public static void TryMergeAllClusters(SetOfClusters clusterSet, int maxNumDisagreements)
{
var clusters = clusterSet.Clusters;
for (var i = 0; i < clusters.Count(); i++)
{
for (var j = i + 1; j < clusters.Count(); j++)
{
var clusterA = clusters[i];
var clusterB = clusters[j];
var canBeMerged = TestCanBeMerged(clusterA, clusterB, maxNumDisagreements);
if (!canBeMerged)
{
continue;
}
clusterSet.RemoveCluster(clusterA.Name);
clusterSet.RemoveCluster(clusterB.Name);
var mergedCluster = MergeClusters(clusterA, clusterB);
clusterSet.AddCluster(mergedCluster);
}
}
}
private SetOfClusters CreateDefaultSetOfClusters()
{
var clusteringParams = new ClusteringParameters()
{
MaxNumberDisagreements = 0, MinNumberAgreements = 0
};
var setOfClusters = new SetOfClusters(clusteringParams);
Assert.Equal(0, setOfClusters.Clusters.Count());
return(setOfClusters);
}
public void GetPhasingProbabilities()
{
var variantSites = new List <VariantSite>
{
new VariantSite(1),
new VariantSite(2),
new VariantSite(45)
};
var clusters = new SetOfClusters(new ClusteringParameters());
// There should be a PhasingResult for each variant in variantSites
var phasingProbabilities = VariantPhasingResult.GetPhasingProbabilities(variantSites, clusters);
Assert.Equal(variantSites.Count, phasingProbabilities.Count);
Assert.Equal(variantSites.Select(x => x), phasingProbabilities.Keys.ToList());
}
public static Cluster MergeAllBestCandidates(SetOfClusters clusters, int maxNumDisagreements,
List <Cluster> bestCandidates, VeadGroup testVeadGroup)
{
var numCandidates = bestCandidates.Count;
var bestcluster = bestCandidates[0];
for (var i = 0; i < numCandidates; i++)
{
for (var j = i + 1; j < numCandidates; j++)
{
// First test if the clusters can be merged. If they can, merge them.
// If they can't, return the better of the two.
var clusterA = bestCandidates[i];
var clusterB = bestCandidates[j];
var canBeMerged = TestCanBeMerged(clusterA, clusterB, maxNumDisagreements, testVeadGroup);
if (canBeMerged)
{
clusters.RemoveCluster(clusterA.Name);
clusters.RemoveCluster(clusterB.Name);
var mergedCluster = MergeClusters(clusterA, clusterB);
clusters.AddCluster(mergedCluster);
bestcluster = mergedCluster;
}
else
{
if (clusterB.NumVeads > clusterA.NumVeads)
{
bestcluster = clusterB;
}
//else, leave it as bestcluster = cA;
}
}
}
return(bestcluster);
}
private void MeetPloidyConstraints(SetOfClusters clusters)
{
while (clusters.NumClusters > _options.ClusterConstraint)
{
Logger.WriteToLog("Num clusters: " + clusters.Clusters.Length);
Logger.WriteToLog("Num cluster constraint " + _options.ClusterConstraint + " is violated. Pruning clusters...");
int maxAllowedToRemove = (clusters.NumClusters - _options.ClusterConstraint);
int numWorstClusters = clusters.RemoveWorstClusters(maxAllowedToRemove);
if (numWorstClusters <= maxAllowedToRemove)
{
Logger.WriteToLog(numWorstClusters + " clusters pruned.");
}
else
{
Logger.WriteToLog(numWorstClusters + " low ranked clusters found. This is not resolveable with our cluster constraints.");
// then we had a tie situation, and we do not know how to proceed.
//if this is not resolvable, we are going to fail our cluster constraint.
break;
}
}
Logger.WriteToLog("Clusters finalized: " + clusters.Clusters.Length);
}
public void MergeAllBestCandidates()
{
var setOfClusters = new SetOfClusters(new ClusteringParameters());
var veadgroups = ClusterTestHelpers.GetSampleVeadGroups(prefix: "Original");
var cluster = new Cluster("test", veadgroups);
setOfClusters.AddCluster(cluster);
var veadgroups2 = ClusterTestHelpers.GetSampleVeadGroups(prefix: "Second");
var cluster2 = new Cluster("test2", veadgroups2);
setOfClusters.AddCluster(cluster2);
Assert.Equal(2, setOfClusters.NumClusters);
var veadgroups3 = ClusterTestHelpers.GetSampleVeadGroups(1, 1, prefix: "Tester");
// If can be merged, we should have 1 cluster.
var bestCluster = ClusterMerger.MergeAllBestCandidates(setOfClusters, 0, new List <Cluster> {
cluster, cluster2
}, veadgroups3.First());
Assert.Equal(1, setOfClusters.NumClusters);
}
public static Dictionary <VariantSite, VariantPhasingResult> GetPhasingProbabilities(List <VariantSite> variantSites, SetOfClusters clusters)
{
var results = new Dictionary <VariantSite, VariantPhasingResult>();
foreach (var variantSite in variantSites)
{
results.Add(variantSite, GetPhasingProbabilitiesForVariant(variantSites.ToList(), clusters, variantSite));
}
//TODO debug Clusters: (VariantGroup, Chr, Pos, Ref, Alt); Neighbors (VariantGroup, Chr, Pos, Ref, Alt, ProbOfAGivenB, ProbOfB, ProbOfAAndB)
return(results);
}
private static VariantPhasingResult GetPhasingProbabilitiesForVariant(List <VariantSite> variantGroup, SetOfClusters clusters, VariantSite variantSiteA)
{
var otherVariants = variantGroup.Where(vs => vs != variantSiteA).ToList();
var phasingResult = new VariantPhasingResult(variantSiteA, otherVariants, clusters.NumClusters);
var relativeWeights = clusters.GetRelativeWeights();
//how many clusters have B in them
//how many clusters have A and B in them?
foreach (var cluster in clusters.Clusters)
{
var supportDict = cluster.GetVeadCountsInCluster(variantGroup);
var weight = relativeWeights[cluster.Name];
foreach (var variantSiteB in otherVariants)
{
if (supportDict[variantSiteB] <= 0)
{
continue;
}
phasingResult.AddSupportForB(variantSiteB, weight);
if (supportDict[variantSiteA] > 0)
{
phasingResult.AddSupportForAandB(variantSiteB, weight);
}
}
}
return(phasingResult);
}
public SetOfClusters ClusterVeadGroups(List <VeadGroup> veadGroups, string nbhdID)
{
try
{
// Make the meatiest clusters first.
veadGroups.Sort();
var clusters = new SetOfClusters(_options);
if (veadGroups.Count == 0)
{
Logger.WriteToLog("No vead groups given to clustering algorithm.");
return(clusters);
}
var maxNumNewClusters = veadGroups[0].SiteResults.Length * _options.MaxNumNewClustersPerSite;
var nbhdStart = veadGroups[0].SiteResults[0].VcfReferencePosition;
var nbhdEnd = veadGroups[0].SiteResults[veadGroups[0].SiteResults.Length - 1].VcfReferencePosition;
Logger.WriteToLog("Maximum num new clusters for this nbhd is " + maxNumNewClusters);
Logger.WriteToLog("There are {0} variant sites in this nbhd, from position {1} to {2}", veadGroups[0].SiteResults.Length, nbhdStart, nbhdEnd);
if (_debug)
{
Logger.WriteToLog("variant-compressed read groups as follows: ");
Logger.WriteToLog("count" + "\t", veadGroups[0].ToPositions());
foreach (var vG in veadGroups)
{
Logger.WriteToLog("\t" + vG.NumVeads + "\t" + vG);
}
}
int outerIteration = 0;
while (veadGroups.Count > 0)
{
//Logger.WriteToLog("ITER: {0}.{1}\tNum clusters: {2} \tUnassigned read groups: {3}", outerIteration, 0, clusters.NumClusters, veadGroups.Count);
CreateNewCluster(veadGroups, clusters);
// Is there any poor-fitting read that should go somewhere else?
if (_options.AllowWorstFitRemoval)
{
clusters.ReAssignWorstFit();
}
//TODO log the contents of clusterSet (previously "PrintContents()").
const int maxNumReallocationIterations = 10;
var reallocationIterationNumber = 1;
// Keep trying to allocate free agents to clusters if possible, then tell how many we have left free.
// If we haven't allocated any in the last round, give up and break off a new cluster.
while (veadGroups.Count > 0)
{
var initialReadsLeft = veadGroups.Count;
//Logger.WriteToLog("ITER: {0}.{1}\tNum clusters: {2} \tUnassigned read groups: {3}", outerIteration, reallocationIterationNumber, clusters.NumClusters, initialReadsLeft);
// Find the best existing cluster for each free agent. Add the free agent to that cluster if it exists. Return anyone left free.
veadGroups = AllocateReadsToClusters(veadGroups, clusters,
_options.MaxNumberDisagreements);
//TODO log the contents of clusterSet (previously "PrintContents()").
if (veadGroups.Count == initialReadsLeft)
{
break; // Reallocation didn't do any good. Give up.
}
reallocationIterationNumber++;
if (reallocationIterationNumber > maxNumReallocationIterations)
{
break;
}
}
if (clusters.NumClusters > maxNumNewClusters)
{
break;
}
outerIteration++;
}
//Logger.WriteToLog("ITER: {0}.{1}\tNum clusters: {2} \tUnassigned read groups: {3}", outerIteration, 0, clusters.NumClusters, veadGroups.Count);
if (clusters != null)
{
//Logger.WriteToLog("Found " + clusters.Clusters.Length + " clusters in Nbhd " + nbhdID);
if (_options.ClusterConstraint > 0)
{
MeetPloidyConstraints(clusters);
}
}
return(clusters);
}
catch (Exception ex)
{
Logger.WriteToLog("Clustering issue.", ex);
throw;
}
}
private List <VeadGroup> AllocateReadsToClusters(List <VeadGroup> veadGroups, SetOfClusters clusters, int maxNumDisagreements)
{
var vgsRemaining = new List <VeadGroup>();
foreach (var vg in veadGroups)
{
var bestFits = clusters.GetClusterFits(vg);
if (bestFits.Count == 0)
{
vgsRemaining.Add(vg);
}
else
{
var bestCandidates = bestFits.Last().Value;
var bestcluster = bestCandidates[0];
if (_options.AllowClusterMerging)
{
bestcluster = ClusterMerger.MergeAllBestCandidates(clusters, maxNumDisagreements, bestCandidates, vg);
}
bestcluster.Add(vg);
}
}
return(vgsRemaining);
}
private void CreateNewCluster(List <VeadGroup> vgs, SetOfClusters clusters)
{
clusters.CreateAndAddCluster(vgs[0]);
vgs.Remove(vgs[0]);
}