/// <summary>
/// Method to merge higher-indexed item with new constructed contig.
/// Merges consumed contig with new contig. For each sequence in consumed contig,
/// compute sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score, consensus, their offsets</param>
/// <param name="consumedContig">Consumed Contig to be merged</param>
private static void MergeHigherIndexedContig(Contig newContig, ItemScore globalBest, Contig consumedContig)
{
foreach (Contig.AssembledSequence aseq in consumedContig.Sequences)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// as the higher-index item, this contig is never reversed or complemented, so:
newASeq.IsReversed = aseq.IsReversed;
newASeq.IsComplemented = aseq.IsComplemented;
// position in the new contig adjusted by alignment of the merged items.
newASeq.Position = globalBest.SecondOffset + aseq.Position;
newASeq.Sequence = SequenceWithoutTerminalGaps(aseq.Sequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"\tseq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
}
/// <summary>
/// Analyze the passed contig and store a consensus into its Consensus property.
/// </summary>
/// <param name="contig">Contig for which consensus is to be constructed</param>
private void MakeConsensus(Contig contig)
{
List <byte> positionItems = new List <byte>(), consensusSequence = new List <byte>();
// there's no simple way to pre-guess the length of the contig
long position = 0;
while (true)
{
// Initialization
positionItems.Clear();
// Add the sequences
positionItems.AddRange(from aseq in contig.Sequences
where position >= aseq.Position && position < aseq.Position + aseq.Sequence.Count
let seqPos = aseq.IsReversed ? (aseq.Sequence.Count() - 1) - (position - aseq.Position) : position - aseq.Position
select aseq.IsComplemented ? aseq.Sequence.GetComplementedSequence()[seqPos] : aseq.Sequence[seqPos]);
if (positionItems.Count == 0)
{
// This means no sequences at this position. We're done
contig.Consensus = new Sequence(Alphabets.AmbiguousAlphabetMap[_sequenceAlphabet], consensusSequence.ToArray());
return;
}
consensusSequence.Add(ConsensusResolver.GetConsensus(positionItems.ToArray()));
position++;
}
}
/// <summary>
/// Analyze the passed contig and store a consensus into its Consensus property.
/// </summary>
/// <param name="contig">Contig for which consensus is to be constructed</param>
private void MakeConsensus(Contig contig)
{
Sequence consensusSequence = new Sequence(_sequenceAlphabet);
List <ISequenceItem> positionItems = new List <ISequenceItem>();
// there's no simple way to pre-guess the length of the contig
int position = 0;
while (true)
{
// Initializations
positionItems.Clear();
foreach (Contig.AssembledSequence aseq in contig.Sequences)
{
if (position >= aseq.Position && position < aseq.Position + aseq.Sequence.Count)
{
int seqPos;
if (aseq.IsReversed)
{
seqPos = (aseq.Sequence.Count - 1) - (position - aseq.Position);
}
else
{
seqPos = position - aseq.Position;
}
if (aseq.IsComplemented)
{
positionItems.Add(aseq.Sequence.Complement[seqPos]);
}
else
{
positionItems.Add(aseq.Sequence[seqPos]);
}
}
}
if (positionItems.Count == 0)
{
// This means no sequences at this position. We're done
contig.Consensus = consensusSequence;
return;
}
else
{
consensusSequence.Add(ConsensusResolver.GetConsensus(positionItems));
}
position++;
}
}
/// <summary>
/// Analyze the passed contig and store a consensus into its Consensus property.
/// </summary>
/// <param name="contig">Contig for which consensus is to be constructed</param>
private void MakeConsensus(Contig contig)
{
List <byte> positionItems = new List <byte>();
List <byte> consensusSequence = new List <byte>();
// there's no simple way to pre-guess the length of the contig
long position = 0;
while (true)
{
// Initializations
positionItems.Clear();
foreach (Contig.AssembledSequence aseq in contig.Sequences)
{
if (position >= aseq.Position && position < aseq.Position + aseq.Sequence.Count)
{
long seqPos;
if (aseq.IsReversed)
{
seqPos = (aseq.Sequence.Count() - 1) - (position - aseq.Position);
}
else
{
seqPos = position - aseq.Position;
}
if (aseq.IsComplemented)
{
positionItems.Add(aseq.Sequence.GetComplementedSequence()[seqPos]);
}
else
{
positionItems.Add(aseq.Sequence[seqPos]);
}
}
}
if (positionItems.Count == 0)
{
// This means no sequences at this position. We're done
contig.Consensus = new Sequence(Alphabets.AmbiguousAlphabetMap[_sequenceAlphabet], consensusSequence.ToArray());
return;
}
else
{
consensusSequence.Add(ConsensusResolver.GetConsensus(positionItems.ToArray()));
}
position++;
}
}
/// <summary>
/// Analyze the passed contig and store a consensus into its Consensus property.
/// Public method to allow testing of consensus generation part.
/// Used by test automation.
/// </summary>
/// <param name="alphabet">Sequence alphabet</param>
/// <param name="contig">Contig for which consensus is to be constructed</param>
public void MakeConsensus(IAlphabet alphabet, Contig contig)
{
_sequenceAlphabet = alphabet;
if (ConsensusResolver == null)
{
ConsensusResolver = new SimpleConsensusResolver(_sequenceAlphabet);
}
else
{
ConsensusResolver.SequenceAlphabet = _sequenceAlphabet;
}
MakeConsensus(contig);
}
/// <summary>
/// Write out a set of contigs to the given file.
/// </summary>
/// <param name="formatter">Formatter</param>
/// <param name="contig">Contig to write</param>
public static void Format(this XsvContigFormatter formatter, Contig contig)
{
if (formatter == null)
{
throw new ArgumentNullException("formatter");
}
if (contig == null)
{
throw new ArgumentNullException("contig");
}
var fs = ParserFormatterExtensions<ISequenceFormatter>.GetOpenStream(formatter, true);
if (fs != null)
formatter.Write(fs, contig);
else
throw new Exception("You must open a formatter before calling Write.");
}
/// <summary>
/// Write contig to application log
/// </summary>
/// <param name="contig">contig to be dumped</param>
private static void Dump(Contig contig)
{
ApplicationLog.WriteLine("contig has {0} seqs, length {1}", contig.Sequences.Count, contig.Length);
ApplicationLog.WriteLine("consensus: {0}", contig.Consensus);
foreach (Contig.AssembledSequence aseq in contig.Sequences)
{
ApplicationLog.WriteLine(
"seq (rev = {0} comp = {1} pos = {2}) {3}",
aseq.IsReversed,
aseq.IsComplemented,
aseq.Position,
aseq.Sequence);
}
ApplicationLog.WriteLine(string.Empty);
}
/// <summary>
/// Write out a set of contigs to the given file.
/// </summary>
/// <param name="formatter">Formatter</param>
/// <param name="contig">Contig to write</param>
/// <param name="filename">Filename</param>
public static void Format(this XsvContigFormatter formatter, Contig contig, string filename)
{
if (formatter == null)
{
throw new ArgumentNullException("formatter");
}
if (contig == null)
{
throw new ArgumentNullException("contig");
}
if (string.IsNullOrWhiteSpace(filename))
{
throw new ArgumentNullException("filename");
}
using (var fs = File.Create(filename))
{
formatter.Write(fs, contig);
}
}
/// <summary>
/// Method to merge higher-indexed item with new constructed contig.
/// Merges consumed sequence with new contig. For the consumed sequence,
/// compute new sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score, consensus, their offsets</param>
/// <param name="consumedSequence">Consumed Sequence to be merged</param>
private static void MergeHigherIndexedSequence(Contig newContig, ItemScore globalBest, ISequence consumedSequence)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// as the higher-index item, this sequence is never reversed or complemented, so:
newASeq.IsReversed = false;
newASeq.IsComplemented = false;
newASeq.Position = globalBest.SecondOffset;
newASeq.Sequence = SequenceWithoutTerminalGaps(consumedSequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"seq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
/// <summary>
/// Method to merge lower-indexed item with new constructed contig
/// Merges consumed sequence with new contig. For the consumed sequence,
/// compute new sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score, consensus, their offsets</param>
/// <param name="consumedSequence">Consumed Sequence to be merged</param>
private static void MergeLowerIndexedSequence(Contig newContig, ItemScore globalBest, ISequence consumedSequence)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// lower-indexed item might be reversed or complemented.
// Retreive information from globalBest
newASeq.IsReversed = globalBest.Reversed;
newASeq.IsComplemented = globalBest.Complemented;
newASeq.Position = globalBest.FirstOffset;
newASeq.Sequence = SequenceWithoutTerminalGaps(consumedSequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"seq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
/// <summary>
/// Method to merge lower-indexed item with new constructed contig.
/// Merges consumed contig with new contig. For each sequence in consumed contig,
/// compute sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score along with offsets information</param>
/// <param name="consumedContig">Contig to be merged</param>
private static void MergeLowerIndexedContig(Contig newContig, ItemScore globalBest, Contig consumedContig)
{
foreach (Contig.AssembledSequence aseq in consumedContig.Sequences)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// lower-indexed item might be reversed or complemented.
// Construct new sequence based on setting in globalBest
// reverse of reverse, or comp of comp, equals no-op. So use xor
newASeq.IsReversed = aseq.IsReversed ^ globalBest.Reversed;
newASeq.IsComplemented = aseq.IsComplemented ^ globalBest.Complemented;
// position in the new contig is adjusted by alignment of the merged items.
// this depends on whether the contig is reverse-aligned.
if (globalBest.Reversed)
{
long rightOffset = consumedContig.Length - (aseq.Sequence.Count + aseq.Position);
newASeq.Position = globalBest.FirstOffset + rightOffset;
}
else
{
newASeq.Position = globalBest.FirstOffset + aseq.Position;
}
newASeq.Sequence = SequenceWithoutTerminalGaps(aseq.Sequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"\tseq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
}
/// <summary>
/// Formats a (sparse) contig to a character-separated value file,
/// writing the consensus first, followed by the sequence separator,
/// and each assembled sequences followed by the sequence separator.
/// The consensus has an offset of 0, while the assembled sequences have the
/// offset as present in AssembledSequence.Position.
/// </summary>
/// <param name="stream">Stream to write to, it is left open at the end.</param>
/// <param name="contig">The contig to format as a set of sparse sequences.</param>
public void Write (Stream stream, Contig contig)
{
if (stream == null)
{
throw new ArgumentNullException("stream");
}
if (contig == null)
{
throw new ArgumentNullException("contig");
}
// Write the consensus sequence out.
base.Format(stream, contig.Consensus);
// Write out the contigs.
using (StreamWriter writer = stream.OpenWrite(leaveOpen: true))
{
foreach (Contig.AssembledSequence aSeq in contig.Sequences)
{
this.Write(writer, aSeq.Sequence, (long)aSeq.Sequence.Metadata[XsvSparseParser.MetadataOffsetKey]);
}
}
}
/// <summary>
/// Method to merge higher-indexed item with new constructed contig.
/// Merges consumed contig with new contig. For each sequence in consumed contig,
/// compute sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score, consensus, their offsets</param>
/// <param name="consumedContig">Consumed Contig to be merged</param>
private static void MergeHigherIndexedContig(Contig newContig, ItemScore globalBest, Contig consumedContig)
{
foreach (Contig.AssembledSequence aseq in consumedContig.Sequences)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// as the higher-index item, this contig is never reversed or complemented, so:
newASeq.IsReversed = aseq.IsReversed;
newASeq.IsComplemented = aseq.IsComplemented;
// position in the new contig adjusted by alignment of the merged items.
newASeq.Position = globalBest.SecondOffset + aseq.Position;
newASeq.Sequence = SequenceWithoutTerminalGaps(aseq.Sequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"\tseq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
}
/// <summary>
/// Method to merge lower-indexed item with new constructed contig
/// Merges consumed sequence with new contig. For the consumed sequence,
/// compute new sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score, consensus, their offsets</param>
/// <param name="consumedSequence">Consumed Sequence to be merged</param>
private static void MergeLowerIndexedSequence(Contig newContig, ItemScore globalBest, ISequence consumedSequence)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// lower-indexed item might be reversed or complemented.
// Retreive information from globalBest
newASeq.IsReversed = globalBest.Reversed;
newASeq.IsComplemented = globalBest.Complemented;
newASeq.Position = globalBest.FirstOffset;
newASeq.Sequence = SequenceWithoutTerminalGaps(consumedSequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"seq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
/// <summary>
/// Method to merge lower-indexed item with new constructed contig.
/// Merges consumed contig with new contig. For each sequence in consumed contig,
/// compute sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score along with offsets information</param>
/// <param name="consumedContig">Contig to be merged</param>
private static void MergeLowerIndexedContig(Contig newContig, ItemScore globalBest, Contig consumedContig)
{
foreach (Contig.AssembledSequence aseq in consumedContig.Sequences)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// lower-indexed item might be reversed or complemented.
// Construct new sequence based on setting in globalBest
// reverse of reverse, or comp of comp, equals no-op. So use xor
newASeq.IsReversed = aseq.IsReversed ^ globalBest.Reversed;
newASeq.IsComplemented = aseq.IsComplemented ^ globalBest.Complemented;
// position in the new contig is adjusted by alignment of the merged items.
// this depends on whether the contig is reverse-aligned.
if (globalBest.Reversed)
{
long rightOffset = consumedContig.Length - (aseq.Sequence.Count + aseq.Position);
newASeq.Position = globalBest.FirstOffset + rightOffset;
}
else
{
newASeq.Position = globalBest.FirstOffset + aseq.Position;
}
newASeq.Sequence = SequenceWithoutTerminalGaps(aseq.Sequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"\tseq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
}
/// <summary>
/// Analyze the passed contig and store a consensus into its Consensus property.
/// Public method to allow testing of consensus generation part.
/// Used by test automation.
/// </summary>
/// <param name="alphabet">Sequence alphabet</param>
/// <param name="contig">Contig for which consensus is to be constructed</param>
public void MakeConsensus(IAlphabet alphabet, Contig contig)
{
_sequenceAlphabet = alphabet;
if (ConsensusResolver == null)
{
ConsensusResolver = new SimpleConsensusResolver(_sequenceAlphabet);
}
else
{
ConsensusResolver.SequenceAlphabet = _sequenceAlphabet;
}
MakeConsensus(contig);
}
/// <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>
/// Analyze the passed contig and store a consensus into its Consensus property.
/// </summary>
/// <param name="contig">Contig for which consensus is to be constructed</param>
private void MakeConsensus(Contig contig)
{
List<byte> positionItems = new List<byte>(), consensusSequence = new List<byte>();
// there's no simple way to pre-guess the length of the contig
long position = 0;
while (true)
{
// Initialization
positionItems.Clear();
// Add the sequences
positionItems.AddRange(from aseq in contig.Sequences
where position >= aseq.Position && position < aseq.Position + aseq.Sequence.Count
let seqPos = aseq.IsReversed ? (aseq.Sequence.Count() - 1) - (position - aseq.Position) : position - aseq.Position
select aseq.IsComplemented ? aseq.Sequence.GetComplementedSequence()[seqPos] : aseq.Sequence[seqPos]);
if (positionItems.Count == 0)
{
// This means no sequences at this position. We're done
contig.Consensus = new Sequence(Alphabets.AmbiguousAlphabetMap[_sequenceAlphabet], consensusSequence.ToArray());
return;
}
consensusSequence.Add(ConsensusResolver.GetConsensus(positionItems.ToArray()));
position++;
}
}
/// <summary>
/// Method to merge higher-indexed item with new constructed contig.
/// Merges consumed sequence with new contig. For the consumed sequence,
/// compute new sequence and offset to be added to new contig.
/// </summary>
/// <param name="newContig">New contig for merging</param>
/// <param name="globalBest">Best Score, consensus, their offsets</param>
/// <param name="consumedSequence">Consumed Sequence to be merged</param>
private static void MergeHigherIndexedSequence(Contig newContig, ItemScore globalBest, ISequence consumedSequence)
{
Contig.AssembledSequence newASeq = new Contig.AssembledSequence();
// as the higher-index item, this sequence is never reversed or complemented, so:
newASeq.IsReversed = false;
newASeq.IsComplemented = false;
newASeq.Position = globalBest.SecondOffset;
newASeq.Sequence = SequenceWithoutTerminalGaps(consumedSequence);
newContig.Sequences.Add(newASeq);
if (Trace.Want(Trace.AssemblyDetails))
{
ApplicationLog.WriteLine(
"seq (rev = {0} comp = {1} pos = {2}) {3}",
newASeq.IsReversed,
newASeq.IsComplemented,
newASeq.Position,
newASeq.Sequence);
}
}
/// <summary>
/// Write contig to application log
/// </summary>
/// <param name="contig">contig to be dumped</param>
private static void Dump(Contig contig)
{
ApplicationLog.WriteLine("contig has {0} seqs, length {1}", contig.Sequences.Count, contig.Length);
ApplicationLog.WriteLine("consensus: {0}", contig.Consensus);
foreach (Contig.AssembledSequence aseq in contig.Sequences)
{
ApplicationLog.WriteLine(
"seq (rev = {0} comp = {1} pos = {2}) {3}",
aseq.IsReversed,
aseq.IsComplemented,
aseq.Position,
aseq.Sequence);
}
ApplicationLog.WriteLine(string.Empty);
}
/// <summary>
/// Initializes a new instance of the PoolItem class.
/// constructor for contig.
/// </summary>
/// <param name="item">Pool object</param>
internal PoolItem(Contig item)
: this(item, true)
{
}
/// <summary>
/// Initializes a new instance of the PoolItem class.
/// constructor for contig.
/// </summary>
/// <param name="item">Pool object</param>
internal PoolItem(Contig item)
: this(item, true)
{
}
/// <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);
}
