cluster_isoforms: Initial clustering of isoforms using dynamicTreeCut

In ConesaLab/acorde: Isoform co-usage networks from single-cell RNA-seq data

Initial clustering of isoforms using dynamicTreeCut

Description

Create groups of isoforms with similar expression across cell types. Based on co-expression values and dynamic hierarchical clustering with the dynamicTreeCut package. Ultimately, this function provides a custom wrapper for the cutreeHybrid function.

Usage

cluster_isoforms(
  cor_matrix,
  deepSplit = 3,
  pamStage = FALSE,
  minClusterSize = 20,
  ...
)

Arguments

cor_matrix	A matrix including co-expression values (e.g. correlations) between isoforms, where column and row names indicate isoform IDs.
deepSplit	Either logical or integer int he range 0 to 4 (defaults to 3). Provides control over sensitivity to cluster splitting (higher value creates more clusters with a smaller number of elements).
pamStage	A logical indicating whether a second clustering step similar to Partition Around Medioids is to be performed. If pamStage = TRUE is set, no isoforms will remain unassigned to clusters. By default, we allow unclustered elements by setting pamStage = FALSE.
minClusterSize	An integer defining minimum cluster size allowed during dynamic clustering. Defaults to 20.
...	Additional parameters supplied to cutreeHybrid.

Details

First, a previously obtained co-expression matrix (for instance, percentile correlations generated using percentile_cor) is transformed into a distance matrix for hierarchical clustering. As the acorde pipeline intends to detect positively correlated isoforms, negative correlation values are automatically discarded by replacing them with zero values. Distance is computed as 1 - |co-expression|.

Next, hierarchical clustering is performed using the hclust function. Using the output dendrogram, isoforms are clustered via the cutreeHybrid function in the dynamicTreeCut package.

Value

A list of generated clusters, that is, of character vectors including the identifiers of isoforms assigned to each of the clusters.

Note: if PAM = FALSE, the first element of the list will correspond to "cluster zero", i.e. a group containing isoforms that were not assigned to any of the clusters.

References

\insertRef

Langfelder2008acorde

\insertRef

Venables2002acorde

See Also

For more details on parameters, refer to cutreeHybrid.

ConesaLab/acorde documentation built on Feb. 25, 2024, 4:16 a.m.