SCORE AlignTwoMSAs(const MSA &msa1, const MSA &msa2, MSA &msaOut, PWPath &Path,
bool bLockLeft, bool bLockRight)
{
const unsigned uLengthA = msa1.GetColCount();
const unsigned uLengthB = msa2.GetColCount();
ProfPos *PA = ProfileFromMSA(msa1);
ProfPos *PB = ProfileFromMSA(msa2);
if (bLockLeft)
{
PA[0].m_scoreGapOpen = MINUS_INFINITY;
PB[0].m_scoreGapOpen = MINUS_INFINITY;
}
if (bLockRight)
{
PA[uLengthA-1].m_scoreGapClose = MINUS_INFINITY;
PB[uLengthB-1].m_scoreGapClose = MINUS_INFINITY;
}
float r = (float) uLengthA/ (float) (uLengthB + 1); // +1 to prevent div 0
if (r < 1)
r = 1/r;
SCORE Score = GlobalAlign(PA, uLengthA, PB, uLengthB, Path);
AlignTwoMSAsGivenPath(Path, msa1, msa2, msaOut);
delete[] PA;
delete[] PB;
return Score;
}
// Objective score defined as the dynamic programming score.
// Input is two alignments, which must be of the same length.
// Result is the same profile-profile score that is optimized
// by dynamic programming.
SCORE ObjScoreDP(const MSA &msa1, const MSA &msa2, SCORE MatchScore[])
{
const unsigned uColCount = msa1.GetColCount();
if (msa2.GetColCount() != uColCount)
Quit("ObjScoreDP, must be same length");
const unsigned uColCount1 = msa1.GetColCount();
const unsigned uColCount2 = msa2.GetColCount();
const ProfPos *PA = ProfileFromMSA(msa1);
const ProfPos *PB = ProfileFromMSA(msa2);
return ObjScoreDP_Profs(PA, PB, uColCount1, MatchScore);
}
// Objective score defined as the sum of profile-sequence
// scores for each sequence in the alignment. The profile
// is computed from the entire alignment, so this includes
// the score of each sequence against itself. This is to
// avoid recomputing the profile each time, so we reduce
// complexity but introduce a questionable approximation.
// The goal is to see if we can exploit the apparent
// improvement in performance of log-expectation score
// over the usual sum-of-pairs by optimizing this
// objective score in the iterative refinement stage.
SCORE ObjScorePS(const MSA &msa, SCORE MatchScore[])
{
if (g_PPScore != PPSCORE_LE)
Quit("FastScoreMSA_LASimple: LA");
const unsigned uSeqCount = msa.GetSeqCount();
const unsigned uColCount = msa.GetColCount();
const ProfPos *Prof = ProfileFromMSA(msa);
if (0 != MatchScore)
for (unsigned uColIndex = 0; uColIndex < uColCount; ++uColIndex)
MatchScore[uColIndex] = 0;
SCORE scoreTotal = 0;
for (unsigned uSeqIndex = 0; uSeqIndex < uSeqCount; ++uSeqIndex)
{
const WEIGHT weightSeq = msa.GetSeqWeight(uSeqIndex);
SCORE scoreSeq = 0;
for (unsigned uColIndex = 0; uColIndex < uColCount; ++uColIndex)
{
const ProfPos &PP = Prof[uColIndex];
if (msa.IsGap(uSeqIndex, uColIndex))
{
bool bOpen = (0 == uColIndex ||
!msa.IsGap(uSeqIndex, uColIndex - 1));
bool bClose = (uColCount - 1 == uColIndex ||
!msa.IsGap(uSeqIndex, uColIndex + 1));
if (bOpen)
scoreSeq += PP.m_scoreGapOpen;
if (bClose)
scoreSeq += PP.m_scoreGapClose;
//if (!bOpen && !bClose)
// scoreSeq += PP.m_scoreGapExtend;
}
else if (msa.IsWildcard(uSeqIndex, uColIndex))
continue;
else
{
unsigned uLetter = msa.GetLetter(uSeqIndex, uColIndex);
const SCORE scoreMatch = PP.m_AAScores[uLetter];
if (0 != MatchScore)
MatchScore[uColIndex] += weightSeq*scoreMatch;
scoreSeq += scoreMatch;
}
}
scoreTotal += weightSeq*scoreSeq;
}
delete[] Prof;
return scoreTotal;
}
static ProfPos *ProfileFromMSALocal(MSA &msa, Tree &tree)
{
const unsigned uSeqCount = msa.GetSeqCount();
for (unsigned uSeqIndex = 0; uSeqIndex < uSeqCount; ++uSeqIndex)
msa.SetSeqId(uSeqIndex, uSeqIndex);
if (TreeNeededForWeighting(g_SeqWeight2.get()))
{
TreeFromMSA(msa, tree, g_Cluster2.get(), g_Distance2.get(), g_Root1.get());
SetMuscleTree(tree);
}
return ProfileFromMSA(msa);
}
ProgNode *ProgressiveAlignE(const SeqVect &v, const Tree &GuideTree, MSA &a)
{
assert(GuideTree.IsRooted());
#if TRACE
Log("GuideTree:\n");
GuideTree.LogMe();
#endif
const unsigned uSeqCount = v.Length();
const unsigned uNodeCount = 2*uSeqCount - 1;
const unsigned uIterCount = uSeqCount - 1;
WEIGHT *Weights = new WEIGHT[uSeqCount];
CalcClustalWWeights(GuideTree, Weights);
ProgNode *ProgNodes = new ProgNode[uNodeCount];
unsigned uJoin = 0;
unsigned uTreeNodeIndex = GuideTree.FirstDepthFirstNode();
SetProgressDesc("Align node");
do
{
if (GuideTree.IsLeaf(uTreeNodeIndex))
{
if (uTreeNodeIndex >= uNodeCount)
Quit("TreeNodeIndex=%u NodeCount=%u\n", uTreeNodeIndex, uNodeCount);
ProgNode &Node = ProgNodes[uTreeNodeIndex];
unsigned uId = GuideTree.GetLeafId(uTreeNodeIndex);
if (uId >= uSeqCount)
Quit("Seq index out of range");
const Seq &s = *(v[uId]);
Node.m_MSA.FromSeq(s);
Node.m_MSA.SetSeqId(0, uId);
Node.m_uLength = Node.m_MSA.GetColCount();
Node.m_Weight = Weights[uId];
// TODO: Term gaps settable
Node.m_Prof = ProfileFromMSA(Node.m_MSA);
Node.m_EstringL = 0;
Node.m_EstringR = 0;
#if TRACE
Log("Leaf id=%u\n", uId);
Log("MSA=\n");
Node.m_MSA.LogMe();
Log("Profile (from MSA)=\n");
ListProfile(Node.m_Prof, Node.m_uLength, &Node.m_MSA);
#endif
}
else
{
Progress(uJoin, uSeqCount - 1);
++uJoin;
const unsigned uMergeNodeIndex = uTreeNodeIndex;
ProgNode &Parent = ProgNodes[uMergeNodeIndex];
const unsigned uLeft = GuideTree.GetLeft(uTreeNodeIndex);
const unsigned uRight = GuideTree.GetRight(uTreeNodeIndex);
if (g_bVerbose)
{
Log("Align: (");
LogLeafNames(GuideTree, uLeft);
Log(") (");
LogLeafNames(GuideTree, uRight);
Log(")\n");
}
ProgNode &Node1 = ProgNodes[uLeft];
ProgNode &Node2 = ProgNodes[uRight];
#if TRACE
Log("AlignTwoMSAs:\n");
#endif
AlignTwoProfs(
Node1.m_Prof, Node1.m_uLength, Node1.m_Weight,
Node2.m_Prof, Node2.m_uLength, Node2.m_Weight,
Parent.m_Path,
&Parent.m_Prof, &Parent.m_uLength);
#if TRACE_LENGTH_DELTA
{
unsigned L = Node1.m_uLength;
unsigned R = Node2.m_uLength;
unsigned P = Parent.m_Path.GetEdgeCount();
unsigned Max = L > R ? L : R;
unsigned d = P - Max;
Log("LD%u;%u;%u;%u\n", L, R, P, d);
}
#endif
PathToEstrings(Parent.m_Path, &Parent.m_EstringL, &Parent.m_EstringR);
Parent.m_Weight = Node1.m_Weight + Node2.m_Weight;
#if VALIDATE
{
#if TRACE
Log("AlignTwoMSAs:\n");
#endif
PWPath TmpPath;
AlignTwoMSAs(Node1.m_MSA, Node2.m_MSA, Parent.m_MSA, TmpPath);
ProfPos *P1 = ProfileFromMSA(Node1.m_MSA, true);
ProfPos *P2 = ProfileFromMSA(Node2.m_MSA, true);
unsigned uLength = Parent.m_MSA.GetColCount();
ProfPos *TmpProf = ProfileFromMSA(Parent.m_MSA, true);
#if TRACE
Log("Node1 MSA=\n");
Node1.m_MSA.LogMe();
Log("Node1 prof=\n");
ListProfile(Node1.m_Prof, Node1.m_MSA.GetColCount(), &Node1.m_MSA);
Log("Node1 prof (from MSA)=\n");
ListProfile(P1, Node1.m_MSA.GetColCount(), &Node1.m_MSA);
AssertProfsEq(Node1.m_Prof, Node1.m_uLength, P1, Node1.m_MSA.GetColCount());
Log("Node2 prof=\n");
ListProfile(Node2.m_Prof, Node2.m_MSA.GetColCount(), &Node2.m_MSA);
Log("Node2 MSA=\n");
Node2.m_MSA.LogMe();
Log("Node2 prof (from MSA)=\n");
ListProfile(P2, Node2.m_MSA.GetColCount(), &Node2.m_MSA);
AssertProfsEq(Node2.m_Prof, Node2.m_uLength, P2, Node2.m_MSA.GetColCount());
TmpPath.AssertEqual(Parent.m_Path);
Log("Parent MSA=\n");
Parent.m_MSA.LogMe();
Log("Parent prof=\n");
ListProfile(Parent.m_Prof, Parent.m_uLength, &Parent.m_MSA);
Log("Parent prof (from MSA)=\n");
ListProfile(TmpProf, Parent.m_MSA.GetColCount(), &Parent.m_MSA);
#endif // TRACE
AssertProfsEq(Parent.m_Prof, Parent.m_uLength,
TmpProf, Parent.m_MSA.GetColCount());
delete[] P1;
delete[] P2;
delete[] TmpProf;
}
#endif // VALIDATE
Node1.m_MSA.Clear();
Node2.m_MSA.Clear();
// Don't delete profiles, may need them for tree refinement.
//delete[] Node1.m_Prof;
//delete[] Node2.m_Prof;
//Node1.m_Prof = 0;
//Node2.m_Prof = 0;
}
uTreeNodeIndex = GuideTree.NextDepthFirstNode(uTreeNodeIndex);
}
while (NULL_NEIGHBOR != uTreeNodeIndex);
ProgressStepsDone();
if (g_bBrenner)
MakeRootMSABrenner((SeqVect &) v, GuideTree, ProgNodes, a);
else
MakeRootMSA(v, GuideTree, ProgNodes, a);
#if VALIDATE
{
unsigned uRootNodeIndex = GuideTree.GetRootNodeIndex();
const ProgNode &RootProgNode = ProgNodes[uRootNodeIndex];
AssertMSAEq(a, RootProgNode.m_MSA);
}
#endif
delete[] Weights;
return ProgNodes;
}
