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. Many of the functions are also exported to facilitate documenting a complete analysis cycle.

NEW! A live demo of the app is running here.

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")

Getting started

To run the app once the package is installed, use the following commands:

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

The app includes multiple help points in the interface (look for the question marks), and there are also three guides on how to use tool:

Usage tips 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.

Contributing

Please see the guide for code contribution and suggestions.