7ubp) frag %6u %s <-> unitig %6u (%7ubp) frag %6u %s <-> length %5.2f %7u and %5.2f %7u BAD\n",
fr->id(), fr->getLength(), frg->ident, (frgRev) ? "rev" : "fwd",
to->id(), to->getLength(), tgt->ident, (tgtRev) ? "rev" : "fwd",
joinChange, joinLen,
discChange, discLen);
return(false);
}
// OK, join.
writeLog("joinUnitigs_examineEnd()-- join unitig %6u (%7ubp) frag %6u %s <-> unitig %6u (%7ubp) frag %6u %s <-> length %5.2f %7u and %5.2f %7u\n",
fr->id(), fr->getLength(), frg->ident, (frgRev) ? "rev" : "fwd",
to->id(), to->getLength(), tgt->ident, (tgtRev) ? "rev" : "fwd",
joinChange, joinLen,
discChange, discLen);
joins.push_back(joinEntry(frg->ident, frFirstEnd, tgt->ident, toFlip, joinLen));
return(true);
}
// Given two fragments that share at least one edge, this will find that edge and construct a new
// edge to make it mutual.
//
// For example, if there is a best edge from aFrg 3' to bFrg 5', this will return that edge in a3,
// and also create the symmetric edge in b5.
//
static
bool
findEdges(ufNode *aFrg, BestEdgeOverlap &a5, BestEdgeOverlap &a3,
ufNode *bFrg, BestEdgeOverlap &b5, BestEdgeOverlap &b3) {
if (OG->isContained(aFrg->ident) ||
OG->isContained(bFrg->ident))
return(false);
// Grab what edges we have.
a5 = *OG->getBestEdgeOverlap(aFrg->ident, false);
a3 = *OG->getBestEdgeOverlap(aFrg->ident, true);
b5 = *OG->getBestEdgeOverlap(bFrg->ident, false);
b3 = *OG->getBestEdgeOverlap(bFrg->ident, true);
// Erase things that aren't correct
if (a5.fragId() != bFrg->ident) a5 = BestEdgeOverlap();
if (a3.fragId() != bFrg->ident) a3 = BestEdgeOverlap();
if (b5.fragId() != aFrg->ident) b5 = BestEdgeOverlap();
if (b3.fragId() != aFrg->ident) b3 = BestEdgeOverlap();
// If we have no edges left, there are no edges!
if ((b5.fragId() != aFrg->ident) && (b3.fragId() != aFrg->ident) &&
(a5.fragId() != bFrg->ident) && (a3.fragId() != bFrg->ident))
return(false);
// If we found TWO edges for any single fragment....that's madness! That means the fragment
// had best dovetail overlaps to the same other fragment off of both ends. We'll complain
// and return failure. Ideally, data like this will be cleaned up by OBT, or filtered from
// our input.
//
if (a5.fragId() == a3.fragId()) {
writeLog("findEdges()-- frag %d has multiple edges to frag %d - a5 %d/%d' a3 %d/%d'\n",
aFrg->ident, a5.fragId(),
a5.fragId(), a5.frag3p() ? 3 : 5,
a5.fragId(), a5.frag3p() ? 3 : 5);
}
if (b5.fragId() == b3.fragId()) {
writeLog("findEdges()-- frag %d has multiple edges to frag %d - b5 %d/%d' b3 %d/%d'\n",
bFrg->ident, b5.fragId(),
b5.fragId(), b5.frag3p() ? 3 : 5,
b5.fragId(), b5.frag3p() ? 3 : 5);
}
if (((a5.fragId() != 0) && (a5.fragId() == a3.fragId())) ||
((b5.fragId() != 0) && (b5.fragId() == b3.fragId()))) {
a5 = BestEdgeOverlap();
a3 = BestEdgeOverlap();
b5 = BestEdgeOverlap();
b3 = BestEdgeOverlap();
return(false);
}
// Now, populate the other edges using whatever we have. Best case is that we have two edges
// (because we're done).
assert(((a5.fragId() == bFrg->ident) +
(a3.fragId() == bFrg->ident) +
(b5.fragId() == aFrg->ident) +
(b3.fragId() == aFrg->ident)) <= 2);
if (((a5.fragId() == bFrg->ident) || (a3.fragId() == bFrg->ident)) &&
((b5.fragId() == aFrg->ident) || (b3.fragId() == aFrg->ident)))
return(true);
// Otherwise, we have exactly one edge, and the other one needs to be created.
assert(((a5.fragId() == bFrg->ident) +
(a3.fragId() == bFrg->ident) +
(b5.fragId() == aFrg->ident) +
(b3.fragId() == aFrg->ident)) == 1);
if (a5.fragId() == bFrg->ident) {
//assert(a5.fragId() == 0);
assert(a3.fragId() == 0);
assert(b5.fragId() == 0);
assert(b3.fragId() == 0);
// Edge off of A's 5' end ('false' below)...
// ...to B's 3' end (so ANTI or NORMAL -- negate the hangs)
// ...to B's 5' end (so INNIE or OUTTIE -- swap the hangs)
if (a5.frag3p())
b3.set(aFrg->ident, false, -a5.ahang(), -a5.bhang());
else
b5.set(aFrg->ident, false, a5.bhang(), a5.ahang());
} else if (a3.fragId() == bFrg->ident) {
assert(a5.fragId() == 0);
//assert(a3.fragId() == 0);
assert(b5.fragId() == 0);
assert(b3.fragId() == 0);
// Edge off of A's 3' end ('true' below)...
// ...to B's 3' end (so INNIE or OUTTIE -- swap the hangs)
// ...to B's 5' end (so ANTI or NORMAL -- negate the hangs)
if (a3.frag3p())
b3.set(aFrg->ident, true, a3.bhang(), a3.ahang());
else
b5.set(aFrg->ident, true, -a3.ahang(), -a3.bhang());
} else if (b5.fragId() == aFrg->ident) {
assert(a5.fragId() == 0);
assert(a3.fragId() == 0);
//assert(b5.fragId() == 0);
assert(b3.fragId() == 0);
if (b5.frag3p())
a3.set(bFrg->ident, false, -b5.ahang(), -b5.bhang());
else
a5.set(bFrg->ident, false, b5.bhang(), b5.ahang());
} else if (b3.fragId() == aFrg->ident) {
assert(a5.fragId() == 0);
assert(a3.fragId() == 0);
assert(b5.fragId() == 0);
//assert(b3.fragId() == 0);
if (b3.frag3p())
a3.set(bFrg->ident, true, b3.bhang(), b3.ahang());
else
a5.set(bFrg->ident, true, -b3.ahang(), -b3.bhang());
} else {
fprintf(stderr, "findEdges()-- Logically impossible!\n");
assert(0);
}
// And now we should have exactly two edges.
assert(((a5.fragId() == bFrg->ident) +
(a3.fragId() == bFrg->ident) +
(b5.fragId() == aFrg->ident) +
(b3.fragId() == aFrg->ident)) == 2);
return(true);
}
void BestOverlapGraph::scoreEdge(const OVSoverlap& olap) {
if (isOverlapBadQuality(olap))
return;
// Store edges from contained frags to help with unhappy mate
// splitting.
//
// From Eli: These are contained, but close either way. We're
// storing the non-containment edges for this fragment, plus a few
// containment edges that are "close" to being dovetails. "I think
// there are cases when a change in the alignemtn (consensus) will
// change which one is contained and screw up the order, so having
// this 10 base fudge factor helps things work out."
//
if (isContained(olap.b_iid)) {
if (((olap.dat.ovl.a_hang >= -10) && (olap.dat.ovl.b_hang <= 0)) ||
((olap.dat.ovl.a_hang >= 0) && (olap.dat.ovl.b_hang <= 10))) {
BestContainment *c = &_bestC[olap.b_iid];
if (c->olaps == NULL) {
c->olaps = new uint32 [c->olapsLen];
c->olapsLen = 0;
}
c->olaps[c->olapsLen++] = olap.a_iid;
}
return;
}
// Skip contained fragments.
if (isContained(olap.a_iid) || isContained(olap.b_iid))
return;
// Skip containment overlaps. Can this happen? Yup. How?
// The overlap could be above our allowed error.
//
if (((olap.dat.ovl.a_hang >= 0) && (olap.dat.ovl.b_hang <= 0)) ||
((olap.dat.ovl.a_hang <= 0) && (olap.dat.ovl.b_hang >= 0)))
return;
uint64 newScr = scoreOverlap(olap);
// If the score is 0, the overlap doesn't pass the scoring
// criteria at all so don't store the overlap whether or not
// it's dovetailing or containment.
if (newScr == 0)
return;
// Dove tailing overlap
bool a3p = AS_OVS_overlapAEndIs3prime(olap);
BestEdgeOverlap *best = getBestEdgeOverlap(olap.a_iid, a3p);
uint64 score = 0;
// Store the overlap if:
// 1.) The score is better than what is already in the graph
// 2.) If the scores are identical, the one with the longer length
//
// Since the order of how the overlaps are read in from the overlap
// store are by A's increasing uint32, by default, if the score and
// length are the same, the uint32 of the lower value will be kept.
if (a3p)
score = _best3score[olap.a_iid];
else
score = _best5score[olap.a_iid];
if (newScr > score) {
best->set(olap);
if (a3p)
_best3score[olap.a_iid] = newScr;
else
_best5score[olap.a_iid] = newScr;
}
}