Bio.Cluster package¶
Module contents¶
Cluster Analysis.
The Bio.Cluster provides commonly used clustering algorithms and was designed with the application to gene expression data in mind. However, this module can also be used for cluster analysis of other types of data.
Bio.Cluster and the underlying C Clustering Library is described in M. de Hoon et al. (2004) https://doi.org/10.1093/bioinformatics/bth078

class
Bio.Cluster.
Node
¶ Bases:
_cluster.Node
A Node object describes a single node in a hierarchical clustering tree. The integer attributes ‘left’ and ‘right’ represent the two members that make up this node; the floating point attribute ‘distance’ contains the distance between the two members of this node.

class
Bio.Cluster.
Tree
¶ Bases:
_cluster.Tree
Hierarchical clustering tree.
A Tree consists of Nodes.

sort
(self, order=None)¶ Sort the hierarchical clustering tree.
Sort the hierarchical clustering tree by switching the left and right subnode of nodes such that the elements in the lefttoright order of the tree tend to have increasing order values.
Return the indices of the elements in the lefttoright order in the hierarchical clustering tree, such that the element with index indices[i] occurs at position i in the dendrogram.

cut
(self, nclusters=None)¶ Create clusters by cutting the hierarchical clustering tree.
Divide the elements in a hierarchical clustering result mytree into clusters, and return an array with the number of the cluster to which each element was assigned.
 Keyword arguments:
nclusters: The desired number of clusters.


Bio.Cluster.
kcluster
(data, nclusters=2, mask=None, weight=None, transpose=False, npass=1, method='a', dist='e', initialid=None)¶ Perform kmeans clustering.
This function performs kmeans clustering on the values in data, and returns the cluster assignments, the withincluster sum of distances of the optimal kmeans clustering solution, and the number of times the optimal solution was found.
 Keyword arguments:
data: nrows x ncolumns array containing the data values.
nclusters: number of clusters (the ‘k’ in kmeans).
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i,j]==0, then data[i,j] is missing.
weight: the weights to be used when calculating distances
transpose:  if False: rows are clustered;  if True: columns are clustered.
npass: number of times the kmeans clustering algorithm is performed, each time with a different (random) initial condition.
method: specifies how the center of a cluster is found:  method == ‘a’: arithmetic mean;  method == ‘m’: median.
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance;  dist == ‘b’: City Block distance;  dist == ‘c’: Pearson correlation;  dist == ‘a’: absolute value of the correlation;  dist == ‘u’: uncentered correlation;  dist == ‘x’: absolute uncentered correlation;  dist == ‘s’: Spearman’s rank correlation;  dist == ‘k’: Kendall’s tau.
initialid: the initial clustering from which the algorithm should start. If initialid is None, the routine carries out npass repetitions of the EM algorithm, each time starting from a different random initial clustering. If initialid is given, the routine carries out the EM algorithm only once, starting from the given initial clustering and without randomizing the order in which items are assigned to clusters (i.e., using the same order as in the data matrix). In that case, the kmeans algorithm is fully deterministic.
 Return values:
clusterid: array containing the number of the cluster to which each item was assigned in the best kmeans clustering solution that was found in the npass runs;
error: the withincluster sum of distances for the returned kmeans clustering solution;
nfound: the number of times this solution was found.

Bio.Cluster.
kmedoids
(distance, nclusters=2, npass=1, initialid=None)¶ Perform kmedoids clustering.
This function performs kmedoids clustering, and returns the cluster assignments, the withincluster sum of distances of the optimal kmedoids clustering solution, and the number of times the optimal solution was found.
 Keyword arguments:
distance: The distance matrix between the items. There are three ways in which you can pass a distance matrix: 1. a 2D Numerical Python array (in which only the leftlower part of the array will be accessed); 2. a 1D Numerical Python array containing the distances consecutively; 3. a list of rows containing the lowertriangular part of the distance matrix.
Examples are:
>>> from numpy import array >>> # option 1: >>> distance = array([[0.0, 1.1, 2.3], ... [1.1, 0.0, 4.5], ... [2.3, 4.5, 0.0]]) >>> # option 2: >>> distance = array([1.1, 2.3, 4.5]) >>> # option 3: >>> distance = [array([]), ... array([1.1]), ... array([2.3, 4.5])]
These three correspond to the same distance matrix.
nclusters: number of clusters (the ‘k’ in kmedoids)
npass: the number of times the kmedoids clustering algorithm is performed, each time with a different (random) initial condition.
initialid: the initial clustering from which the algorithm should start. If initialid is not given, the routine carries out npass repetitions of the EM algorithm, each time starting from a different random initial clustering. If initialid is given, the routine carries out the EM algorithm only once, starting from the initial clustering specified by initialid and without randomizing the order in which items are assigned to clusters (i.e., using the same order as in the data matrix). In that case, the kmedoids algorithm is fully deterministic.
 Return values:
clusterid: array containing the number of the cluster to which each item was assigned in the best kmeans clustering solution that was found in the npass runs;
error: the withincluster sum of distances for the returned kmeans clustering solution;
nfound: the number of times this solution was found.

Bio.Cluster.
treecluster
(data, mask=None, weight=None, transpose=False, method='m', dist='e', distancematrix=None)¶ Perform hierarchical clustering, and return a Tree object.
This function implements the pairwise single, complete, centroid, and average linkage hierarchical clustering methods.
 Keyword arguments:
data: nrows x ncolumns array containing the data values.
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i][j]==0, then data[i][j] is missing.
weight: the weights to be used when calculating distances.
transpose:  if False, rows are clustered;  if True, columns are clustered.
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
method: specifies which linkage method is used:  method == ‘s’: Single pairwise linkage  method == ‘m’: Complete (maximum) pairwise linkage (default)  method == ‘c’: Centroid linkage  method == ‘a’: Average pairwise linkage
distancematrix: The distance matrix between the items. There are three ways in which you can pass a distance matrix: 1. a 2D Numerical Python array (in which only the leftlower part of the array will be accessed); 2. a 1D Numerical Python array containing the distances consecutively; 3. a list of rows containing the lowertriangular part of the distance matrix.
Examples are:
>>> from numpy import array >>> # option 1: >>> distance = array([[0.0, 1.1, 2.3], ... [1.1, 0.0, 4.5], ... [2.3, 4.5, 0.0]]) >>> # option 2: >>> distance = array([1.1, 2.3, 4.5]) >>> # option 3: >>> distance = [array([]), ... array([1.1]), ... array([2.3, 4.5])]
These three correspond to the same distance matrix.
PLEASE NOTE: As the treecluster routine may shuffle the values in the distance matrix as part of the clustering algorithm, be sure to save this array in a different variable before calling treecluster if you need it later.
Either data or distancematrix should be None. If distancematrix is None, the hierarchical clustering solution is calculated from the values stored in the argument data. If data is None, the hierarchical clustering solution is instead calculated from the distance matrix. Pairwise centroidlinkage clustering can be performed only from the data values and not from the distance matrix. Pairwise single, maximum, and averagelinkage clustering can be calculated from the data values or from the distance matrix.
Return value: treecluster returns a Tree object describing the hierarchical clustering result. See the description of the Tree class for more information.

Bio.Cluster.
somcluster
(data, mask=None, weight=None, transpose=False, nxgrid=2, nygrid=1, inittau=0.02, niter=1, dist='e')¶ Calculate a SelfOrganizing Map.
This function implements a SelfOrganizing Map on a rectangular grid.
 Keyword arguments:
data: nrows x ncolumns array containing the data values;
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i][j]==0, then data[i][j] is missing.
weight: the weights to be used when calculating distances
transpose:  if False: rows are clustered;  if True: columns are clustered.
nxgrid: the horizontal dimension of the rectangular SOM map
nygrid: the vertical dimension of the rectangular SOM map
inittau: the initial value of tau (the neighborbood function)
niter: the number of iterations
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
Return values:
clusterid: array with two columns, with the number of rows equal to the items that are being clustered. Each row in the array contains the x and y coordinates of the cell in the rectangular SOM grid to which the item was assigned.
celldata: an array with dimensions [nxgrid, nygrid, number of columns] if rows are being clustered, or [nxgrid, nygrid, number of rows) if columns are being clustered. Each element [ix, iy] of this array is a 1D vector containing the data values for the centroid of the cluster in the SOM grid cell with coordinates [ix, iy].

Bio.Cluster.
clusterdistance
(data, mask=None, weight=None, index1=None, index2=None, method='a', dist='e', transpose=False)¶ Calculate and return the distance between two clusters.
 Keyword arguments:
data: nrows x ncolumns array containing the data values.
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i, j]==0, then data[i, j] is missing.
weight: the weights to be used when calculating distances
index1: 1D array identifying which items belong to the first cluster. If the cluster contains only one item, then index1 can also be written as a single integer.
index2: 1D array identifying which items belong to the second cluster. If the cluster contains only one item, then index2 can also be written as a single integer.
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
method: specifies how the distance between two clusters is defined:  method == ‘a’: the distance between the arithmetic means of the two clusters  method == ‘m’: the distance between the medians of the two clusters  method == ‘s’: the smallest pairwise distance between members of the two clusters  method == ‘x’: the largest pairwise distance between members of the two clusters  method == ‘v’: average of the pairwise distances between members of the two clusters
transpose:  if False: clusters of rows are considered;  if True: clusters of columns are considered.

Bio.Cluster.
clustercentroids
(data, mask=None, clusterid=None, method='a', transpose=False)¶ Calculate and return the centroid of each cluster.
The clustercentroids routine calculates the cluster centroids, given to which cluster each item belongs. The centroid is defined as either the mean or the median over all items for each dimension.
 Keyword arguments:
data: nrows x ncolumns array containing the data values.
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i, j]==0, then data[i, j] is missing.
clusterid: array containing the cluster number for each item. The cluster number should be nonnegative.
method: specifies whether the centroid is calculated from the arithmetic mean (method == ‘a’, default) or the median (method == ‘m’) over each dimension.
 transpose: if False, each row contains the data for one item;
if True, each column contains the data for one item.
 Return values:
cdata: 2D array containing the cluster centroids. If transpose is False, then the dimensions of cdata are nclusters x ncolumns. If transpose is True, then the dimensions of cdata are nrows x nclusters.
cmask: 2D array of integers describing which items in cdata, if any, are missing.

Bio.Cluster.
distancematrix
(data, mask=None, weight=None, transpose=False, dist='e')¶ Calculate and return a distance matrix from the data.
This function returns the distance matrix calculated from the data.
 Keyword arguments:
data: nrows x ncolumns array containing the data values.
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i, j]==0, then data[i, j] is missing.
weight: the weights to be used when calculating distances.
 transpose: if False: the distances between rows are calculated;
if True: the distances between columns are calculated.
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
Return value: The distance matrix is returned as a list of 1D arrays containing the distance matrix calculated from the data. The number of columns in eac row is equal to the row number. Hence, the first row has zero length. For example:
>>> from numpy import array >>> from Bio.Cluster import distancematrix >>> data = array([[0, 1, 2, 3], ... [4, 5, 6, 7], ... [8, 9, 10, 11], ... [1, 2, 3, 4]]) >>> distances = distancematrix(data, dist='e') >>> distances [array([], dtype=float64), array([ 16.]), array([ 64., 16.]), array([ 1., 9., 49.])]
 which can be rewritten as
 distances = [array([], dtype=float64),
array([ 16.]), array([ 64., 16.]), array([ 1., 9., 49.])]
This corresponds to the distance matrix:
[ 0., 16., 64., 1.] [16., 0., 16., 9.] [64., 16., 0., 49.] [ 1., 9., 49., 0.]

Bio.Cluster.
pca
(data)¶ Perform principal component analysis.
 Keyword arguments:
data: nrows x ncolumns array containing the data values.
Return value: This function returns an array containing the mean of each column, the principal components as an nmin x ncolumns array, as well as the coordinates (an nrows x nmin array) of the data along the principal components, and the associated eigenvalues. The principal components, the coordinates, and the eigenvalues are sorted by the magnitude of the eigenvalue, with the largest eigenvalues appearing first. Here, nmin is the smaller of nrows and ncolumns. Adding the column means to the dot product of the coordinates and the principal components recreates the data matrix:
>>> from numpy import array, dot, amax, amin >>> from Bio.Cluster import pca >>> matrix = array([[ 0., 0., 0.], ... [ 1., 0., 0.], ... [ 7., 3., 0.], ... [ 4., 2., 6.]]) >>> columnmean, coordinates, pc, _ = pca(matrix) >>> m = matrix  (columnmean + dot(coordinates, pc)) >>> amax(m) < 1e12 and amin(m) > 1e12 True

class
Bio.Cluster.
Record
(handle=None)¶ Bases:
object
Store gene expression data.
A Record stores the gene expression data and related information contained in a data file following the file format defined for Michael Eisen’s Cluster/TreeView program.
 Attributes:
data: a matrix containing the gene expression data
mask: a matrix containing only 1’s and 0’s, denoting which values are present (1) or missing (0). If all items of mask are one (no missing data), then mask is set to None.
geneid: a list containing a unique identifier for each gene (e.g., ORF name)
genename: a list containing an additional description for each gene (e.g., gene name)
gweight: the weight to be used for each gene when calculating the distance
gorder: an array of real numbers indicating the preferred order of the genes in the output file
expid: a list containing a unique identifier for each sample.
eweight: the weight to be used for each sample when calculating the distance
eorder: an array of real numbers indication the preferred order of the samples in the output file
uniqid: the string that was used instead of UNIQID in the input file.

__init__
(self, handle=None)¶ Read gene expression data from the file handle and return a Record.
The file should be in the format defined for Michael Eisen’s Cluster/TreeView program.

treecluster
(self, transpose=False, method='m', dist='e')¶ Apply hierarchical clustering and return a Tree object.
The pairwise single, complete, centroid, and average linkage hierarchical clustering methods are available.
 Keyword arguments:
 transpose: if False: rows are clustered;
if True: columns are clustered.
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
method: specifies which linkage method is used:  method == ‘s’: Single pairwise linkage  method == ‘m’: Complete (maximum) pairwise linkage (default)  method == ‘c’: Centroid linkage  method == ‘a’: Average pairwise linkage
See the description of the Tree class for more information about the Tree object returned by this method.

kcluster
(self, nclusters=2, transpose=False, npass=1, method='a', dist='e', initialid=None)¶ Apply kmeans or kmedian clustering.
This method returns a tuple (clusterid, error, nfound).
 Keyword arguments:
nclusters: number of clusters (the ‘k’ in kmeans)
 transpose: if False, genes (rows) are clustered;
if True, samples (columns) are clustered.
npass: number of times the kmeans clustering algorithm is performed, each time with a different (random) initial condition.
method: specifies how the center of a cluster is found:  method == ‘a’: arithmetic mean  method == ‘m’: median
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
initialid: the initial clustering from which the algorithm should start. If initialid is None, the routine carries out npass repetitions of the EM algorithm, each time starting from a different random initial clustering. If initialid is given, the routine carries out the EM algorithm only once, starting from the given initial clustering and without randomizing the order in which items are assigned to clusters (i.e., using the same order as in the data matrix). In that case, the kmeans algorithm is fully deterministic.
 Return values:
clusterid: array containing the number of the cluster to which each gene/sample was assigned in the best kmeans clustering solution that was found in the npass runs;
error: the withincluster sum of distances for the returned kmeans clustering solution;
nfound: the number of times this solution was found.

somcluster
(self, transpose=False, nxgrid=2, nygrid=1, inittau=0.02, niter=1, dist='e')¶ Calculate a selforganizing map on a rectangular grid.
The somcluster method returns a tuple (clusterid, celldata).
 Keyword arguments:
 transpose: if False, genes (rows) are clustered;
if True, samples (columns) are clustered.
nxgrid: the horizontal dimension of the rectangular SOM map
nygrid: the vertical dimension of the rectangular SOM map
inittau: the initial value of tau (the neighborbood function)
niter: the number of iterations
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
 Return values:
clusterid: array with two columns, while the number of rows is equal to the number of genes or the number of samples depending on whether genes or samples are being clustered. Each row in the array contains the x and y coordinates of the cell in the rectangular SOM grid to which the gene or samples was assigned.
celldata: an array with dimensions (nxgrid, nygrid, number of samples) if genes are being clustered, or (nxgrid, nygrid, number of genes) if samples are being clustered. Each item [ix, iy] of this array is a 1D vector containing the gene expression data for the centroid of the cluster in the SOM grid cell with coordinates [ix, iy].

clustercentroids
(self, clusterid=None, method='a', transpose=False)¶ Calculate the cluster centroids and return a tuple (cdata, cmask).
The centroid is defined as either the mean or the median over all items for each dimension.
 Keyword arguments:
data: nrows x ncolumns array containing the expression data
mask: nrows x ncolumns array of integers, showing which data are missing. If mask[i, j]==0, then data[i, j] is missing.
 transpose: if False, gene (row) clusters are considered;
if True, sample (column) clusters are considered.
clusterid: array containing the cluster number for each gene or sample. The cluster number should be nonnegative.
method: specifies how the centroid is calculated:  method == ‘a’: arithmetic mean over each dimension. (default)  method == ‘m’: median over each dimension.
 Return values:
cdata: 2D array containing the cluster centroids. If transpose is False, then the dimensions of cdata are nclusters x ncolumns. If transpose is True, then the dimensions of cdata are nrows x nclusters.
cmask: 2D array of integers describing which items in cdata, if any, are missing.

clusterdistance
(self, index1=0, index2=0, method='a', dist='e', transpose=False)¶ Calculate the distance between two clusters.
 Keyword arguments:
index1: 1D array identifying which genes/samples belong to the first cluster. If the cluster contains only one gene, then index1 can also be written as a single integer.
index2: 1D array identifying which genes/samples belong to the second cluster. If the cluster contains only one gene, then index2 can also be written as a single integer.
 transpose: if False, genes (rows) are clustered;
if True, samples (columns) are clustered.
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
method: specifies how the distance between two clusters is defined:  method == ‘a’: the distance between the arithmetic means of the two clusters  method == ‘m’: the distance between the medians of the two clusters  method == ‘s’: the smallest pairwise distance between members of the two clusters  method == ‘x’: the largest pairwise distance between members of the two clusters  method == ‘v’: average of the pairwise distances between members of the two clusters
 transpose: if False: clusters of rows are considered;
if True: clusters of columns are considered.

distancematrix
(self, transpose=False, dist='e')¶ Calculate the distance matrix and return it as a list of arrays.
 Keyword arguments:
 transpose:
if False: calculate the distances between genes (rows); if True: calculate the distances between samples (columns).
dist: specifies the distance function to be used:  dist == ‘e’: Euclidean distance  dist == ‘b’: City Block distance  dist == ‘c’: Pearson correlation  dist == ‘a’: absolute value of the correlation  dist == ‘u’: uncentered correlation  dist == ‘x’: absolute uncentered correlation  dist == ‘s’: Spearman’s rank correlation  dist == ‘k’: Kendall’s tau
Return value:
The distance matrix is returned as a list of 1D arrays containing the distance matrix between the gene expression data. The number of columns in each row is equal to the row number. Hence, the first row has zero length. An example of the return value is:
 matrix = [[],
array([1.]), array([7., 3.]), array([4., 2., 6.])]
This corresponds to the distance matrix:
[0., 1., 7., 4.] [1., 0., 3., 2.] [7., 3., 0., 6.] [4., 2., 6., 0.]

save
(self, jobname, geneclusters=None, expclusters=None)¶ Save the clustering results.
The saved files follow the convention for the Java TreeView program, which can therefore be used to view the clustering result.
 Keyword arguments:
jobname: The base name of the files to be saved. The filenames are jobname.cdt, jobname.gtr, and jobname.atr for hierarchical clustering, and jobnameK*.cdt, jobnameK*.kgg, jobnameK*.kag for kmeans clustering results.
geneclusters: For hierarchical clustering results, geneclusters is a Tree object as returned by the treecluster method. For kmeans clustering results, geneclusters is a vector containing ngenes integers, describing to which cluster a given gene belongs. This vector can be calculated by kcluster.
expclusters: For hierarchical clustering results, expclusters is a Tree object as returned by the treecluster method. For kmeans clustering results, expclusters is a vector containing nexps integers, describing to which cluster a given sample belongs. This vector can be calculated by kcluster.

Bio.Cluster.
read
(handle)¶ Read gene expression data from the file handle and return a Record.
The file should be in the file format defined for Michael Eisen’s Cluster/TreeView program.