Package 'MosaiClusteR'

Description

A deep-learning variant of ABCpp.SingleInMultiple that keeps the ABC algorithm of (Amaratunga et al. 2008) intact and replaces only Step 4: instead of clustering the raw feature sub-matrix, a neural autoencoder embeds the objects into a latent space and the latent vectors are clustered into NC base clusters. The stacked label matrix is fused across sources by f.clustABC.MultiSource, exactly as in M_ABCpp.

Usage

ABCdeep.SingleInMultiple(
  data,
  weighting = "var",
  normalize = NULL,
  gr = c(),
  bag = TRUE,
  numsim = 1000,
  numvar = 100,
  NC = NULL,
  topk_frac = 0.2,
  latent_dim = "auto",
  hidden_units = 32L,
  ae_epochs = 80L,
  ae_lr = 0.001,
  cluster_in_latent = "ward",
  linkage = "ward.D2",
  nugget_type = "between",
  nugget_args = list(),
  min_samples = 40L,
  seed = NULL
)

Arguments

Value

A data frame, numsim rows by one column per object; entries are base-cluster labels (0 = object not selected that iteration) – the same contract as ABCpp.SingleInMultiple.

Speed (trained once, forward pass per iteration)

The autoencoder is trained a single time on the source's top-topk_frac most-informative features; each of the numsim iterations then performs only a forward pass over the bootstrapped objects. The network is a self-contained, vectorised R autoencoder (He initialisation, ReLU, Adam, MSE) – there is no Python, TensorFlow or torch dependency, and no PCA or other fallback: a single code path.

References

Amaratunga D, Cabrera J, Kovtun V (2008). “Microarray Learning with ABC.” Biostatistics, 9(1), 128–136. doi:10.1093/biostatistics/kxm017. https://academic.oup.com/biostatistics/article-abstract/9/1/128/254198.

See Also

M_ABCdeep, ABCpp.SingleInMultiple, f.clustABC.MultiSource

Examples

data(mosaic_toy)
lab <- ABCdeep.SingleInMultiple(mosaic_toy$List[[1]], numsim = 40, NC = 3,
                                ae_epochs = 40, seed = 1)
dim(lab)

Description

A variant of ABCpp.SingleInMultiple that accumulates the mean scaled pairwise distance between co-selected objects across iterations, yielding a [0, 1] dissimilarity matrix that M_ABCdist fuses across sources.

Usage

ABCdist.SingleInMultiple(
  data,
  distmeasure = "euclidean",
  weighting = "var",
  normalize = NULL,
  gr = c(),
  bag = TRUE,
  numsim = 1000,
  numvar = 100,
  linkage = "ward.D",
  NC = NULL,
  mds = FALSE,
  nfeat = NULL,
  nugget_type = "between",
  nugget_args = list()
)

Arguments

Value

A list with D ($N \times N$ dissimilarity), NC and ng (per-iteration feature count); if mds = TRUE also mds_coords.

References

(Amaratunga et al. 2008)

See Also

M_ABCdist, ABCpp.SingleInMultiple

Examples

data(mosaic_toy)
out <- ABCdist.SingleInMultiple(mosaic_toy$List[[1]], numsim = 40)
dim(out$D)

Description

The canonical single-source ABC engine of (Amaratunga et al. 2008). Over numsim iterations it (optionally) bootstraps the objects, draws a weighted subsample of features, clusters the sub-matrix with fastcluster::hclust, and records each selected object's base-cluster label. M_ABCpp turns the stacked label matrix into a consensus dissimilarity using the C++ co-clustering kernel.

Usage

ABCpp.SingleInMultiple(
  data,
  distmeasure = "euclidean",
  weighting = "var",
  normalize = NULL,
  gr = c(),
  bag = TRUE,
  numsim = 1000,
  numvar = 100,
  linkage = "ward.D2",
  NC = NULL,
  mds = FALSE,
  nfeat = NULL,
  nugget_type = "between",
  nugget_args = list()
)

Arguments

Value

A data frame, numsim rows by one column per object; entries are base-cluster labels (0 = object not selected that iteration).

Feature weighting (incl. data nuggets)

Features are subsampled with probability proportional to a rank-transformed importance score:

• "var" – feature variance (the classic ABC choice);

• "cv" – coefficient of variation;

• "nugget" – NEW: the data-nugget between-group weighted variance (create_data_nuggets), a robust, Big-Data-friendly alternative to the raw variance;

• anything else – equal weights.

References

Amaratunga D, Cabrera J, Kovtun V (2008). “Microarray Learning with ABC.” Biostatistics, 9(1), 128–136. doi:10.1093/biostatistics/kxm017. https://academic.oup.com/biostatistics/article-abstract/9/1/128/254198.

See Also

M_ABCpp, create_data_nuggets

Examples

data(mosaic_toy)
lab <- ABCpp.SingleInMultiple(mosaic_toy$List[[1]], numsim = 40, NC = 3,
                              weighting = "nugget")
dim(lab)

Description

Aggregated Data Clustering (ADC) is a direct clustering multi-source technique. The data matrices are merged into a single fused data matrix on which a dissimilarity matrix is computed. Hierarchical clustering is then performed on this dissimilarity matrix.

Usage

ADC(
  List,
  distmeasure = "tanimoto",
  normalize = FALSE,
  method = NULL,
  clust = "agnes",
  linkage = "flexible",
  alpha = 0.625
)

Arguments

Value

The returned value is a list with the following three elements.

The value has class 'ADC'. The Clust element will be of interest for further applications.

References

Fodeh, S. J., Punch, W. & Tan, P.-N. (2013). On unifying multi-source information from probabilistically generated functional data. Proteins 81:1298-1310.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List
res_ADC <- ADC(List=L, distmeasure="euclidean", normalize=FALSE, method=NULL,
               clust="agnes", linkage="flexible", alpha=0.625)

## End(Not run)

Description

Aggregated Data Ensemble Clustering (ADEC) is a direct clustering multi-source technique. ADEC is an iterative procedure which starts with the merging of the data sets. In each iteration, a random sample of the features is selected and/or a resulting dendrogram is divided into k clusters for a range of values of k.

Usage

ADEC(
  List,
  distmeasure = "tanimoto",
  normalize = FALSE,
  method = NULL,
  t = 10,
  r = NULL,
  nrclusters = NULL,
  clust = "agnes",
  linkage = "flexible",
  alpha = 0.625
)

Arguments

Details

If r is specified and nrclusters is a fixed number, only a random sampling of the features will be performed for the t iterations (ADECa). If r is NULL and the nrclusters is a sequence, the clustering is performed on all features and the dendrogam is divided into clusters for the values of nrclusters (ADECb). If both r is specified and nrclusters is a sequence, the combination is performed (ADECc). After every iteration, either be random sampling, multiple divisions of the dendrogram or both, an incidence matrix is set up. All incidence matrices are summed and represent the distance matrix on which a final clustering is performed.

Value

The returned value is a list with the following three elements.

The value has class 'ADEC'. The Clust element will be of interest for further applications.

References

Fodeh, S. J., Punch, W. & Tan, P.-N. (2013). On unifying multi-source information from probabilistically generated functional data. Proteins 81:1298-1310.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List
res_ADEC <- ADEC(List=L, distmeasure="euclidean", normalize=FALSE, method=NULL,
                 t=5, r=50, nrclusters=5, clust="agnes", linkage="flexible", alpha=0.625)

## End(Not run)

Description

A tool to visualize characteristic binary features of a set of objects in comparison with the remaining objects for multiple data sets. The result is a matrix with coloured cells. Columns represent objects and rows represent the specified features. A feature which is present is give a coloured cell while an absent feature is represented by a grey cell. The labels on the right indicate the names of the features while the labels on the bottom are the names of the objects.

Usage

BinFeaturesPlot_MultipleData(
  leadCpds,
  orderLab,
  features = list(),
  data = list(),
  validate = NULL,
  colorLab = NULL,
  nrclusters = NULL,
  cols = NULL,
  name = c("Data1", "Data2"),
  colors1 = c("gray90", "blue"),
  colors2 = c("gray90", "green"),
  margins = c(5.5, 3.5, 0.5, 5.5),
  cexB = 0.8,
  cexL = 0.8,
  cexR = 0.8,
  spaceNames = 0.2,
  plottype = "new",
  location = NULL
)

Arguments

Value

A plot visualizing the presence and absence of the characteristic binary features across multiple data sets.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

Comps=FindCluster(list(MCF7_F),nrclusters=10,select=c(1,8))

MCF7_Char=CharacteristicFeatures(List=NULL,Selection=Comps,binData=
list(fingerprintMat,targetMat),datanames=c("FP","TP"),nrclusters=NULL,
topC=NULL,sign=0.05,fusionsLog=TRUE,weightclust=TRUE,names=c("FP","TP"))

FeatFP=MCF7_Char$Selection$Characteristics$FP$TopFeat$Names[c(1:10)]
FeatTP=MCF7_Char$Selection$Characteristics$TP$TopFeat$Names[c(1:10)]

BinFeaturesPlot_MultipleData(leadCpds=Comps,orderLab=MCF7_Char$Selection$
objects$OrderedCpds,features=list(FeatFP,FeatTP),data=list(fingerprintMat,targetMat),
validate=NULL,colorLab=NULL,nrclusters=NULL,cols=NULL,name=c("FP","TP"),colors1=
c('gray90','blue'),colors2=c('gray90','green'),margins=c(5.5,3.5,0.5,5.5),cexB=0.80,
cexL=0.80,cexR=0.80,spaceNames=0.20,plottype="new",location=NULL)

## End(Not run)

Description

A tool to visualize characteristic binary features of a set of objects in comparison with the remaining objects for a single data set. The result is a matrix with coloured cells. Columns represent objects and rows represent the specified features. A feature which is present is give a coloured cell while an absent feature is represented by a grey cell. The labels on the right indicate the names of the features while the labels on the bottom are the names of the objects.

Usage

BinFeaturesPlot_SingleData(
  leadCpds = c(),
  orderLab = c(),
  features = c(),
  data = NULL,
  colorLab = NULL,
  nrclusters = NULL,
  cols = NULL,
  name = c("Data"),
  colors1 = c("gray90", "blue"),
  colors2 = c("gray90", "green"),
  highlightFeat = NULL,
  margins = c(5.5, 3.5, 0.5, 5.5),
  plottype = "new",
  location = NULL
)

Arguments

Value

A plot visualizing the presence and absence of the characteristic binary features.

Examples

## Not run: 
data(fingerprintMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

Comps=FindCluster(list(MCF7_F),nrclusters=10,select=c(1,8))

MCF7_Char=CharacteristicFeatures(List=list(fingerprintMat),Selection=Comps,
binData=list(fingerprintMat),datanames=c("FP"),nrclusters=NULL,topC=NULL,
sign=0.05,fusionsLog=TRUE,weightclust=TRUE,names=c("FP"))
Feat=MCF7_Char$Selection$Characteristics$FP$TopFeat$Names[c(1:10)]

BinFeaturesPlot_SingleData(leadCpds=Comps,orderLab=MCF7_Char$Selection$
objects$OrderedCpds,features=Feat,data=fingerprintMat,colorLab=NULL,
nrclusters=NULL,cols=NULL,name=c("FP"),colors1=c('gray90','blue'),colors2=
c('gray90','green'),highlightFeat=NULL,margins=c(5.5,3.5,0.5,5.5),
plottype="new",location=NULL)

## End(Not run)

Description

Given two distance matrices, the function categorizes one distance matrix and produces a box plot from the other distance matrix against the created categories. The option is available to choose one of the plots or to have both plots. The function also works on outputs from ADEC and CEC functions which do not have distance matrices but incidence matrices.

Usage

BoxPlotDistance(
  Data1,
  Data2,
  type = c("data", "dist", "clusters"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  lab1,
  lab2,
  limits1 = NULL,
  limits2 = NULL,
  plot = 1,
  StopRange = FALSE,
  plottype = "new",
  location = NULL
)

Arguments

Value

One/multiple box plots.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)


BoxPlotDistance(MCF7_F,MCF7_T,type="cluster",lab1="FP",lab2="TP",limits1=c(0.3,0.7),
limits2=c(0.3,0.7),plot=1,StopRange=FALSE,plottype="new", location=NULL)

## End(Not run)

Description

Complementary Ensemble Clustering (CEC) Complementary Ensemble Clustering (CEC, Fodeh2013) shows similarities with ADEC. However, instead of merging the data matrices, ensemble clustering is performedon each data matrix separately. The resulting incidence matrices for each data sets are combined in a weighted linear equation. The weighted incidence matrix is the input for the final clustering algorithm. Similarly as ADEC, there are versions depending of the specification of the number of features to sample and the number of clusters.

Usage

CEC(
  List,
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  t = 10,
  r = NULL,
  nrclusters = NULL,
  weight = NULL,
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  weightclust = 0.5
)

Arguments

Details

If r is specified and nrclusters is a fixed number, only a random sampling of the features will be performed for the t iterations (CECa). If r is NULL and the nrclusters is a sequence, the clustering is performedon all features and the dendrogam is divided into clusters for the values of nrclusters (CECb). If both r is specified and nrclusters is a sequence, the combination is performed (CECc). After every iteration, either be random sampling, multiple divisions of the dendrogram or both, an incidence matrix is set up. All incidence matrices are summed and represent the distance matrix on which a final clustering is performed.

Value

The returned value is a list of four elements:

The value has class 'CEC'.

References

Fodeh SJ, Punch W, Tan P (2013). “On Unifying Multi-source Information to Improve Subcellular Protein Localization Prediction.” Proteins: Structure, Function, and Bioinformatics, 81, 1298–1310.

Description

CharacteristicFeatures identifies the features that are characteristic for the objects of each cluster of one or more clustering results. Binary features are evaluated with Fisher's exact test, continuous features with the t-test. Multiplicity correction is included.

Usage

CharacteristicFeatures(
  List,
  Selection = NULL,
  binData = NULL,
  contData = NULL,
  datanames = NULL,
  nrclusters = NULL,
  sign = 0.05,
  topChar = NULL,
  fusionsLog = TRUE,
  weightclust = TRUE,
  names = NULL
)

Arguments

Value

A list with for each method the found (top) characteristics of the feature data per cluster.

References

Perualila-Tan N. et al. (2016). Weighted similarity-based clustering of chemical structures and bioactivity data in early drug discovery.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(MCF7_F,MCF7_T)
MCF7_Char=CharacteristicFeatures(List=L,Selection=NULL,binData=list(fingerprintMat),
contData=NULL,datanames=c("FP"),nrclusters=7,topChar=20)

## End(Not run)

Description

If desired, the function produces a dendrogram of a clustering result. One or multiple clusters can be indicated by a mouse click. From these clusters DE genes and characteristic features are determined. It is also possible to provide the objects of interest without producing the plot.

Usage

ChooseCluster(
  Interactive = TRUE,
  leadCpds = NULL,
  clusterResult = NULL,
  colorLab = NULL,
  binData = NULL,
  contData = NULL,
  datanames = c("FP"),
  geneExpr = NULL,
  topChar = 20,
  topG = 20,
  sign = 0.05,
  nrclusters = NULL,
  cols = NULL,
  n = 1
)

Arguments

Details

This function may require the suggested packages 'a4Base', 'a4Core' and 'limma' when a gene expression matrix is supplied.

Value

A list with one element per cluster of interest with elements objects, Characteristics and (optionally) Genes.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)
data(geneMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

MCF7_Interactive=ChooseCluster(Interactive=TRUE,leadCpds=NULL,clusterResult=MCF7_T,
colorLab=MCF7_F,binData=list(fingerprintMat),datanames=c("FP"),geneExpr=geneMat,
topChar = 20, topG = 20,nrclusters=7,n=1)

## End(Not run)

Description

Cluster one data (or distance) matrix with agglomerative hierarchical clustering. This is the Layer-2 baseline primitive of the MoSaIC framework: run it on each tile on its own to see what each modality recovers before integrating. It also underpins HierarchicalEnsembleClustering.

Usage

Cluster(
  Data,
  type = c("data", "dist"),
  distmeasure = "euclidean",
  normalize = FALSE,
  method = NULL,
  clust = "agnes",
  linkage = "ward",
  alpha = 0.625,
  StopRange = TRUE,
  ...
)

Arguments

Value

A list with DistM (the dissimilarity matrix) and Clust (an agnes object).

See Also

Distance, M_ABCpp

Examples

data(mosaic_toy)
base1 <- Cluster(t(mosaic_toy$List[[1]]), type = "data",
                 distmeasure = "euclidean")
mosaic_labels(base1, k = 3)[1:6]

Description

Computes the adjusted Rand index (ARI), normalised mutual information (NMI) and pairwise Jaccard between two label vectors of equal length.

Usage

cluster_agreement(a, b)

Arguments

Details

ARI rule of thumb: >0.90 excellent; 0.75-0.90 good; 0.50-0.75 moderate; <0.50 weak.

Value

A named numeric vector c(ARI, NMI, Jaccard).

Examples

cluster_agreement(c(1, 1, 2, 2), c(2, 2, 1, 1))   # identical up to relabel

Description

Internal helper applied via dendrapply that colours the leaves and edges of a dendrogram according to the cluster a leaf belongs to, or to a specific selection of objects. Used by ClusterPlot.

Usage

ClusterCols(x, Data, nrclusters = NULL, cols = NULL, colorComps = NULL)

Arguments

Value

The dendrogram node x with coloured nodePar and edgePar attributes.

Description

The ClusteringAggregation includes the ensemble clustering methods Balls, Agglomerative (Aggl.) and Furthest which are graph-based consensus methods.

Usage

ClusteringAggregation(
  List,
  type = c("data", "dist", "clust"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  nrclusters = c(7, 7),
  gap = FALSE,
  maxK = 15,
  agglMethod = c("Balls", "Aggl", "Furthest", "LocalSearch"),
  improve = TRUE,
  distThresh_B = 0.5,
  distThresh_A = 0.8
)

Arguments

Details

Gionis, Mannila and Tsaparas (2007) propose heuristic algorithms in order to find a solution for the consensus problem. In a first step, a weighted graph is built from the objects with weights between two vertices determined by the fraction of clusterings that place the two vertices in different clusters. In a second step, an algorithm searches for the partition that minimizes the total number of disagreements with the given partitions. The Balls algorithm is an iterative process which finds a cluster for the consensus partition in each iteration. For each object, all objects at a distance of at most 0.5 are collected and the average distance of this set to the object of interest is calculated. If the average distance is less or equal to a parameter the objects form a cluster; otherwise the object forms a singleton. The Agglomerative (Aggl.) algorithm starts by considering every object as a singleton cluster. Next, the two closest clusters are merged if the average distance between the clusters is less than 0.5. If there are no two clusters with an average distance smaller than 0.5, the algorithm stops and returns the created clusters as a solution. The Furthest algorithm starts by placing all objects into a single cluster. In each iteration, the pair of objects that are the furthest apart are considered as new cluster centers. The remaining objects are appointed to the center that increases the cost of the partition the least and the new cost is computed. The cost is the sum of the all distances between the obtained partition and the partitions in the ensemble. The iteration continues until the cost of the new partition is higher than the previous partition.

Value

The returned value is a list of two elements:

The value has class 'Ensemble'.

References

Gionis, A., Mannila, H. and Tsaparas, P. (2007). Clustering aggregation. ACM Transactions on Knowledge Discovery from Data, 1(1), 4-es.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List
Aggl_toy <- ClusteringAggregation(List = L, type = "data",
  distmeasure = c("euclidean", "euclidean"), normalize = c(FALSE, FALSE),
  method = c(NULL, NULL), clust = "agnes", linkage = c("ward", "ward"),
  alpha = 0.625, nrclusters = c(7, 7), agglMethod = "Aggl",
  improve = TRUE, distThresh_B = 0.5, distThresh_A = 0.8)

## End(Not run)

Description

Plot a dendrogram with leaves colored by a result of choice.

Usage

ClusterPlot(
  Data1,
  Data2 = NULL,
  nrclusters = NULL,
  cols = NULL,
  colorComps = NULL,
  hangdend = 0.02,
  plottype = "new",
  location = NULL,
  ...
)

Arguments

Value

A plot of the dendrogram of the first clustering result with colored leaves. If a second clustering result is given in Data2, the colors are based on this clustering result.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)
data(Colors1)

MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

ClusterPlot(MCF7_T ,nrclusters=7,cols=Colors1,plottype="new",location=NULL,
main="Clustering on Target Predictions: Dendrogram",ylim=c(-0.1,1.8))

## End(Not run)

Description

In order to facilitate the visualization of the influence of the different methods on the clustering of the objects, colours can be used. The function ColorPalette is able to pick out as many colours as there are clusters. This is done with the help of the ColorRampPalette function of the grDevices package

Usage

ColorPalette(colors = c("red", "green"), ncols = 5)

Arguments

Value

A vector containing the hex codes of the chosen colors.

Examples

Colors1<-ColorPalette(c("cadetblue2","chocolate","firebrick2",
"darkgoldenrod2", "darkgreen","blue2","darkorchid3","deeppink2"), ncols=8)

Description

The ColorsNames function is used on the output of the ReorderToReference and matches the cluster numbers indicated by the cell with the names of the colors. This is necessary to produce the plot of the ComparePlot function and is therefore an internal function of this function but can also be applied separately.

Usage

ColorsNames(matrixColors, cols = NULL)

Arguments

Value

A matrix containing the hex code of the color that corresponds to each cell of the matrix to be colored. This function is called upon by the ComparePlot function.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)
data(Colors1)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(MCF7_F,MCF7_T)
names=c("FP","TP")

MatrixColors=ReorderToReference(List=L,nrclusters=7,fusionsLog=TRUE,weightclust=TRUE,
names=names)

Names=ColorsNames(matrixColors=MatrixColors,cols=Colors1)

## End(Not run)

Description

Clusters two dissimilarity matrices with the same linkage and number of clusters and reports their agreement (ARI, NMI, pairwise Jaccard) and the contingency table.

Usage

compare_clusterings(
  D1,
  D2,
  NC = NULL,
  linkage = "ward.D2",
  labels = c("D1", "D2"),
  verbose = TRUE
)

Arguments

Value

Invisibly, a list with the two label vectors, ARI, NMI, pair_Jaccard and the contingency table.

Examples

set.seed(1)
D1 <- as.matrix(dist(matrix(rnorm(60), 20)))
D2 <- as.matrix(dist(matrix(rnorm(60), 20)))
compare_clusterings(D1, D2, NC = 3, verbose = FALSE)$ARI

Description

The function CompareInteractive produces an interactive plot of the multiple source clustering results in ListM. By identifying a cluster or a method, a comparison with the single source clustering results in ListS is shown in a new plotting window.

Usage

CompareInteractive(
  ListM,
  ListS,
  nrclusters = NULL,
  cols = NULL,
  fusionsLogM = FALSE,
  fusionsLogS = FALSE,
  weightclustM = FALSE,
  weightclustS = FALSE,
  namesM = NULL,
  namesS = NULL,
  marginsM = c(2, 2.5, 2, 2.5),
  marginsS = c(8, 2.5, 2, 2.5),
  Interactive = TRUE,
  n = 1
)

Arguments

Value

A plot of the comparison of the elements of ListM, on which multiple clusters and/or methods can be identified to compare them to the elements in ListS.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)
data(Colors1)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(fingerprintMat,targetMat)

MCF7_W=WeightedClust(List=L,type="data",distmeasure=c("tanimoto","tanimoto"),
normalize=c(FALSE,FALSE),method=c(NULL,NULL),weight=seq(1,0,-0.1),weightclust=0.5,
clust="agnes",linkage="ward",StopRange=FALSE)

ListM=list(MCF7_W)
namesM=c(seq(1.0,0.0,-0.1))

ListS=list(MCF7_F,MCF7_T)
namesS=c("FP","TP")

CompareInteractive(ListM,ListS,nrclusters=7,cols=Colors1,fusionsLogM=FALSE,
fusionsLogS=FALSE,weightclustM=FALSE,weightclustS=TRUE,namesM,namesS,
marginsM=c(2,2.5,2,2.5),marginsS=c(8,2.5,2,2.5),Interactive=TRUE,n=1)

## End(Not run)

Description

A visual comparison of the clustering results of several methods. The function relies on ReorderToReference and ColorsNames and renders the resulting matrix either as a rectangular heatmap (using the plotrix package) or in a circular format (using the circlize package).

Usage

ComparePlot(
  List,
  nrclusters = NULL,
  cols = NULL,
  fusionsLog = FALSE,
  weightclust = FALSE,
  names = NULL,
  margins = c(8.1, 3.1, 3.1, 4.1),
  circle = FALSE,
  canvaslims = c(-1, 1, -1, 1),
  Highlight = NULL,
  substr = NULL,
  cex.highlight = 1,
  cex.labels = 0.7,
  trackheightdend = 0.5,
  plottype = "new",
  location = NULL
)

Arguments

Value

A plot which translates the matrix output of ReorderToReference.

Description

The function CompareSilCluster compares the results of two medoid clusterings. The null hypothesis is that the clustering is identical. A test statistic is calculated and a p-value obtained with bootstrapping.

Usage

CompareSilCluster(
  List,
  type = c("data", "dist"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  nrclusters = NULL,
  names = NULL,
  nboot = 100,
  plottype = "new",
  location = NULL
)

Arguments

Value

A plot of the density of the statistic under the null hypothesis. Further a list with two elements:

.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

List=list(fingerprintMat,targetMat)

Comparison=CompareSilCluster(List=List,type="data",
distmeasure=c("tanimoto","tanimoto"),normalize=c(FALSE,FALSE),method=c(NULL,NULL),
nrclusters=7,names=NULL,nboot=100,plottype="new",location=NULL)

Comparison

## End(Not run)

Description

The function CompareSvsM plots the comparison of the single source clustering results on the left and that of the multiple source clustering results on the right such that a visual comparison is possible.

Usage

CompareSvsM(
  ListS,
  ListM,
  nrclusters = NULL,
  cols = NULL,
  fusionsLogS = FALSE,
  fusionsLogM = FALSE,
  weightclustS = FALSE,
  weightclustM = FALSE,
  namesS = NULL,
  namesM = NULL,
  margins = c(8.1, 3.1, 3.1, 4.1),
  plottype = "new",
  location = NULL
)

Arguments

Value

A plot with on the left the comparison over the objects in ListS and on the right a comparison over the objects in ListM.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)
data(Colors1)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(fingerprintMat,targetMat)

MCF7_W=WeightedClust(List=L,type="data", distmeasure=c("tanimoto","tanimoto"),
normalize=c(FALSE,FALSE),method=c(NULL,NULL),weight=seq(1,0,-0.1),weightclust=0.5
,clust="agnes",linkage="ward",StopRange=FALSE)

ListM=list(MCF7_W)
namesM=seq(1.0,0.0,-0.1)

ListS=list(MCF7_F,MCF7_T)
namesS=c("FP","TP")

CompareSvsM(ListS,ListM,nrclusters=7,cols=Colors1,fusionsLogS=FALSE,
fusionsLogM=FALSE,weightclustS=FALSE,weightclustM=FALSE,namesS,
namesM,plottype="new",location=NULL)

## End(Not run)

Description

Consensus Clustering performs voting-based consensus on a set of single-source partitions. Each data matrix is clustered separately and the resulting label matrix is reconciled into a single consensus partition by an iterative voting scheme. Three voting methods are available: Iterative Voting Consensus (IVC), Iterative Probabilistic Voting Consensus (IPVC) and Iterative Pairwise Consensus (IPC).

Usage

ConsensusClustering(
  List,
  type = c("data", "dist", "clust"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  nrclusters = c(7, 7),
  gap = FALSE,
  maxK = 15,
  votingMethod = c("IVC", "IPVC", "IPC"),
  optimalk = 7
)

Arguments

Value

The returned value is a list of two elements:

The value has class 'Ensemble'.

References

Nguyen N. and Caruana R. (2007). Consensus Clusterings. Seventh IEEE International Conference on Data Mining (ICDM 2007), 607-612.

Examples

## Not run: 
data(mosaic_toy)
fit <- ConsensusClustering(List = mosaic_toy$List, type = "data",
  distmeasure = c("euclidean", "euclidean"), normalize = c(FALSE, FALSE),
  method = c(NULL, NULL), clust = "agnes", linkage = c("ward", "ward"),
  nrclusters = c(3, 3), gap = FALSE, votingMethod = "IVC", optimalk = 3)

## End(Not run)

Description

The function ContFeaturesPlot plots the values of continuous features. It is possible to separate between objects of interest and the other objects.

Usage

ContFeaturesPlot(
  leadCpds,
  data,
  nrclusters = NULL,
  orderLab = NULL,
  colorLab = NULL,
  cols = NULL,
  ylab = "features",
  addLegend = TRUE,
  margins = c(5.5, 3.5, 0.5, 8.7),
  plottype = "new",
  location = NULL
)

Arguments

Value

A plot in which the values of the features of the leadCpds are separeted from the others.

Examples

## Not run: 
data(Colors1)
Comps=c("Cpd1", "Cpd2", "Cpd3", "Cpd4", "Cpd5")

Data=matrix(sample(15, size = 50*5, replace = TRUE), nrow = 50, ncol = 5)
colnames(Data)=colnames(Data, do.NULL = FALSE, prefix = "col")
rownames(Data)=rownames(Data, do.NULL = FALSE, prefix = "row")
for(i in 1:50){
	rownames(Data)[i]=paste("Cpd",i,sep="")
}

ContFeaturesPlot(leadCpds=Comps,orderLab=rownames(Data),colorLab=NULL,data=Data,
nrclusters=7,cols=Colors1,ylab="features",addLegend=TRUE,margins=c(5.5,3.5,0.5,8.7),
plottype="new",location=NULL)

## End(Not run)

Description

create_data_nuggets compresses n observations into M <= n representative "nuggets". Each nugget stores a centre (location), a weight (the number of member observations) and a scale (within-nugget variability). Data nuggets preserve the covariance structure of the data - including the periphery - far better than a random sub-sample, which makes them a robust, memory-light surrogate for the full data set when clustering or computing feature weights.

Usage

create_data_nuggets(
  x,
  max_nuggets = NULL,
  min_size = 1L,
  engine = c("native", "datanugget"),
  scale_features = TRUE,
  seed = NULL,
  ...
)

Arguments

Details

With the native engine the number of nuggets defaults to min(nrow(x), max(2, floor(nrow(x) / 10))) when n > 50, and to n otherwise (in which case every observation is its own nugget and the nugget-weighted variance reduces to the ordinary variance - a safe fallback for small samples). The partition is obtained with stats::kmeans; missing values are mean-imputed for the partition step only.

Value

An object of class "data_nugget": a list with elements

References

Cherasia KE, Cabrera J, Fernholz LT, Fernholz R (2023). “Data Nuggets in Supervised Learning.” In Robust and Multivariate Statistical Methods: Festschrift in Honor of David E. Tyler, 429–449. Springer.

Dey T, Duan Y, Cabrera J, Cheng X (2025). WCluster: Clustering and PCA with Weights, and Data Nuggets Clustering. R package version 1.3.0.

See Also

nugget_feature_weights, M_ABCpp

Examples

set.seed(1)
x <- rbind(matrix(rnorm(200 * 5,  0), ncol = 5),
           matrix(rnorm(200 * 5,  4), ncol = 5))
dn <- create_data_nuggets(x, max_nuggets = 40)
dn

Description

CVAA performs cumulative voting consensus clustering. A reference partition is used to align the individual source partitions, which are then aggregated into a consensus stochastic membership matrix. Two variants are available: cumulative voting ("CVAA") and the information-weighted variant ("W-CVAA").

Usage

CVAA(
  Reference = NULL,
  nrclustersR = 7,
  List,
  typeL = c("data", "dist", "clust"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  nrclusters = c(7, 7),
  gap = FALSE,
  maxK = 15,
  votingMethod = c("CVAA", "W-CVAA"),
  optimalk = nrclustersR
)

Arguments

Details

Cumulative voting aggregation aligns each source membership matrix to a running consensus by least-squares voting, then averages cumulatively. The weighted variant orders the partitions by average information content and weights their contribution accordingly. When optimalk is smaller than nrclustersR the JS-ALink agglomeration is used to merge reference clusters.

Value

A list of two elements:

The value has class 'Ensemble'.

References

Ayad H.G. and Kamel M.S. (2008). Cumulative voting consensus method for partitions with variable number of clusters. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(1), 160-173.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List
Ref <- Cluster(L[[1]], type = "data", distmeasure = "euclidean")
attr(Ref, "method") <- "Single"
res <- CVAA(Reference = Ref, nrclustersR = 5, List = L, typeL = "data",
  distmeasure = c("euclidean","euclidean"), votingMethod = "CVAA", optimalk = 5)

## End(Not run)

Description

A visual comparison of all methods is handy to see which objects will always cluster together independent of the applied methods. To this aid the function Cyclogram has been written. The function relies on methods of the circlize package.

Usage

Cyclogram(
  List,
  nrclusters = NULL,
  cols = NULL,
  fusionsLog = FALSE,
  weightclust = FALSE,
  names = NULL,
  canvaslims = c(-1, 1, -1, 1),
  margins = c(8.1, 3.1, 3.1, 4.1),
  Highlight = NULL,
  cex.highlight = 1,
  substr = NULL,
  cex.labels = 0.7,
  plottype = "new",
  location = NULL
)

Arguments

Details

This function makes use of the functions ReorderToReference and Colorsnames. Given a list with the outputs of several methods, the first step is to call upon ReorderToReference and to produce a matrix of which the columns are ordered according to the ordering of the objects of the first method in the list. Each cell represent the number of the cluster the object belongs to for a specific method indicated by the rows. The clusters are arranged in such a way that these correspond to that one cluster of the referenced method that they have the most in common with. The circlize package is used to visualize the matrix is a circular format. The inner element of the circle is the first element of the List parameter, the second inner element is the second element of the list and so on. The object names are written on the circular dendrogram portayed in the center of the circle.

Value

A plot which translates the matrix output of the function ReorderToReference into a circular format.

Description

The function DetermineWeight_SilClust determines an optimal weight for weighted similarity clustering by calculating silhouettes widths. For each given weight, a linear combination of the distance matrices of the single data sources is obtained, medoid clustering is performed and the silhouette widths are regressed against the cluster memberships. A statistic is derived and a p-value obtained via bootstrapping.

Usage

DetermineWeight_SilClust(
  List,
  type = c("data", "dist", "clusters"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  weight = seq(0, 1, by = 0.01),
  nrclusters = NULL,
  names = NULL,
  nboot = 10,
  StopRange = FALSE,
  plottype = "new",
  location = NULL
)

Arguments

Value

Two plots and a list with two elements:

.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(MCF7_F,MCF7_T)

MC7_Weight=DetermineWeight_SilClust(List=L,type="clusters",distmeasure=
c("tanimoto","tanimoto"),normalize=c(FALSE,FALSE),method=c(NULL,NULL),
weight=seq(0,1,by=0.01),nrclusters=c(7,7),names=c("FP","TP"),nboot=10,
StopRange=FALSE,plottype="new",location=NULL)

## End(Not run)

Description

The function DetermineWeight_SimClust determines an optimal weight for performing weighted similarity clustering. For each given weight, each separate clustering is compared to the clustering on a weighted dissimilarity matrix and a Jaccard coefficient is calculated. The ratio of the Jaccard coefficients closest to one indicates an optimal weight.

Usage

DetermineWeight_SimClust(
  List,
  type = c("data", "dist", "clusters"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  weight = seq(0, 1, by = 0.01),
  nrclusters = NULL,
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  linkageF = "ward",
  alpha = 0.625,
  gap = FALSE,
  maxK = 15,
  names = NULL,
  StopRange = FALSE,
  plottype = "new",
  location = NULL
)

Arguments

Value

The returned value is a list with three elements:

.

References

Perualila-Tan, N. et al. (2016). Weighted similarity-based clustering of chemical structures and bioactivity data in early drug discovery. Journal of Bioinformatics and Computational Biology.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",alpha=0.625,gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",alpha=0.625,gap=FALSE,maxK=55,StopRange=FALSE)

L=list(MCF7_F,MCF7_T)

MCF7_Weight=DetermineWeight_SimClust(List=L,type="clusters",weight=seq(0,1,by=0.01),
nrclusters=c(7,7),distmeasure=c("tanimoto","tanimoto"),normalize=c(FALSE,FALSE),
method=c(NULL,NULL),clust="agnes",linkage=c("flexible","flexible"),linkageF="ward",
alpha=0.625,gap=FALSE,maxK=50,names=c("FP","TP"),StopRange=FALSE,plottype="new",location=NULL)

## End(Not run)

Description

The DiffGenes function looks for differentially expressed genes for each cluster of a clustering result by comparing the objects of a cluster against all other objects. The limma method is applied to find the differentially expressed genes.

Usage

DiffGenes(
  List,
  Selection = NULL,
  geneExpr = NULL,
  nrclusters = NULL,
  method = "limma",
  sign = 0.05,
  topG = NULL,
  fusionsLog = TRUE,
  weightclust = TRUE,
  names = NULL
)

Arguments

Details

This function relies on the suggested Bioconductor package 'limma'.

Value

The result is a list with an element per method. Each element is again a list with for each cluster the found differentially expressed genes.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)
data(geneMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)
MCF7_T = Cluster(targetMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(MCF7_F,MCF7_T)
names=c('FP','TP')

MCF7_FT_DE = DiffGenes(List=L,geneExpr=geneMat,nrclusters=7,method="limma",sign=0.05,
topG=10,fusionsLog=TRUE,weightclust=TRUE,names=names)

## End(Not run)

Description

Internal function of DiffGenes.

Usage

DiffGenesSelection(
  List,
  Selection,
  geneExpr = NULL,
  nrclusters = NULL,
  method = "limma",
  sign = 0.05,
  topG = NULL,
  fusionsLog = TRUE,
  weightclust = TRUE,
  names = NULL
)

Arguments

Examples

## Not run: 
MCF7_FT_DE = DiffGenes(List=L,Selection=c("Cpd1","Cpd2"),geneExpr=geneMat,method="limma")

## End(Not run)

Description

Dispatcher computing an object-by-object distance matrix under a range of measures suitable for continuous, binary and mixed-type data.

Usage

Distance(
  Data,
  distmeasure = c("tanimoto", "jaccard", "euclidean", "hamming", "cont tanimoto",
    "MCA_coord", "gower", "chi.squared", "cosine"),
  normalize = NULL,
  method = NULL
)

Distance_v2(Data, distmeasure = "tanimoto", normalize = NULL)

Arguments

Details

"cosine", "chi.squared" and "MCA_coord" require the suggested packages lsa, analogue and FactoMineR respectively.

Value

A symmetric numeric distance matrix.

Examples

m <- matrix(rbinom(40, 1, 0.5), nrow = 8)
D <- Distance(m, distmeasure = "tanimoto")
dim(D)

Description

Internal function of HeatmapPlot

Usage

distanceheatmaps(Data1, Data2, names = NULL, nrclusters = 7)

Arguments

Description

Ensemble Hierarchical Clustering (EHC) builds a cluster association graph from the agglomerative clustering results of each data source. Strengths of association between pairs of objects are accumulated across the dendrograms (weighted by cluster depth and intra-cluster proximity), yielding a weighted adjacency matrix that is then partitioned.

Two partitioning strategies are available through graphPartitioning:

"METIS"

Delegates the partitioning of the association graph to an external graph-partitioning executable (METIS / gpmetis / shmetis, or the corresponding MATLAB hmetis routine). These binaries are not bundled with the package and must be configured through the executable argument. When executable=FALSE (the default), or when the configured executable cannot be located, the function stops with an informative error.

"MST"

A pure-R minimum spanning tree partitioning that needs no external executable. It does require the igraph package; if igraph is not installed the function stops with an informative error.

Usage

EHC(
  List,
  type = c("data", "dist", "clust"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  gap = FALSE,
  maxK = 15,
  graphPartitioning = c("METIS", "MST"),
  optimalk = 7,
  waitingtime = 300,
  file_number = 0,
  executable = FALSE
)

Arguments

Value

The returned value is a list of two elements:

The value has class 'Ensemble'.

References

Mirzaei A. and Rahmati M. (2010). A novel hierarchical-clustering-combination scheme based on fuzzy-similarity relations. IEEE Transactions on Fuzzy Systems, 18(1), 27-39.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List

# METIS path requires an external graph-partitioning executable / MATLAB
# configured through the 'executable' argument.
res <- EHC(List = L, type = "data",
  distmeasure = c("euclidean", "euclidean"), normalize = c(FALSE, FALSE),
  method = c(NULL, NULL), clust = "agnes",
  linkage = c("flexible", "flexible"), alpha = 0.625, gap = FALSE,
  maxK = 15, graphPartitioning = "METIS", optimalk = 7,
  executable = "./MetisAlgorithm")

# MST path is pure R but needs the igraph package.
res2 <- EHC(List = L, type = "data", graphPartitioning = "MST")

## End(Not run)

Description

Cluster-based ensemble clustering that combines several single-source partitions into a single consensus partition through the graph-partitioning algorithms of Strehl and Ghosh: the Cluster-based Similarity Partitioning Algorithm (CSPA), the HyperGraph Partitioning Algorithm (HGPA) and the Meta-CLustering Algorithm (MCLA), as well as a "Best" selection strategy.

External executable required. This method does not run in pure R. It writes the stacked label matrix to disk and delegates the actual graph partitioning to an external graph-partitioning executable (e.g. a compiled CSPA/HGPA/MCLA binary backed by METIS / gpmetis / shmetis) or to a MATLAB session. These binaries are not bundled with the package; the user must supply and configure them. The path to the executable is selected through the executable argument. When executable=FALSE (the default), or when the configured executable cannot be located, the function stops with an informative error rather than attempting to launch an unavailable program.

Usage

EnsembleClustering(
  List,
  type = c("data", "dist", "clust"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  nrclusters = c(7, 7),
  gap = FALSE,
  maxK = 15,
  ensembleMethod = c("CSPA", "HGPA", "MCLA", "Best"),
  waitingtime = 300,
  file_number = 0,
  executable = FALSE
)

Arguments

Value

The returned value is a list of two elements:

The value has class 'Ensemble'.

References

Strehl A. and Ghosh J. (2002). Cluster ensembles - a knowledge reuse framework for combining multiple partitions. Journal of Machine Learning Research, 3, 583-617.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List

# Requires an external graph-partitioning executable / MATLAB to be
# configured through the 'executable' argument.
res <- EnsembleClustering(List = L, type = "data",
  distmeasure = c("euclidean", "euclidean"), normalize = c(FALSE, FALSE),
  method = c(NULL, NULL), clust = "agnes",
  linkage = c("flexible", "flexible"), nrclusters = c(7, 7),
  gap = FALSE, maxK = 15, ensembleMethod = "CSPA",
  executable = "./EnsembleClusteringC")

## End(Not run)

Description

EvidenceAccumulation builds a co-association (similarity) matrix from the individual source partitions and extracts a consensus partition by graph partitioning. Three graph-partitioning options are available: a minimum-spanning-tree procedure ("MST", requires the igraph package), a thresholded single-link procedure ("SL"), and an agnes single-link procedure ("SL_agnes").

Usage

EvidenceAccumulation(
  List,
  type = c("data", "dist", "clust"),
  distmeasure = c("tanimoto", "tanimoto"),
  normalize = c(FALSE, FALSE),
  method = c(NULL, NULL),
  clust = "agnes",
  linkage = c("flexible", "flexible"),
  alpha = 0.625,
  nrclusters = c(7, 7),
  gap = FALSE,
  maxK = 15,
  graphPartitioning = c("MST", "SL", "SL_agnes"),
  t = NULL
)

Arguments

Details

The co-association matrix S records the fraction of source partitions in which two objects are co-clustered. "MST" prunes a minimum spanning tree of the similarity graph (optionally at threshold t); "SL" performs a thresholded single-link grouping; and "SL_agnes" runs agnes single linkage on the dissimilarity 1-S. The "MST" path requires the suggested package igraph.

Value

A list of two elements:

The value has class 'Ensemble' (or 'Single Clustering' for "SL_agnes").

References

Fred A.L.N. and Jain A.K. (2005). Combining multiple clusterings using evidence accumulation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 27(6), 835-850.

Examples

## Not run: 
data(mosaic_toy)
L <- mosaic_toy$List
res <- EvidenceAccumulation(List = L, type = "data",
  distmeasure = c("euclidean","euclidean"), nrclusters = c(5,5),
  graphPartitioning = "SL_agnes")

## End(Not run)

Description

Turns a row-stacked matrix of per-iteration cluster labels ($K \cdot R$ rows by $N$ columns) into a co-clustering dissimilarity and applies cluster::agnes. $S_{ij}$ is the co-clustering probability over iterations selecting both objects (B6). The co-clustering counts use the compiled C++ kernel count_coclustering_cpp (from src/mcpp_sampling.cpp) when available, otherwise a pure-R loop.

Usage

f.clustABC.MultiSource(
  res,
  numclust = NULL,
  dissimilarity = "mabc",
  linkage = "ward",
  alpha = 0.625,
  mds = FALSE
)

Arguments

Value

A list with DistM, Clust and (if numclust given) cut.

References

(Amaratunga et al. 2008)

See Also

M_ABCpp

Description

FeatSelection determines the characteristic features of a selection of objects or of a specific cluster across the methods. Binary features are evaluated with Fisher's exact test, continuous features with the t-test.

Usage

FeatSelection(
  List,
  Selection = NULL,
  binData = NULL,
  contData = NULL,
  datanames = NULL,
  nrclusters = NULL,
  topChar = NULL,
  sign = 0.05,
  fusionsLog = TRUE,
  weightclust = TRUE
)

Arguments

Value

A list with the found (top) characteristics of the feature data.

Examples

## Not run: 
data(fingerprintMat)
data(targetMat)

MCF7_F = Cluster(fingerprintMat,type="data",distmeasure="tanimoto",normalize=FALSE,
method=NULL,clust="agnes",linkage="flexible",gap=FALSE,maxK=55,StopRange=FALSE)

L=list(MCF7_F)
MCF7_FeatSel=FeatSelection(List=L,Selection=4,binData=list(fingerprintMat),
datanames=c("FP"),nrclusters=7,topChar=20)

## End(Not run)