visxhclust: visual exploration of hierarchical clustering

visxhclust is a package that includes a Shiny application for visual exploration of hierarchical clustering. It is aimed at facilitating iterative workflows of hierarchical clustering on numeric data. For that, the app allows users to quickly change parameters and analyse and evaluate results with typical heatmaps with dendrograms and other charts. Additionally, it includes lightweight data overview plots such as correlation heatmaps, annotated MDS and PCA plots. On the evaluation side, it builds on existing packages to compute internal validation scores and Gap statistic, as well as Dunn’s test to evaluate significant differences between clusters.

The app includes multiple help points in the interface and a vignette documenting how to load data and the basic clustering loop – see vignette("visxhclust"). The package exports various functions to help with documenting and reproducing a clustering workflow with R or R Markdown – see vignette("clusterworkflow") and vignette("clusterevaluation") for examples.

Installation

The latest release can be installed from CRAN:

install.packages("visxhclust")

The latest development version can be installed from GitHub:

remotes::install_github("rhenkin/visxhclust")

Most dependencies are found in CRAN. However, the heatmap drawing package is part of Bioconductor and may require a separate installation:

install.packages("BiocManager")
BiocManager::install("ComplexHeatmap")

Usage and data requirements

To use your data with the tool, you can save a data frame or tibble in an RDS file, or use comma or tab-delimited files, with .csv, .tsv or .txt extensions. The clustering method supported by the tool works only on numeric values; columns containing text will be set aside to annotate the heatmap if so desired. If a column named ID exists, it will be used as an internal identifier for rows.

Clustering requires complete datasets with no missing values, NULLs or NAs. If any column contains missing values, it will be set aside to be used as a heatmap annotation. Badly formatted data will also lead to unexpected results in the tool. As an alternative, imputation packages can be used to fill missing data and faulty rows (e.g. text in numeric columns) should be removed before loading the file into the tool. The tool provides limited abilities to help with diagnosing issues and preprocessing data.

To run the app once the package is installed:

library(visxhclust)
# Increases max file size to 30 MB
options(shiny.maxRequestSize = 30*1024^2)
run_app()