/// <summary>
/// Generate IProfileAlignment from a set of aligned sequences
/// </summary>
/// <param name="sequences">aligned sequences</param>
public static IProfileAlignment GenerateProfileAlignment(ICollection <ISequence> sequences)
{
IProfiles profileMatrix = Profiles.GenerateProfiles(sequences);
IProfileAlignment profileAlignment = new ProfileAlignment();
profileAlignment.NumberOfSequences = sequences.Count;
profileAlignment.ProfilesMatrix = profileMatrix;
return(profileAlignment);
}
/// <summary>
/// Generate a profileAlignment from one single sequence
/// The set of sequence items of the seq should be the same as
/// 'static ItemSet' of the IProfiles.
/// </summary>
/// <param name="seq">an input sequence</param>
/// <param name="weight">sequence weight</param>
public static IProfileAlignment GenerateProfileAlignment(ISequence seq, float weight)
{
IProfiles profileMatrix = Profiles.GenerateProfiles(seq, weight);
IProfileAlignment profileAlignment = new ProfileAlignment();
profileAlignment.NumberOfSequences = 1;
profileAlignment.ProfilesMatrix = profileMatrix;
return(profileAlignment);
}
/// <summary>
/// Construct a node with assigned ID
/// </summary>
/// <param name="id">zero-based node ID</param>
public BinaryGuideTreeNode(int id)
{
LeftChildren = null;
RightChildren = null;
Parent = null;
ProfileAlignment = new ProfileAlignment();
_id = id;
_sequenceID = id;
_eString = new List <int>();
NeedReAlignment = true;
}
/// <summary>
/// Combine two profileAlignments into one if they are aligned already
/// </summary>
/// <param name="profileAlignmentA">first profile alignment</param>
/// <param name="profileAlignmentB">second profile alignment</param>
public static IProfileAlignment GenerateProfileAlignment(IProfileAlignment profileAlignmentA, IProfileAlignment profileAlignmentB)
{
IProfiles profileMatrix = Profiles.GenerateProfiles(
profileAlignmentA.ProfilesMatrix, profileAlignmentB.ProfilesMatrix,
profileAlignmentA.NumberOfSequences, profileAlignmentB.NumberOfSequences);
IProfileAlignment profileAlignment = new ProfileAlignment();
profileAlignment.NumberOfSequences = profileAlignmentA.NumberOfSequences + profileAlignmentB.NumberOfSequences;
profileAlignment.ProfilesMatrix = profileMatrix;
return(profileAlignment);
}
/// <summary>
/// The profiles of two subsets is extracted from the current multiple alignment.
/// Columns containing no residues, i.e. indels only, are discarded.
///
/// This method is used in alignment refinement, when the guide tree is cut into two,
/// the sequences (leaf nodes) are separated into two subsets. This method generates
/// two profileAlignments for the two subtrees by extracting profiles of the two subsets
/// of sequences.
/// </summary>
/// <param name="alignedSequences">a set of aligned sequences</param>
/// <param name="sequenceIndicesA">the subset sequence indices of subtree A</param>
/// <param name="sequenceIndicesB">the subset sequence indices of subtree B</param>
/// <param name="allIndelPositions">the list of all-indel positions that have been removed when constructing</param>
public static IProfileAlignment[] ProfileExtraction(List <ISequence> alignedSequences,
List <int> sequenceIndicesA, List <int> sequenceIndicesB,
out List <int>[] allIndelPositions)
{
allIndelPositions = new List <int> [2];
IProfiles profileA = Profiles.GenerateProfiles(alignedSequences, sequenceIndicesA, out allIndelPositions[0]);
IProfiles profileB = Profiles.GenerateProfiles(alignedSequences, sequenceIndicesB, out allIndelPositions[1]);
IProfileAlignment profileAlignmentA = new ProfileAlignment();
IProfileAlignment profileAlignmentB = new ProfileAlignment();
profileAlignmentA.ProfilesMatrix = profileA;
profileAlignmentB.ProfilesMatrix = profileB;
profileAlignmentA.NumberOfSequences = sequenceIndicesA.Count;
profileAlignmentB.NumberOfSequences = sequenceIndicesB.Count;
return(new IProfileAlignment[2] {
profileAlignmentA, profileAlignmentB
});
}
/// <summary>
/// Combine two profileAlignments with alignment operation array from dynamic programming.
/// The dynamic programming algorithm returns two arrays containing the alignment operations
/// on the two profiles. This method applies the operation information in the two arrays to
/// the two original profiles, and combine them into a new aligned profile, and put into the
/// newly generated profileAlignment.
/// </summary>
/// <param name="profileAlignmentA">first profile alignment</param>
/// <param name="profileAlignmentB">second profile alignment</param>
/// <param name="aAligned">aligned interger array generated by dynamic programming</param>
/// <param name="bAligned">aligned interger array generated by dynamic programming</param>
/// <param name="gapCode">the gap integer code defined in dynamic programming class</param>
public static IProfileAlignment GenerateProfileAlignment(
IProfileAlignment profileAlignmentA,
IProfileAlignment profileAlignmentB,
int[] aAligned,
int[] bAligned,
int gapCode)
{
IProfiles profileMatrix = Profiles.GenerateProfiles(
profileAlignmentA.ProfilesMatrix, profileAlignmentB.ProfilesMatrix,
profileAlignmentA.NumberOfSequences, profileAlignmentB.NumberOfSequences,
aAligned, bAligned, gapCode);
IProfileAlignment profileAlignment = new ProfileAlignment();
profileAlignment.NumberOfSequences = profileAlignmentA.NumberOfSequences +
profileAlignmentB.NumberOfSequences;
profileAlignment.ProfilesMatrix = profileMatrix;
return(profileAlignment);
}
/// <summary>
/// Main pregressive alignment algorithm aligns a set of sequences guided by
/// a binary tree.
/// </summary>
/// <param name="sequences">input sequences</param>
/// <param name="tree">a binary guide tree</param>
public void Align(IList <ISequence> sequences, BinaryGuideTree tree)
{
SequenceWeighting sequenceWeighting = null;
if (PAMSAMMultipleSequenceAligner.UseWeights)
{
sequenceWeighting = new SequenceWeighting(tree);
/*
* for (int i = 0; i < sequenceWeighting.Weights.Length; ++i)
* {
* sequenceWeighting.Weights[i] = 1;
* }
*/
}
if (sequences.Count == 0)
{
throw new ArgumentException("Empty set of sequences");
}
IAlphabet alphabet = sequences[0].Alphabet;
Parallel.For(1, sequences.Count, PAMSAMMultipleSequenceAligner.parallelOption, i =>
{
if (sequences[i].Alphabet != alphabet)
{
throw new ArgumentException("Inconsistent sequence alphabet");
}
});
if (PAMSAMMultipleSequenceAligner.UseWeights)
{
// Generate profile for leaf nodes
Parallel.For(0, sequences.Count, PAMSAMMultipleSequenceAligner.parallelOption, i =>
{
tree.Nodes[i].ProfileAlignment = ProfileAlignment.GenerateProfileAlignment(sequences[i], sequenceWeighting.Weights[i]);
tree.Nodes[i].Weight = sequenceWeighting.Weights[i];
});
}
else
{
// Generate profile for leaf nodes
Parallel.For(0, sequences.Count, PAMSAMMultipleSequenceAligner.parallelOption, i =>
{
tree.Nodes[i].ProfileAlignment = ProfileAlignment.GenerateProfileAlignment(sequences[i]);
});
}
// Iterate internal nodes;
// as defined in the tree, the last node is the root
for (int i = sequences.Count; i < tree.Nodes.Count; ++i)
{
if (tree.Nodes[i].NeedReAlignment)
{
// pull out its children
_nodeA = tree.Nodes[i].LeftChildren;
_nodeB = tree.Nodes[i].RightChildren;
if (PAMSAMMultipleSequenceAligner.UseWeights)
{
_profileAligner.Weights = new float[2];
_profileAligner.Weights[0] = _nodeA.Weight;
_profileAligner.Weights[1] = _nodeB.Weight;
tree.Nodes[i].Weight = _nodeA.Weight + _nodeB.Weight;
}
// align two profiles
ProfileAlignment result = null;
if (_nodeA.ProfileAlignment.NumberOfSequences < _nodeB.ProfileAlignment.NumberOfSequences)
{
result = (ProfileAlignment)_profileAligner.Align(
_nodeA.ProfileAlignment, _nodeB.ProfileAlignment);
// assign aligned profiles to the current internal node
tree.Nodes[i].ProfileAlignment = result;
// generate eString for the children nodes
_nodeA.EString = _profileAligner.GenerateEString(_profileAligner.AlignedA);
_nodeB.EString = _profileAligner.GenerateEString(_profileAligner.AlignedB);
}
else
{
result = (ProfileAlignment)_profileAligner.Align(
_nodeB.ProfileAlignment, _nodeA.ProfileAlignment);
// assign aligned profiles to the current internal node
tree.Nodes[i].ProfileAlignment = result;
// generate eString for the children nodes
_nodeA.EString = _profileAligner.GenerateEString(_profileAligner.AlignedB);
_nodeB.EString = _profileAligner.GenerateEString(_profileAligner.AlignedA);
}
// children node profiles can be deleted
_nodeA.ProfileAlignment.Clear();
_nodeB.ProfileAlignment.Clear();
}
}
// Convert original unaligned sequences to aligned ones by applying alignment paths in eStrings
try
{
_alignedSequences = new List <ISequence>(sequences.Count);
}
catch (OutOfMemoryException ex)
{
throw new Exception("Out of memory", ex.InnerException);
}
for (int i = 0; i < sequences.Count; ++i)
{
_alignedSequences.Add(null);
}
Parallel.For(0, sequences.Count, PAMSAMMultipleSequenceAligner.parallelOption, i =>
{
ISequence seq = sequences[i];
BinaryGuideTreeNode node;
node = tree.Nodes[i];
while (!node.IsRoot)
{
seq = _profileAligner.GenerateSequenceFromEString(node.EString, seq);
node = node.Parent;
}
_alignedSequences[i] = seq;
});
}
/// <summary>
/// Performs Stage 1, 2, and 3 as described in class description.
/// </summary>
/// <param name="sequences">input unaligned sequences</param>
public IList <MBF.Algorithms.Alignment.ISequenceAlignment> Align(IList <ISequence> sequences)
{
// Initializations
if (sequences.Count > 0)
{
if (ConsensusResolver == null)
{
ConsensusResolver = new SimpleConsensusResolver(sequences[0].Alphabet);
}
else
{
ConsensusResolver.SequenceAlphabet = sequences[0].Alphabet;
}
}
// Get ProfileAligner ready
IProfileAligner profileAligner = null;
switch (_profileAlignerName)
{
case (ProfileAlignerNames.NeedlemanWunschProfileAligner):
if (_degreeOfParallelism == 1)
{
profileAligner = new NeedlemanWunschProfileAlignerSerial(
SimilarityMatrix, _profileProfileFunctionName, GapOpenCost, GapExtensionCost, _numberOfPartitions);
}
else
{
profileAligner = new NeedlemanWunschProfileAlignerParallel(
SimilarityMatrix, _profileProfileFunctionName, GapOpenCost, GapExtensionCost, _numberOfPartitions);
}
break;
case (ProfileAlignerNames.SmithWatermanProfileAligner):
if (_degreeOfParallelism == 1)
{
profileAligner = new SmithWatermanProfileAlignerSerial(
SimilarityMatrix, _profileProfileFunctionName, GapOpenCost, GapExtensionCost, _numberOfPartitions);
}
else
{
profileAligner = new SmithWatermanProfileAlignerParallel(
SimilarityMatrix, _profileProfileFunctionName, GapOpenCost, GapExtensionCost, _numberOfPartitions);
}
break;
default:
throw new ArgumentException("Invalid profile aligner name");
}
_alignedSequences = new List <ISequence>(sequences.Count);
float currentScore = 0;
// STAGE 1
Performance.Snapshot("Stage 1");
// Generate DistanceMatrix
KmerDistanceMatrixGenerator kmerDistanceMatrixGenerator =
new KmerDistanceMatrixGenerator(sequences, _kmerLength, _moleculeType, _distanceFunctionName);
// Hierarchical clustering
IHierarchicalClustering hierarcicalClustering =
new HierarchicalClusteringParallel
(kmerDistanceMatrixGenerator.DistanceMatrix, _hierarchicalClusteringMethodName);
// Generate Guide Tree
BinaryGuideTree binaryGuideTree =
new BinaryGuideTree(hierarcicalClustering);
// Progressive Alignment
IProgressiveAligner progressiveAlignerA = new ProgressiveAligner(profileAligner);
progressiveAlignerA.Align(sequences, binaryGuideTree);
currentScore = MsaUtils.MultipleAlignmentScoreFunction(progressiveAlignerA.AlignedSequences, SimilarityMatrix, GapOpenCost, GapExtensionCost);
if (currentScore > _alignmentScoreA)
{
_alignmentScoreA = currentScore;
_alignedSequencesA = progressiveAlignerA.AlignedSequences;
}
if (_alignmentScoreA > _alignmentScore)
{
_alignmentScore = _alignmentScoreA;
_alignedSequences = _alignedSequencesA;
}
if (PAMSAMMultipleSequenceAligner.FasterVersion)
{
_alignedSequencesB = _alignedSequencesA;
_alignedSequencesC = _alignedSequencesA;
_alignmentScoreB = _alignmentScoreA;
_alignmentScoreC = _alignmentScoreA;
}
else
{
BinaryGuideTree binaryGuideTreeB = null;
IHierarchicalClustering hierarcicalClusteringB = null;
KimuraDistanceMatrixGenerator kimuraDistanceMatrixGenerator = new KimuraDistanceMatrixGenerator();
if (PAMSAMMultipleSequenceAligner.UseStageB)
{
// STAGE 2
Performance.Snapshot("Stage 2");
// Generate DistanceMatrix from Multiple Sequence Alignment
int iterateTime = 0;
while (true)
{
++iterateTime;
kimuraDistanceMatrixGenerator.GenerateDistanceMatrix(_alignedSequences);
// Hierarchical clustering
hierarcicalClusteringB = new HierarchicalClusteringParallel
(kimuraDistanceMatrixGenerator.DistanceMatrix, _hierarchicalClusteringMethodName);
// Generate Guide Tree
binaryGuideTreeB = new BinaryGuideTree(hierarcicalClusteringB);
BinaryGuideTree.CompareTwoTrees(binaryGuideTreeB, binaryGuideTree);
binaryGuideTree = binaryGuideTreeB;
// Progressive Alignment
IProgressiveAligner progressiveAlignerB = new ProgressiveAligner(profileAligner);
progressiveAlignerB.Align(sequences, binaryGuideTreeB);
currentScore = MsaUtils.MultipleAlignmentScoreFunction(progressiveAlignerB.AlignedSequences, SimilarityMatrix, GapOpenCost, GapExtensionCost);
if (currentScore > _alignmentScoreB)
{
_alignmentScoreB = currentScore;
_alignedSequencesB = progressiveAlignerB.AlignedSequences;
break;
}
else
{
break;
}
}
if (_alignmentScoreB > _alignmentScore)
{
_alignmentScore = _alignmentScoreB;
_alignedSequences = _alignedSequencesB;
}
}
else
{
binaryGuideTreeB = binaryGuideTree;
}
// STAGE 3
Performance.Snapshot("Stage 3");
// refinement
//int maxRefineMentTime = sequences.Count * 2 - 2;
int maxRefineMentTime = 1;
if (sequences.Count == 2)
{
maxRefineMentTime = 0;
}
int refinementTime = 0;
_alignedSequencesC = new List <ISequence>(sequences.Count);
for (int i = 0; i < sequences.Count; ++i)
{
_alignedSequencesC.Add(new Sequence(_alphabet, _alignedSequences[i].ToString()));
}
List <int>[] leafNodeIndices = null;
List <int>[] allIndelPositions = null;
IProfileAlignment[] separatedProfileAlignments = null;
List <int>[] eStrings = null;
while (refinementTime < maxRefineMentTime)
{
++refinementTime;
Performance.Snapshot("Refinement iter " + refinementTime.ToString());
bool needRefinement = false;
for (int edgeIndex = 0; edgeIndex < binaryGuideTreeB.NumberOfEdges; ++edgeIndex)
{
leafNodeIndices = binaryGuideTreeB.SeparateSequencesByCuttingTree(edgeIndex);
allIndelPositions = new List <int> [2];
separatedProfileAlignments = ProfileAlignment.ProfileExtraction(_alignedSequencesC, leafNodeIndices[0], leafNodeIndices[1], out allIndelPositions);
eStrings = new List <int> [2];
if (separatedProfileAlignments[0].NumberOfSequences < separatedProfileAlignments[1].NumberOfSequences)
{
profileAligner.Align(separatedProfileAlignments[0], separatedProfileAlignments[1]);
eStrings[0] = profileAligner.GenerateEString(profileAligner.AlignedA);
eStrings[1] = profileAligner.GenerateEString(profileAligner.AlignedB);
}
else
{
profileAligner.Align(separatedProfileAlignments[1], separatedProfileAlignments[0]);
eStrings[0] = profileAligner.GenerateEString(profileAligner.AlignedB);
eStrings[1] = profileAligner.GenerateEString(profileAligner.AlignedA);
}
for (int set = 0; set < 2; ++set)
{
Parallel.ForEach(leafNodeIndices[set], PAMSAMMultipleSequenceAligner.parallelOption, i =>
{
Sequence seq = new Sequence(_alphabet, "");
seq.IsReadOnly = false;
int indexAllIndel = 0;
for (int j = 0; j < _alignedSequencesC[i].Count; ++j)
{
if (indexAllIndel < allIndelPositions[set].Count && j == allIndelPositions[set][indexAllIndel])
{
++indexAllIndel;
}
else
{
seq.Add(_alignedSequencesC[i][j]);
}
}
seq = profileAligner.GenerateSequenceFromEString(eStrings[set], seq);
seq.IsReadOnly = true;
_alignedSequencesC[i] = seq;
});
}
currentScore = MsaUtils.MultipleAlignmentScoreFunction(_alignedSequencesC, SimilarityMatrix, GapOpenCost, GapExtensionCost);
if (currentScore > _alignmentScoreC)
{
_alignmentScoreC = currentScore;
needRefinement = true;
// recreate the tree
kimuraDistanceMatrixGenerator.GenerateDistanceMatrix(_alignedSequencesC);
hierarcicalClusteringB = new HierarchicalClusteringParallel
(kimuraDistanceMatrixGenerator.DistanceMatrix, _hierarchicalClusteringMethodName);
binaryGuideTreeB = new BinaryGuideTree(hierarcicalClusteringB);
break;
}
}
if (!needRefinement)
{
refinementTime = maxRefineMentTime;
break;
}
}
if (_alignmentScoreC > _alignmentScore)
{
_alignmentScore = _alignmentScoreC;
_alignedSequences = _alignedSequencesC;
}
Performance.Snapshot("Stop Stage 3");
}
//just for the purpose of integrating PW and MSA with the same output
IList <MBF.Algorithms.Alignment.ISequenceAlignment> results = new List <MBF.Algorithms.Alignment.ISequenceAlignment>();
return(results);
}