/// <summary>
/// Assemble the input sequences into the largest possible contigs.
/// </summary>
/// <remarks>
/// The algorithm is:
/// 1. initialize list of contigs to empty list. List of seqs is passed as argument.
/// 2. compute pairwise overlap scores for each pair of input seqs (with reversal and
/// complementation as appropriate).
/// 3. choose best overlap score. the “merge items” (can be seqs or contigs) are the
/// items with that score. If best score is less than threshold, assembly is finished.
/// 4. merge the merge items into a single contig and remove them from their list(s)
/// 5. compute the overlap between new item and all existing items
/// 6. go to step 3
/// </remarks>
/// <param name="inputSequences">The sequences to assemble.</param>
/// <returns>Returns the OverlapDeNovoAssembly instance which contains list of
/// contigs and list of unmerged sequences which are result of this assembly.</returns>
public IDeNovoAssembly Assemble(IEnumerable<ISequence> inputSequences)
{
if (null == inputSequences)
{
throw new ArgumentNullException(Properties.Resource.ParameterNameInputSequences);
}
// numbering convention: every pool item (whether sequence or contig)
// gets a fixed number.
// sequence index = index into inputs (which we won't modify)
// contig index = nSequences + index into contigs
List<PoolItem> pool = inputSequences.Select(seq => new PoolItem(seq)).ToList();
// Initialization
int sequenceCount = pool.Count;
if (sequenceCount > 0)
{
_sequenceAlphabet = pool[0].Sequence.Alphabet;
if (ConsensusResolver == null)
{
ConsensusResolver = new SimpleConsensusResolver(_sequenceAlphabet);
}
else
{
ConsensusResolver.SequenceAlphabet = _sequenceAlphabet;
}
}
// put all the initial sequences into the pool, and generate the pair scores.
// there are no contigs in the pool yet.
// to save an iteration, we'll also find the best global score as we go.
ItemScore globalBest = new ItemScore(-1, -1, false, false, 0, 0);
int globalBestLargerIndex = -1;
int unconsumedCount = sequenceCount;
// Compute alignment scores for all combinations between input sequences
// Store these scores in the poolItem corresponding to each sequence
for (int newSeq = 0; newSeq < pool.Count; ++newSeq)
{
PoolItem newItem = pool[newSeq];
for (int oldSeq = 0; oldSeq < newSeq; ++oldSeq)
{
PoolItem oldItem = pool[oldSeq];
ItemScore score = AlignSequence(oldItem.SequenceOrConsensus, newItem.SequenceOrConsensus, oldSeq, newSeq);
newItem.Scores.Add(score);
if (score.OverlapScore > globalBest.OverlapScore)
{
globalBest = new ItemScore(score);
globalBestLargerIndex = newSeq;
}
}
}
// Merge sequence if best score is above threshold
// and add new contig to pool
if (globalBest.OverlapScore >= MergeThreshold)
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine("Merging (overlap score {0}):", globalBest.OverlapScore);
}
PoolItem mergeItem1 = pool[globalBest.OtherItem];
PoolItem mergeItem2 = pool[globalBestLargerIndex];
Contig newContig = new Contig();
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"new pool item {0} will merge old items {1} and {2}",
pool.Count,
globalBest.OtherItem,
globalBestLargerIndex);
}
MergeLowerIndexedSequence(newContig, globalBest, mergeItem1.Sequence);
MergeHigherIndexedSequence(newContig, globalBest, mergeItem2.Sequence);
MakeConsensus(newContig);
// Set ConsumedBy value and
// free memory as these sequences are no longer used
mergeItem1.ConsumedBy = pool.Count;
mergeItem2.ConsumedBy = pool.Count;
mergeItem1.FreeSequences();
mergeItem2.FreeSequences();
pool.Add(new PoolItem(newContig));
unconsumedCount--;
while (unconsumedCount > 1)
{
// Compute scores for each unconsumed sequence with new contig
int newSeq = pool.Count - 1;
PoolItem newItem = pool[newSeq];
for (int oldSeq = 0; oldSeq < pool.Count - 1; ++oldSeq)
{
PoolItem oldItem = pool[oldSeq];
if (oldItem.ConsumedBy >= 0)
{
// already consumed - just add dummy score to maintain correct indices
newItem.Scores.Add(new ItemScore());
}
else
{
ItemScore score = AlignSequence(oldItem.SequenceOrConsensus, newItem.SequenceOrConsensus, oldSeq, newSeq);
newItem.Scores.Add(score);
}
}
// find best global score in the modified pool.
globalBest = new ItemScore(-1, -1, false, false, 0, 0);
globalBestLargerIndex = -1;
for (int current = 0; current < pool.Count; ++current)
{
PoolItem curItem = pool[current];
if (curItem.ConsumedBy < 0)
{
for (int other = 0; other < current; ++other)
{
if (pool[other].ConsumedBy < 0)
{
ItemScore itemScore = curItem.Scores[other];
if (itemScore.OverlapScore > globalBest.OverlapScore)
{
globalBest = new ItemScore(itemScore); // copy the winner so far
globalBestLargerIndex = current;
}
}
}
}
}
if (globalBest.OverlapScore >= MergeThreshold)
{
// Merge sequences / contigs if above threshold
mergeItem1 = pool[globalBest.OtherItem];
mergeItem2 = pool[globalBestLargerIndex];
newContig = new Contig();
if (mergeItem1.IsContig)
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a contig (reversed = {1}, complemented = {2}, offset = {3}",
globalBest.OtherItem,
globalBest.Reversed,
globalBest.Complemented,
globalBest.FirstOffset);
}
MergeLowerIndexedContig(newContig, globalBest, mergeItem1.Contig);
}
else
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a sequence (reversed = {1}, complemented = {2}, offset = {3}",
globalBest.OtherItem,
globalBest.Reversed,
globalBest.Complemented,
globalBest.FirstOffset);
}
MergeLowerIndexedSequence(newContig, globalBest, mergeItem1.Sequence);
}
if (mergeItem2.IsContig)
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a contig (offset = {1}",
globalBestLargerIndex,
globalBest.SecondOffset);
}
MergeHigherIndexedContig(newContig, globalBest, mergeItem2.Contig);
}
else
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a sequence (offset = {1}",
globalBestLargerIndex,
globalBest.SecondOffset);
}
MergeHigherIndexedSequence(newContig, globalBest, mergeItem2.Sequence);
}
MakeConsensus(newContig);
if (Trace.Want(Trace.AssemblyDetails))
{
Dump(newContig);
}
// Set ConsumedBy value for these poolItems and
// free memory as these sequences are no longer used
mergeItem1.ConsumedBy = pool.Count;
mergeItem2.ConsumedBy = pool.Count;
mergeItem1.FreeSequences();
mergeItem2.FreeSequences();
pool.Add(new PoolItem(newContig));
unconsumedCount--;
}
else
{
// None of the alignment scores cross threshold
// No more merges possible. So end iteration.
break;
}
}
}
// no further qualifying merges, so we're done.
// populate contigs and unmergedSequences
OverlapDeNovoAssembly sequenceAssembly = new OverlapDeNovoAssembly();
foreach (PoolItem curItem in pool)
{
if (curItem.ConsumedBy < 0)
{
if (curItem.IsContig)
{
sequenceAssembly.Contigs.Add(curItem.Contig);
}
else
{
sequenceAssembly.UnmergedSequences.Add(curItem.Sequence);
}
}
}
return sequenceAssembly;
}
/// <summary>
/// Assemble the input sequences into the largest possible contigs.
/// </summary>
/// <remarks>
/// The algorithm is:
/// 1. initialize list of contigs to empty list. List of seqs is passed as argument.
/// 2. compute pairwise overlap scores for each pair of input seqs (with reversal and
/// complementation as appropriate).
/// 3. choose best overlap score. the “merge items” (can be seqs or contigs) are the
/// items with that score. If best score is less than threshold, assembly is finished.
/// 4. merge the merge items into a single contig and remove them from their list(s)
/// 5. compute the overlap between new item and all existing items
/// 6. go to step 3
/// </remarks>
/// <param name="inputSequences">The sequences to assemble.</param>
/// <returns>Returns the OverlapDeNovoAssembly instance which contains list of
/// contigs and list of unmerged sequences which are result of this assembly.</returns>
public IDeNovoAssembly Assemble(IEnumerable <ISequence> inputSequences)
{
if (null == inputSequences)
{
throw new ArgumentNullException(Properties.Resource.ParameterNameInputSequences);
}
// Initializations
if (inputSequences.Count() > 0)
{
_sequenceAlphabet = inputSequences.First().Alphabet;
if (ConsensusResolver == null)
{
ConsensusResolver = new SimpleConsensusResolver(_sequenceAlphabet);
}
else
{
ConsensusResolver.SequenceAlphabet = _sequenceAlphabet;
}
}
OverlapDeNovoAssembly sequenceAssembly = null;
// numbering convention: every pool item (whether sequence or contig)
// gets a fixed number.
// sequence index = index into inputs (which we won't modify)
// contig index = nSequences + index into contigs
List <PoolItem> pool = new List <PoolItem>();
foreach (ISequence seq in inputSequences)
{
pool.Add(new PoolItem(seq));
}
// put all the initial sequences into the pool, and generate the pair scores.
// there are no contigs in the pool yet.
// to save an iteration, we'll also find the best global score as we go.
ItemScore globalBest = new ItemScore(-1, -1, false, false, 0, 0);
int globalBestLargerIndex = -1;
int unconsumedCount = inputSequences.Count();
// Compute alignment scores for all combinations between input sequences
// Store these scores in the poolItem correspodning to each sequence
for (int newSeq = 0; newSeq < pool.Count; ++newSeq)
{
PoolItem newItem = pool[newSeq];
for (int oldSeq = 0; oldSeq < newSeq; ++oldSeq)
{
PoolItem oldItem = pool[oldSeq];
ItemScore score = AlignSequence(oldItem.SequenceOrConsensus, newItem.SequenceOrConsensus, oldSeq, newSeq);
newItem.Scores.Add(score);
if (score.OverlapScore > globalBest.OverlapScore)
{
globalBest = new ItemScore(score);
globalBestLargerIndex = newSeq;
}
}
}
// Merge sequence if best score is above threshold
// and add new contig to pool
if (globalBest.OverlapScore >= MergeThreshold)
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine("Merging (overlap score {0}):", globalBest.OverlapScore);
}
PoolItem mergeItem1 = pool[globalBest.OtherItem];
PoolItem mergeItem2 = pool[globalBestLargerIndex];
Contig newContig = new Contig();
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"new pool item {0} will merge old items {1} and {2}",
pool.Count,
globalBest.OtherItem,
globalBestLargerIndex);
}
MergeLowerIndexedSequence(newContig, globalBest, mergeItem1.Sequence);
MergeHigherIndexedSequence(newContig, globalBest, mergeItem2.Sequence);
MakeConsensus(newContig);
// Set ConsumedBy value and
// free memory as these sequences are no longer used
mergeItem1.ConsumedBy = pool.Count;
mergeItem2.ConsumedBy = pool.Count;
mergeItem1.FreeSequences();
mergeItem2.FreeSequences();
pool.Add(new PoolItem(newContig));
unconsumedCount--;
while (unconsumedCount > 1)
{
// Compute scores for each unconsumed sequence with new contig
globalBest = new ItemScore(-1, -1, false, false, 0, 0);
globalBestLargerIndex = -1;
int newSeq = pool.Count - 1;
PoolItem newItem = pool[newSeq];
for (int oldSeq = 0; oldSeq < pool.Count - 1; ++oldSeq)
{
PoolItem oldItem = pool[oldSeq];
if (oldItem.ConsumedBy >= 0)
{
// already consumed - just add dummy score to maintain correct indices
newItem.Scores.Add(new ItemScore());
}
else
{
ItemScore score = AlignSequence(oldItem.SequenceOrConsensus, newItem.SequenceOrConsensus, oldSeq, newSeq);
newItem.Scores.Add(score);
}
}
// find best global score in the modified pool.
globalBest = new ItemScore(-1, -1, false, false, 0, 0);
globalBestLargerIndex = -1;
for (int current = 0; current < pool.Count; ++current)
{
PoolItem curItem = pool[current];
if (curItem.ConsumedBy < 0)
{
for (int other = 0; other < current; ++other)
{
if (pool[other].ConsumedBy < 0)
{
ItemScore itemScore = curItem.Scores[other];
if (itemScore.OverlapScore > globalBest.OverlapScore)
{
globalBest = new ItemScore(itemScore); // copy the winner so far
globalBestLargerIndex = current;
}
}
}
}
}
if (globalBest.OverlapScore >= MergeThreshold)
{
// Merge sequences / contigs if above threshold
mergeItem1 = pool[globalBest.OtherItem];
mergeItem2 = pool[globalBestLargerIndex];
newContig = new Contig();
if (mergeItem1.IsContig)
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a contig (reversed = {1}, complemented = {2}, offset = {3}",
globalBest.OtherItem,
globalBest.Reversed,
globalBest.Complemented,
globalBest.FirstOffset);
}
MergeLowerIndexedContig(newContig, globalBest, mergeItem1.Contig);
}
else
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a sequence (reversed = {1}, complemented = {2}, offset = {3}",
globalBest.OtherItem,
globalBest.Reversed,
globalBest.Complemented,
globalBest.FirstOffset);
}
MergeLowerIndexedSequence(newContig, globalBest, mergeItem1.Sequence);
}
if (mergeItem2.IsContig)
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a contig (offset = {1}",
globalBestLargerIndex,
globalBest.SecondOffset);
}
MergeHigherIndexedContig(newContig, globalBest, mergeItem2.Contig);
}
else
{
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"item {0} is a sequence (offset = {1}",
globalBestLargerIndex,
globalBest.SecondOffset);
}
MergeHigherIndexedSequence(newContig, globalBest, mergeItem2.Sequence);
}
MakeConsensus(newContig);
if (Trace.Want(Trace.AssemblyDetails))
{
Dump(newContig);
}
// Set ConsumedBy value for these poolItems and
// free memory as these sequences are no longer used
mergeItem1.ConsumedBy = pool.Count;
mergeItem2.ConsumedBy = pool.Count;
mergeItem1.FreeSequences();
mergeItem2.FreeSequences();
pool.Add(new PoolItem(newContig));
unconsumedCount--;
}
else
{
// None of the alignment scores cross threshold
// No more merges possible. So end iteration.
break;
}
}
}
// no further qualifying merges, so we're done.
// populate contigs and unmergedSequences
sequenceAssembly = new OverlapDeNovoAssembly();
foreach (PoolItem curItem in pool)
{
if (curItem.ConsumedBy < 0)
{
if (curItem.IsContig)
{
sequenceAssembly.Contigs.Add(curItem.Contig);
}
else
{
sequenceAssembly.UnmergedSequences.Add(curItem.Sequence);
}
}
}
return(sequenceAssembly);
}
