static void LogGap(unsigned uStart, unsigned uEnd, unsigned uGapLength,
bool bNTerm, bool bCTerm)
{
Log("%16.16s ", "");
for (unsigned i = 0; i < uStart; ++i)
Log(" ");
unsigned uMyLength = 0;
for (unsigned i = uStart; i <= uEnd; ++i)
{
bool bGap1 = g_ptrMSA1.get()->IsGap(g_uSeqIndex1.get(), i);
bool bGap2 = g_ptrMSA2.get()->IsGap(g_uSeqIndex2.get(), i);
if (!bGap1 && !bGap2)
Quit("Error -- neither gapping");
if (bGap1 && bGap2)
Log(".");
else
{
++uMyLength;
Log("-");
}
}
SCORE s = GapPenalty(uGapLength, bNTerm || bCTerm);
Log(" L=%d N%d C%d s=%.3g", uGapLength, bNTerm, bCTerm, s);
Log("\n");
if (uMyLength != uGapLength)
Quit("Lengths differ");
}
static const char *LocalScoreToStr(SCORE s)
{
static TLS<char[16]> str;
if (s < -100000)
return " *";
sprintf(str.get(), "%6.1f", s);
return str.get();
}
static inline unsigned TriangleSubscript(unsigned uIndex1, unsigned uIndex2)
{
#if DEBUG
if (uIndex1 >= g_uLeafCount.get() || uIndex2 >= g_uLeafCount.get())
Quit("TriangleSubscript(%u,%u) %u", uIndex1, uIndex2, g_uLeafCount.get());
#endif
unsigned v;
if (uIndex1 >= uIndex2)
v = uIndex2 + (uIndex1*(uIndex1 - 1))/2;
else
v = uIndex1 + (uIndex2*(uIndex2 - 1))/2;
assert(v < (g_uLeafCount.get()*(g_uLeafCount.get() - 1))/2);
return v;
}
void SaveCurrentAlignment()
{
extern TLS<MSA *>ptrBestMSA;
static TLS<bool> bCalled(false);
if (bCalled.get())
{
fprintf(stderr,
"\nRecursive call to SaveCurrentAlignment, giving up attempt to save.\n");
exit(EXIT_FatalError);
}
if (0 == ptrBestMSA.get())
{
fprintf(stderr, "\nAlignment not completed, cannot save.\n");
Log("Alignment not completed, cannot save.\n");
exit(EXIT_FatalError);
}
if (0 == pstrOutputFileName.get())
{
fprintf(stderr, "\nOutput file name not specified, cannot save.\n");
exit(EXIT_FatalError);
}
fprintf(stderr, "\nSaving current alignment ...\n");
TextFile fileOut(pstrOutputFileName.get(), true);
ptrBestMSA.get()->ToFASTAFile(fileOut);
fprintf(stderr, "Current alignment saved to \"%s\".\n", pstrOutputFileName.get());
Log("Current alignment saved to \"%s\".\n", pstrOutputFileName.get());
}
static bool _indication_proc(Instance* cimple_inst, void* client_data)
{
TRACE;
// This function is called by the CIMPLE Indication_Handler<> in order to
// deliver a single indication.
Adapter* adapter = (Adapter*)client_data;
// If this is the final call, just return.
if (cimple_inst == 0)
return false;
// Convert CIMPLE instance to CMPI instance:
CMPIInstance* cmpi_inst = 0;
CMPIrc rc = make_cmpi_instance(adapter->broker,
cimple_inst, _INDICATIONS_NAMESPACE, 0, cmpi_inst);
// Deliver the indication (we cannot do anything about failures).
if (rc == CMPI_RC_OK)
{
// Grab the CMPI context from thread-specific-data.
const CMPIContext* context = (const CMPIContext*)_context_tls.get();
// Deliver the indication:
CBDeliverIndication(
adapter->broker, context, _INDICATIONS_NAMESPACE, cmpi_inst);
}
// Keep them coming!
return true;
}
SCORE NWDASimple(const ProfPos *PA, unsigned uLengthA, const ProfPos *PB,
unsigned uLengthB, PWPath &Path)
{
assert(uLengthB > 0 && uLengthA > 0);
const unsigned uPrefixCountA = uLengthA + 1;
const unsigned uPrefixCountB = uLengthB + 1;
// Allocate DP matrices
const size_t LM = uPrefixCountA*uPrefixCountB;
SCORE *DPL_ = new SCORE[LM];
SCORE *DPM_ = new SCORE[LM];
SCORE *DPD_ = new SCORE[LM];
SCORE *DPE_ = new SCORE[LM];
SCORE *DPI_ = new SCORE[LM];
SCORE *DPJ_ = new SCORE[LM];
char *TBM_ = new char[LM];
char *TBD_ = new char[LM];
char *TBE_ = new char[LM];
char *TBI_ = new char[LM];
char *TBJ_ = new char[LM];
memset(TBM_, '?', LM);
memset(TBD_, '?', LM);
memset(TBE_, '?', LM);
memset(TBI_, '?', LM);
memset(TBJ_, '?', LM);
DPM(0, 0) = 0;
DPD(0, 0) = MINUS_INFINITY;
DPE(0, 0) = MINUS_INFINITY;
DPI(0, 0) = MINUS_INFINITY;
DPJ(0, 0) = MINUS_INFINITY;
DPM(1, 0) = MINUS_INFINITY;
DPD(1, 0) = PA[0].m_scoreGapOpen;
DPE(1, 0) = PA[0].m_scoreGapOpen2;
TBD(1, 0) = 'D';
TBE(1, 0) = 'E';
DPI(1, 0) = MINUS_INFINITY;
DPJ(1, 0) = MINUS_INFINITY;
DPM(0, 1) = MINUS_INFINITY;
DPD(0, 1) = MINUS_INFINITY;
DPE(0, 1) = MINUS_INFINITY;
DPI(0, 1) = PB[0].m_scoreGapOpen;
DPJ(0, 1) = PB[0].m_scoreGapOpen2;
TBI(0, 1) = 'I';
TBJ(0, 1) = 'J';
// Empty prefix of B is special case
for (unsigned uPrefixLengthA = 2; uPrefixLengthA < uPrefixCountA; ++uPrefixLengthA)
{
DPM(uPrefixLengthA, 0) = MINUS_INFINITY;
DPD(uPrefixLengthA, 0) = DPD(uPrefixLengthA - 1, 0) + g_scoreGapExtend.get();
DPE(uPrefixLengthA, 0) = DPE(uPrefixLengthA - 1, 0) + g_scoreGapExtend2.get();
TBD(uPrefixLengthA, 0) = 'D';
TBE(uPrefixLengthA, 0) = 'E';
DPI(uPrefixLengthA, 0) = MINUS_INFINITY;
DPJ(uPrefixLengthA, 0) = MINUS_INFINITY;
}
// Empty prefix of A is special case
for (unsigned uPrefixLengthB = 2; uPrefixLengthB < uPrefixCountB; ++uPrefixLengthB)
{
DPM(0, uPrefixLengthB) = MINUS_INFINITY;
DPD(0, uPrefixLengthB) = MINUS_INFINITY;
DPE(0, uPrefixLengthB) = MINUS_INFINITY;
DPI(0, uPrefixLengthB) = DPI(0, uPrefixLengthB - 1) + g_scoreGapExtend.get();
DPJ(0, uPrefixLengthB) = DPJ(0, uPrefixLengthB - 1) + g_scoreGapExtend2.get();
TBI(0, uPrefixLengthB) = 'I';
TBJ(0, uPrefixLengthB) = 'J';
}
// Special case to agree with NWFast, no D-I transitions so...
DPD(uLengthA, 0) = MINUS_INFINITY;
DPE(uLengthA, 0) = MINUS_INFINITY;
// DPI(0, uLengthB) = MINUS_INFINITY;
// DPJ(0, uLengthB) = MINUS_INFINITY;
// ============
// Main DP loop
// ============
SCORE scoreGapCloseB = MINUS_INFINITY;
SCORE scoreGapClose2B = MINUS_INFINITY;
for (unsigned uPrefixLengthB = 1; uPrefixLengthB < uPrefixCountB; ++uPrefixLengthB)
{
const ProfPos &PPB = PB[uPrefixLengthB - 1];
SCORE scoreGapCloseA = MINUS_INFINITY;
SCORE scoreGapClose2A = MINUS_INFINITY;
for (unsigned uPrefixLengthA = 1; uPrefixLengthA < uPrefixCountA; ++uPrefixLengthA)
{
const ProfPos &PPA = PA[uPrefixLengthA - 1];
{
// Match M=LetterA+LetterB
SCORE scoreLL = ScoreProfPos2(PPA, PPB);
DPL(uPrefixLengthA, uPrefixLengthB) = scoreLL;
SCORE scoreMM = DPM(uPrefixLengthA-1, uPrefixLengthB-1);
SCORE scoreDM = DPD(uPrefixLengthA-1, uPrefixLengthB-1) + scoreGapCloseA;
SCORE scoreEM = DPE(uPrefixLengthA-1, uPrefixLengthB-1) + scoreGapClose2A;
SCORE scoreIM = DPI(uPrefixLengthA-1, uPrefixLengthB-1) + scoreGapCloseB;
SCORE scoreJM = DPJ(uPrefixLengthA-1, uPrefixLengthB-1) + scoreGapClose2B;
SCORE scoreBest;
if (scoreMM >= scoreDM && scoreMM >= scoreEM && scoreMM >= scoreIM && scoreMM >= scoreJM)
{
scoreBest = scoreMM;
TBM(uPrefixLengthA, uPrefixLengthB) = 'M';
}
else if (scoreDM >= scoreMM && scoreDM >= scoreEM && scoreDM >= scoreIM && scoreDM >= scoreJM)
{
scoreBest = scoreDM;
TBM(uPrefixLengthA, uPrefixLengthB) = 'D';
}
else if (scoreEM >= scoreMM && scoreEM >= scoreDM && scoreEM >= scoreIM && scoreEM >= scoreJM)
{
scoreBest = scoreEM;
TBM(uPrefixLengthA, uPrefixLengthB) = 'E';
}
else if (scoreIM >= scoreMM && scoreIM >= scoreDM && scoreIM >= scoreEM && scoreIM >= scoreJM)
{
scoreBest = scoreIM;
TBM(uPrefixLengthA, uPrefixLengthB) = 'I';
}
else
{
assert(scoreJM >= scoreMM && scoreJM >= scoreDM && scoreJM >= scoreEM && scoreJM >= scoreIM);
scoreBest = scoreJM;
TBM(uPrefixLengthA, uPrefixLengthB) = 'J';
}
DPM(uPrefixLengthA, uPrefixLengthB) = scoreBest + scoreLL;
}
{
// Delete D=LetterA+GapB
SCORE scoreMD = DPM(uPrefixLengthA-1, uPrefixLengthB) +
PA[uPrefixLengthA-1].m_scoreGapOpen;
SCORE scoreDD = DPD(uPrefixLengthA-1, uPrefixLengthB) + g_scoreGapExtend.get();
SCORE scoreBest;
if (scoreMD >= scoreDD)
{
scoreBest = scoreMD;
TBD(uPrefixLengthA, uPrefixLengthB) = 'M';
}
else
{
assert(scoreDD >= scoreMD);
scoreBest = scoreDD;
TBD(uPrefixLengthA, uPrefixLengthB) = 'D';
}
DPD(uPrefixLengthA, uPrefixLengthB) = scoreBest;
}
{
// Delete E=LetterA+GapB
SCORE scoreME = DPM(uPrefixLengthA-1, uPrefixLengthB) +
PA[uPrefixLengthA-1].m_scoreGapOpen2;
SCORE scoreEE = DPE(uPrefixLengthA-1, uPrefixLengthB) + g_scoreGapExtend2.get();
SCORE scoreBest;
if (scoreME >= scoreEE)
{
scoreBest = scoreME;
TBE(uPrefixLengthA, uPrefixLengthB) = 'M';
}
else
{
assert(scoreEE >= scoreME);
scoreBest = scoreEE;
TBE(uPrefixLengthA, uPrefixLengthB) = 'E';
}
DPE(uPrefixLengthA, uPrefixLengthB) = scoreBest;
}
// Insert I=GapA+LetterB
{
SCORE scoreMI = DPM(uPrefixLengthA, uPrefixLengthB-1) +
PB[uPrefixLengthB - 1].m_scoreGapOpen;
SCORE scoreII = DPI(uPrefixLengthA, uPrefixLengthB-1) + g_scoreGapExtend.get();
SCORE scoreBest;
if (scoreMI >= scoreII)
{
scoreBest = scoreMI;
TBI(uPrefixLengthA, uPrefixLengthB) = 'M';
}
else
{
assert(scoreII > scoreMI);
scoreBest = scoreII;
TBI(uPrefixLengthA, uPrefixLengthB) = 'I';
}
DPI(uPrefixLengthA, uPrefixLengthB) = scoreBest;
}
// Insert J=GapA+LetterB
{
SCORE scoreMJ = DPM(uPrefixLengthA, uPrefixLengthB-1) +
PB[uPrefixLengthB - 1].m_scoreGapOpen2;
SCORE scoreJJ = DPJ(uPrefixLengthA, uPrefixLengthB-1) + g_scoreGapExtend2.get();
SCORE scoreBest;
if (scoreMJ >= scoreJJ)
{
scoreBest = scoreMJ;
TBJ(uPrefixLengthA, uPrefixLengthB) = 'M';
}
else
{
assert(scoreJJ > scoreMJ);
scoreBest = scoreJJ;
TBJ(uPrefixLengthA, uPrefixLengthB) = 'J';
}
DPJ(uPrefixLengthA, uPrefixLengthB) = scoreBest;
}
scoreGapCloseA = PPA.m_scoreGapClose;
scoreGapClose2A = PPA.m_scoreGapClose2;
}
scoreGapCloseB = PPB.m_scoreGapClose;
scoreGapClose2B = PPB.m_scoreGapClose2;
}
#if TRACE
Log("\n");
Log("DA Simple DPL:\n");
ListDP(DPL_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple DPM:\n");
ListDP(DPM_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple DPD:\n");
ListDP(DPD_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple DPE:\n");
ListDP(DPE_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple DPI:\n");
ListDP(DPI_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple DPJ:\n");
ListDP(DPJ_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple TBM:\n");
ListTB(TBM_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple TBD:\n");
ListTB(TBD_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple TBE:\n");
ListTB(TBE_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple TBI:\n");
ListTB(TBI_, PA, PB, uPrefixCountA, uPrefixCountB);
Log("\n");
Log("DA Simple TBJ:\n");
ListTB(TBJ_, PA, PB, uPrefixCountA, uPrefixCountB);
#endif
// Trace-back
// ==========
Path.Clear();
// Find last edge
SCORE M = DPM(uLengthA, uLengthB);
SCORE D = DPD(uLengthA, uLengthB) + PA[uLengthA-1].m_scoreGapClose;
SCORE E = DPE(uLengthA, uLengthB) + PA[uLengthA-1].m_scoreGapClose2;
SCORE I = DPI(uLengthA, uLengthB) + PB[uLengthB-1].m_scoreGapClose;
SCORE J = DPJ(uLengthA, uLengthB) + PB[uLengthB-1].m_scoreGapClose2;
char cEdgeType = '?';
SCORE BestScore = M;
cEdgeType = 'M';
if (D > BestScore)
{
cEdgeType = 'D';
BestScore = D;
}
if (E > BestScore)
{
cEdgeType = 'E';
BestScore = E;
}
if (I > BestScore)
{
cEdgeType = 'I';
BestScore = I;
}
if (J > BestScore)
{
cEdgeType = 'J';
BestScore = J;
}
#if TRACE
Log("DA Simple: MAB=%.4g DAB=%.4g EAB=%.4g IAB=%.4g JAB=%.4g best=%c\n",
M, D, E, I, J, cEdgeType);
#endif
unsigned PLA = uLengthA;
unsigned PLB = uLengthB;
for (;;)
{
PWEdge Edge;
Edge.cType = XlatEdgeType(cEdgeType);
Edge.uPrefixLengthA = PLA;
Edge.uPrefixLengthB = PLB;
#if TRACE
Log("Prepend %c%d.%d\n", Edge.cType, PLA, PLB);
#endif
Path.PrependEdge(Edge);
switch (cEdgeType)
{
case 'M':
assert(PLA > 0);
assert(PLB > 0);
cEdgeType = TBM(PLA, PLB);
--PLA;
--PLB;
break;
case 'D':
assert(PLA > 0);
cEdgeType = TBD(PLA, PLB);
--PLA;
break;
case 'E':
assert(PLA > 0);
cEdgeType = TBE(PLA, PLB);
--PLA;
break;
case 'I':
assert(PLB > 0);
cEdgeType = TBI(PLA, PLB);
--PLB;
break;
case 'J':
assert(PLB > 0);
cEdgeType = TBJ(PLA, PLB);
--PLB;
break;
default:
Quit("Invalid edge %c", cEdgeType);
}
if (0 == PLA && 0 == PLB)
break;
}
Path.Validate();
// SCORE Score = TraceBack(PA, uLengthA, PB, uLengthB, DPM_, DPD_, DPI_, Path);
#if TRACE
SCORE scorePath = FastScorePath2(PA, uLengthA, PB, uLengthB, Path);
Path.LogMe();
Log("Score = %s Path = %s\n", LocalScoreToStr(BestScore), LocalScoreToStr(scorePath));
#endif
if (g_bKeepSimpleDP.get())
{
g_DPM.get() = DPM_;
g_DPD.get() = DPD_;
g_DPE.get() = DPE_;
g_DPI.get() = DPI_;
g_DPJ.get() = DPJ_;
g_TBM.get() = TBM_;
g_TBD.get() = TBD_;
g_TBE.get() = TBE_;
g_TBI.get() = TBI_;
g_TBJ.get() = TBJ_;
}
else
{
delete[] DPM_;
delete[] DPD_;
delete[] DPE_;
delete[] DPI_;
delete[] DPJ_;
delete[] TBM_;
delete[] TBD_;
delete[] TBE_;
delete[] TBI_;
delete[] TBJ_;
}
return BestScore;
}
static void ListState()
{
Log("Dist matrix\n");
Log(" ");
for (unsigned i = 0; i < g_uLeafCount.get(); ++i)
{
if (uInsane == g_uNodeIndex.get()[i])
continue;
Log(" %5u", g_uNodeIndex.get()[i]);
}
Log("\n");
for (unsigned i = 0; i < g_uLeafCount.get(); ++i)
{
if (uInsane == g_uNodeIndex.get()[i])
continue;
Log("%5u ", g_uNodeIndex.get()[i]);
for (unsigned j = 0; j < g_uLeafCount.get(); ++j)
{
if (uInsane == g_uNodeIndex.get()[j])
continue;
if (i == j)
Log(" ");
else
{
unsigned v = TriangleSubscript(i, j);
Log("%5.2g ", g_Dist.get()[v]);
}
}
Log("\n");
}
Log("\n");
Log(" i Node NrNb Dist\n");
Log("----- ----- ----- --------\n");
for (unsigned i = 0; i < g_uLeafCount.get(); ++i)
{
if (uInsane == g_uNodeIndex.get()[i])
continue;
Log("%5u %5u %5u %8.3f\n",
i,
g_uNodeIndex.get()[i],
g_uNearestNeighbor.get()[i],
g_MinDist.get()[i]);
}
Log("\n");
Log(" Node L R Height LLength RLength\n");
Log("----- ----- ----- ------ ------- -------\n");
for (unsigned i = 0; i <= g_uInternalNodeIndex.get(); ++i)
Log("%5u %5u %5u %6.2g %6.2g %6.2g\n",
i,
g_uLeft.get()[i],
g_uRight.get()[i],
g_Height.get()[i],
g_LeftLength.get()[i],
g_RightLength.get()[i]);
}
void UPGMA2(const DistCalc &DC, Tree &tree, LINKAGE Linkage)
{
g_uLeafCount.get() = DC.GetCount();
g_uTriangleSize.get() = (g_uLeafCount.get()*(g_uLeafCount.get() - 1))/2;
g_uInternalNodeCount.get() = g_uLeafCount.get() - 1;
g_Dist.get() = new dist_t[g_uTriangleSize.get()];
g_uNodeIndex.get() = new unsigned[g_uLeafCount.get()];
g_uNearestNeighbor.get() = new unsigned[g_uLeafCount.get()];
g_MinDist.get() = new dist_t[g_uLeafCount.get()];
unsigned *Ids = new unsigned [g_uLeafCount.get()];
char **Names = new char *[g_uLeafCount.get()];
g_uLeft.get() = new unsigned[g_uInternalNodeCount.get()];
g_uRight.get() = new unsigned[g_uInternalNodeCount.get()];
g_Height.get() = new dist_t[g_uInternalNodeCount.get()];
g_LeftLength.get() = new dist_t[g_uInternalNodeCount.get()];
g_RightLength.get() = new dist_t[g_uInternalNodeCount.get()];
for (unsigned i = 0; i < g_uLeafCount.get(); ++i)
{
g_MinDist.get()[i] = BIG_DIST;
g_uNodeIndex.get()[i] = i;
g_uNearestNeighbor.get()[i] = uInsane;
Ids[i] = DC.GetId(i);
Names[i] = strsave(DC.GetName(i));
}
for (unsigned i = 0; i < g_uInternalNodeCount.get(); ++i)
{
g_uLeft.get()[i] = uInsane;
g_uRight.get()[i] = uInsane;
g_LeftLength.get()[i] = BIG_DIST;
g_RightLength.get()[i] = BIG_DIST;
g_Height.get()[i] = BIG_DIST;
}
// Compute initial NxN triangular distance matrix.
// Store minimum distance for each full (not triangular) row.
// Loop from 1, not 0, because "row" is 0, 1 ... i-1,
// so nothing to do when i=0.
for (unsigned i = 1; i < g_uLeafCount.get(); ++i)
{
dist_t *Row = g_Dist.get() + TriangleSubscript(i, 0);
DC.CalcDistRange(i, Row);
for (unsigned j = 0; j < i; ++j)
{
const dist_t d = Row[j];
if (d < g_MinDist.get()[i])
{
g_MinDist.get()[i] = d;
g_uNearestNeighbor.get()[i] = j;
}
if (d < g_MinDist.get()[j])
{
g_MinDist.get()[j] = d;
g_uNearestNeighbor.get()[j] = i;
}
}
}
#if TRACE
Log("Initial state:\n");
ListState();
#endif
for (g_uInternalNodeIndex.get() = 0; g_uInternalNodeIndex.get() < g_uLeafCount.get() - 1;
++g_uInternalNodeIndex.get())
{
#if TRACE
Log("\n");
Log("Internal node index %5u\n", g_uInternalNodeIndex.get());
Log("-------------------------\n");
#endif
// Find nearest neighbors
unsigned Lmin = uInsane;
unsigned Rmin = uInsane;
dist_t dtMinDist = BIG_DIST;
for (unsigned j = 0; j < g_uLeafCount.get(); ++j)
{
if (uInsane == g_uNodeIndex.get()[j])
continue;
dist_t d = g_MinDist.get()[j];
if (d < dtMinDist)
{
dtMinDist = d;
Lmin = j;
Rmin = g_uNearestNeighbor.get()[j];
assert(uInsane != Rmin);
assert(uInsane != g_uNodeIndex.get()[Rmin]);
}
}
assert(Lmin != uInsane);
assert(Rmin != uInsane);
assert(dtMinDist != BIG_DIST);
#if TRACE
Log("Nearest neighbors Lmin %u[=%u] Rmin %u[=%u] dist %.3g\n",
Lmin,
g_uNodeIndex.get()[Lmin],
Rmin,
g_uNodeIndex.get()[Rmin],
dtMinDist);
#endif
// Compute distances to new node
// New node overwrites row currently assigned to Lmin
dist_t dtNewMinDist = BIG_DIST;
unsigned uNewNearestNeighbor = uInsane;
for (unsigned j = 0; j < g_uLeafCount.get(); ++j)
{
if (j == Lmin || j == Rmin)
continue;
if (uInsane == g_uNodeIndex.get()[j])
continue;
const unsigned vL = TriangleSubscript(Lmin, j);
const unsigned vR = TriangleSubscript(Rmin, j);
const dist_t dL = g_Dist.get()[vL];
const dist_t dR = g_Dist.get()[vR];
dist_t dtNewDist;
switch (Linkage)
{
case LINKAGE_Avg:
dtNewDist = AVG(dL, dR);
break;
case LINKAGE_Min:
dtNewDist = MIN(dL, dR);
break;
case LINKAGE_Max:
dtNewDist = MAX(dL, dR);
break;
case LINKAGE_Biased:
dtNewDist = g_dSUEFF.get()*AVG(dL, dR) + (1 - g_dSUEFF.get())*MIN(dL, dR);
break;
default:
Quit("UPGMA2: Invalid LINKAGE_%u", Linkage);
}
// Nasty special case.
// If nearest neighbor of j is Lmin or Rmin, then make the new
// node (which overwrites the row currently occupied by Lmin)
// the nearest neighbor. This situation can occur when there are
// equal distances in the matrix. If we don't make this fix,
// the nearest neighbor pointer for j would become invalid.
// (We don't need to test for == Lmin, because in that case
// the net change needed is zero due to the change in row
// numbering).
if (g_uNearestNeighbor.get()[j] == Rmin)
g_uNearestNeighbor.get()[j] = Lmin;
#if TRACE
Log("New dist to %u = (%u/%.3g + %u/%.3g)/2 = %.3g\n",
j, Lmin, dL, Rmin, dR, dtNewDist);
#endif
g_Dist.get()[vL] = dtNewDist;
if (dtNewDist < dtNewMinDist)
{
dtNewMinDist = dtNewDist;
uNewNearestNeighbor = j;
}
}
assert(g_uInternalNodeIndex.get() < g_uLeafCount.get() - 1 || BIG_DIST != dtNewMinDist);
assert(g_uInternalNodeIndex.get() < g_uLeafCount.get() - 1 || uInsane != uNewNearestNeighbor);
const unsigned v = TriangleSubscript(Lmin, Rmin);
const dist_t dLR = g_Dist.get()[v];
const dist_t dHeightNew = dLR/2;
const unsigned uLeft = g_uNodeIndex.get()[Lmin];
const unsigned uRight = g_uNodeIndex.get()[Rmin];
const dist_t HeightLeft =
uLeft < g_uLeafCount.get() ? 0 : g_Height.get()[uLeft - g_uLeafCount.get()];
const dist_t HeightRight =
uRight < g_uLeafCount.get() ? 0 : g_Height.get()[uRight - g_uLeafCount.get()];
g_uLeft.get()[g_uInternalNodeIndex.get()] = uLeft;
g_uRight.get()[g_uInternalNodeIndex.get()] = uRight;
g_LeftLength.get()[g_uInternalNodeIndex.get()] = dHeightNew - HeightLeft;
g_RightLength.get()[g_uInternalNodeIndex.get()] = dHeightNew - HeightRight;
g_Height.get()[g_uInternalNodeIndex.get()] = dHeightNew;
// Row for left child overwritten by row for new node
g_uNodeIndex.get()[Lmin] = g_uLeafCount.get() + g_uInternalNodeIndex.get();
g_uNearestNeighbor.get()[Lmin] = uNewNearestNeighbor;
g_MinDist.get()[Lmin] = dtNewMinDist;
// Delete row for right child
g_uNodeIndex.get()[Rmin] = uInsane;
#if TRACE
Log("\nInternalNodeIndex=%u Lmin=%u Rmin=%u\n",
g_uInternalNodeIndex.get(), Lmin, Rmin);
ListState();
#endif
}
unsigned uRoot = g_uLeafCount.get() - 2;
tree.Create(g_uLeafCount.get(), uRoot, g_uLeft.get(), g_uRight.get(), g_LeftLength.get(), g_RightLength.get(),
Ids, Names);
#if TRACE
tree.LogMe();
#endif
delete[] g_Dist.get();
delete[] g_uNodeIndex.get();
delete[] g_uNearestNeighbor.get();
delete[] g_MinDist.get();
delete[] g_Height.get();
delete[] g_uLeft.get();
delete[] g_uRight.get();
delete[] g_LeftLength.get();
delete[] g_RightLength.get();
for (unsigned i = 0; i < g_uLeafCount.get(); ++i)
free(Names[i]);
delete[] Names;
delete[] Ids;
}
// WARNING: Sequences MUST be stripped of gaps and upper case!
void DistKmer20_3(const SeqVect &v, DistFunc &DF)
{
const unsigned uSeqCount = v.Length();
DF.SetCount(uSeqCount);
if (0 == uSeqCount)
return;
for (unsigned uSeq1 = 0; uSeq1 < uSeqCount; ++uSeq1)
{
DF.SetDist(uSeq1, uSeq1, 0);
for (unsigned uSeq2 = 0; uSeq2 < uSeq1; ++uSeq2)
DF.SetDist(uSeq1, uSeq2, 0);
}
const unsigned uTripleArrayBytes = TRIPLE_COUNT*sizeof(TripleCount);
TripleCounts.get() = (TripleCount *) malloc(uTripleArrayBytes);
if (0 == TripleCounts.get())
Quit("Not enough memory (TripleCounts)");
memset(TripleCounts.get(), 0, uTripleArrayBytes);
for (unsigned uWord = 0; uWord < TRIPLE_COUNT; ++uWord)
{
TripleCount &tc = *(TripleCounts.get() + uWord);
const unsigned uBytes = uSeqCount*sizeof(short);
tc.m_Counts = (unsigned short *) malloc(uBytes);
memset(tc.m_Counts, 0, uBytes);
}
for (unsigned uSeqIndex = 0; uSeqIndex < uSeqCount; ++uSeqIndex)
{
Seq &s = *(v[uSeqIndex]);
const unsigned uSeqLength = s.Length();
for (unsigned uPos = 0; uPos < uSeqLength - 2; ++uPos)
{
const unsigned uLetter1 = CharToLetterEx(s[uPos]);
if (uLetter1 >= 20)
continue;
const unsigned uLetter2 = CharToLetterEx(s[uPos+1]);
if (uLetter2 >= 20)
continue;
const unsigned uLetter3 = CharToLetterEx(s[uPos+2]);
if (uLetter3 >= 20)
continue;
const unsigned uWord = uLetter1 + uLetter2*20 + uLetter3*20*20;
assert(uWord < TRIPLE_COUNT);
TripleCount &tc = *(TripleCounts.get() + uWord);
const unsigned uOldCount = tc.m_Counts[uSeqIndex];
if (0 == uOldCount)
++(tc.m_uSeqCount);
++(tc.m_Counts[uSeqIndex]);
}
}
#if TRACE
{
Log("TripleCounts\n");
unsigned uGrandTotal = 0;
for (unsigned uWord = 0; uWord < TRIPLE_COUNT; ++uWord)
{
const TripleCount &tc = *(TripleCounts.get() + uWord);
if (0 == tc.m_uSeqCount)
continue;
const unsigned uLetter3 = uWord/(20*20);
const unsigned uLetter2 = (uWord - uLetter3*20*20)/20;
const unsigned uLetter1 = uWord%20;
Log("Word %6u %c%c%c %6u",
uWord,
LetterToCharAmino(uLetter1),
LetterToCharAmino(uLetter2),
LetterToCharAmino(uLetter3),
tc.m_uSeqCount);
unsigned uSeqCountWithThisWord = 0;
for (unsigned uSeqIndex = 0; uSeqIndex < uSeqCount; ++uSeqIndex)
{
const unsigned uCount = tc.m_Counts[uSeqIndex];
if (uCount > 0)
{
++uSeqCountWithThisWord;
Log(" %u=%u", uSeqIndex, uCount);
uGrandTotal += uCount;
}
}
if (uSeqCountWithThisWord != tc.m_uSeqCount)
Log(" *** SQ ERROR *** %u %u", tc.m_uSeqCount, uSeqCountWithThisWord);
Log("\n");
}
unsigned uTotalBySeqLength = 0;
for (unsigned uSeqIndex = 0; uSeqIndex < uSeqCount; ++uSeqIndex)
{
Seq &s = *(v[uSeqIndex]);
const unsigned uSeqLength = s.Length();
uTotalBySeqLength += uSeqLength - 2;
}
if (uGrandTotal != uTotalBySeqLength)
Log("*** TOTALS DISAGREE *** %u %u\n", uGrandTotal, uTotalBySeqLength);
}
#endif
const unsigned uSeqListBytes = uSeqCount*sizeof(unsigned);
unsigned short *SeqList = (unsigned short *) malloc(uSeqListBytes);
for (unsigned uWord = 0; uWord < TRIPLE_COUNT; ++uWord)
{
const TripleCount &tc = *(TripleCounts.get() + uWord);
if (0 == tc.m_uSeqCount)
continue;
unsigned uSeqCountFound = 0;
memset(SeqList, 0, uSeqListBytes);
for (unsigned uSeqIndex = 0; uSeqIndex < uSeqCount; ++uSeqIndex)
{
if (tc.m_Counts[uSeqIndex] > 0)
{
SeqList[uSeqCountFound] = uSeqIndex;
++uSeqCountFound;
if (uSeqCountFound == tc.m_uSeqCount)
break;
}
}
assert(uSeqCountFound == tc.m_uSeqCount);
for (unsigned uSeq1 = 0; uSeq1 < uSeqCountFound; ++uSeq1)
{
const unsigned uSeqIndex1 = SeqList[uSeq1];
const unsigned uCount1 = tc.m_Counts[uSeqIndex1];
for (unsigned uSeq2 = 0; uSeq2 < uSeq1; ++uSeq2)
{
const unsigned uSeqIndex2 = SeqList[uSeq2];
const unsigned uCount2 = tc.m_Counts[uSeqIndex2];
const unsigned uMinCount = uCount1 < uCount2 ? uCount1 : uCount2;
const double d = DF.GetDist(uSeqIndex1, uSeqIndex2);
DF.SetDist(uSeqIndex1, uSeqIndex2, (float) (d + uMinCount));
}
}
}
delete[] SeqList;
free(TripleCounts.get());
unsigned uDone = 0;
const unsigned uTotal = (uSeqCount*(uSeqCount - 1))/2;
for (unsigned uSeq1 = 0; uSeq1 < uSeqCount; ++uSeq1)
{
DF.SetDist(uSeq1, uSeq1, 0.0);
const Seq &s1 = *(v[uSeq1]);
const unsigned uLength1 = s1.Length();
for (unsigned uSeq2 = 0; uSeq2 < uSeq1; ++uSeq2)
{
const Seq &s2 = *(v[uSeq2]);
const unsigned uLength2 = s2.Length();
unsigned uMinLength = uLength1 < uLength2 ? uLength1 : uLength2;
if (uMinLength < 3)
{
DF.SetDist(uSeq1, uSeq2, 1.0);
continue;
}
const double dTripleCount = DF.GetDist(uSeq1, uSeq2);
if (dTripleCount == 0)
{
DF.SetDist(uSeq1, uSeq2, 1.0);
continue;
}
double dNormalizedTripletScore = dTripleCount/(uMinLength - 2);
//double dEstimatedPairwiseIdentity = exp(0.3912*log(dNormalizedTripletScore));
//if (dEstimatedPairwiseIdentity > 1)
// dEstimatedPairwiseIdentity = 1;
// DF.SetDist(uSeq1, uSeq2, (float) (1.0 - dEstimatedPairwiseIdentity));
DF.SetDist(uSeq1, uSeq2, (float) dNormalizedTripletScore);
#if TRACE
{
Log("%s - %s Triplet count = %g Lengths %u, %u Estimated pwid = %g\n",
s1.GetName(), s2.GetName(), dTripleCount, uLength1, uLength2,
dEstimatedPairwiseIdentity);
}
#endif
if (uDone%1000 == 0)
Progress(uDone, uTotal);
}
}
ProgressStepsDone();
}
static SCORE ScoreSeqPair(const MSA &msa1, unsigned uSeqIndex1,
const MSA &msa2, unsigned uSeqIndex2, SCORE *ptrLetters, SCORE *ptrGaps)
{
g_ptrMSA1.get() = &msa1;
g_ptrMSA2.get() = &msa2;
g_uSeqIndex1.get() = uSeqIndex1;
g_uSeqIndex2.get() = uSeqIndex2;
const unsigned uColCount = msa1.GetColCount();
const unsigned uColCount2 = msa2.GetColCount();
if (uColCount != uColCount2)
Quit("ScoreSeqPair, different lengths");
#if TRACE
Log("ScoreSeqPair\n");
Log("%16.16s ", msa1.GetSeqName(uSeqIndex1));
for (unsigned i = 0; i < uColCount; ++i)
Log("%c", msa1.GetChar(uSeqIndex1, i));
Log("\n");
Log("%16.16s ", msa2.GetSeqName(uSeqIndex2));
for (unsigned i = 0; i < uColCount; ++i)
Log("%c", msa1.GetChar(uSeqIndex2, i));
Log("\n");
#endif
SCORE scoreTotal = 0;
// Substitution scores
unsigned uFirstLetter1 = uInsane;
unsigned uFirstLetter2 = uInsane;
unsigned uLastLetter1 = uInsane;
unsigned uLastLetter2 = uInsane;
for (unsigned uColIndex = 0; uColIndex < uColCount; ++uColIndex)
{
bool bGap1 = msa1.IsGap(uSeqIndex1, uColIndex);
bool bGap2 = msa2.IsGap(uSeqIndex2, uColIndex);
bool bWildcard1 = msa1.IsWildcard(uSeqIndex1, uColIndex);
bool bWildcard2 = msa2.IsWildcard(uSeqIndex2, uColIndex);
if (!bGap1)
{
if (uInsane == uFirstLetter1)
uFirstLetter1 = uColIndex;
uLastLetter1 = uColIndex;
}
if (!bGap2)
{
if (uInsane == uFirstLetter2)
uFirstLetter2 = uColIndex;
uLastLetter2 = uColIndex;
}
if (bGap1 || bGap2 || bWildcard1 || bWildcard2)
continue;
unsigned uLetter1 = msa1.GetLetter(uSeqIndex1, uColIndex);
unsigned uLetter2 = msa2.GetLetter(uSeqIndex2, uColIndex);
SCORE scoreMatch = (*g_ptrScoreMatrix.get())[uLetter1][uLetter2];
scoreTotal += scoreMatch;
#if TRACE
Log("%c <-> %c = %7.1f %10.1f\n",
msa1.GetChar(uSeqIndex1, uColIndex),
msa2.GetChar(uSeqIndex2, uColIndex),
scoreMatch,
scoreTotal);
#endif
}
*ptrLetters = scoreTotal;
// Gap penalties
unsigned uGapLength = uInsane;
unsigned uGapStartCol = uInsane;
bool bGapping1 = false;
bool bGapping2 = false;
for (unsigned uColIndex = 0; uColIndex < uColCount; ++uColIndex)
{
bool bGap1 = msa1.IsGap(uSeqIndex1, uColIndex);
bool bGap2 = msa2.IsGap(uSeqIndex2, uColIndex);
if (bGap1 && bGap2)
continue;
if (bGapping1)
{
if (bGap1)
++uGapLength;
else
{
bGapping1 = false;
bool bNTerm = (uFirstLetter2 == uGapStartCol);
bool bCTerm = (uLastLetter2 + 1 == uColIndex);
SCORE scoreGap = GapPenalty(uGapLength, bNTerm || bCTerm);
scoreTotal += scoreGap;
#if TRACE
LogGap(uGapStartCol, uColIndex - 1, uGapLength, bNTerm, bCTerm);
Log("GAP %7.1f %10.1f\n",
scoreGap,
scoreTotal);
#endif
}
continue;
}
else
{
if (bGap1)
{
uGapStartCol = uColIndex;
bGapping1 = true;
uGapLength = 1;
continue;
}
}
if (bGapping2)
{
if (bGap2)
++uGapLength;
else
{
bGapping2 = false;
bool bNTerm = (uFirstLetter1 == uGapStartCol);
bool bCTerm = (uLastLetter1 + 1 == uColIndex);
SCORE scoreGap = GapPenalty(uGapLength, bNTerm || bCTerm);
scoreTotal += scoreGap;
#if TRACE
LogGap(uGapStartCol, uColIndex - 1, uGapLength, bNTerm, bCTerm);
Log("GAP %7.1f %10.1f\n",
scoreGap,
scoreTotal);
#endif
}
}
else
{
if (bGap2)
{
uGapStartCol = uColIndex;
bGapping2 = true;
uGapLength = 1;
}
}
}
if (bGapping1 || bGapping2)
{
SCORE scoreGap = GapPenalty(uGapLength, true);
scoreTotal += scoreGap;
#if TRACE
LogGap(uGapStartCol, uColCount - 1, uGapLength, false, true);
Log("GAP %7.1f %10.1f\n",
scoreGap,
scoreTotal);
#endif
}
*ptrGaps = scoreTotal - *ptrLetters;
return scoreTotal;
}
void SetCurrentAlignment(MSA &msa)
{
extern TLS<MSA *>ptrBestMSA;
ptrBestMSA.get() = &msa;
}
void SetOutputFileName(const char *out)
{
pstrOutputFileName.get() = out;
}
