https://github.com/yhoogstrate/recursivecorplot
Recursive correlation based natural clustering (using ggplot, labels and historgrams)
https://github.com/yhoogstrate/recursivecorplot
Last synced: 5 months ago
JSON representation
Recursive correlation based natural clustering (using ggplot, labels and historgrams)
- Host: GitHub
- URL: https://github.com/yhoogstrate/recursivecorplot
- Owner: yhoogstrate
- License: gpl-3.0
- Created: 2022-01-20T13:06:48.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2024-07-10T11:22:47.000Z (11 months ago)
- Last Synced: 2024-07-11T12:18:49.142Z (11 months ago)
- Language: R
- Size: 4.99 MB
- Stars: 3
- Watchers: 2
- Forks: 1
- Open Issues: 1
-
Metadata Files:
- Readme: README.Rmd
- License: LICENSE
Awesome Lists containing this project
README
---
title: "recursiveCorPlot - natural clustering of RNA-seq data"
output:
github_document:
toc: true # table of content true
toc_depth: 3 # upto three depths of headings (specified by #, ## and ###)
---
---
[]() 
# Introduction
For classical hierarchical clustering of RNA-seq data, the use of Euclidean distances as distance metric often result in unnatural clusters. For example, if the clustering contains genes with only a few samples with strong up-regulation by hyper-amplifications, these will weigh heavily at the Euclidean distance(s). This distance metric is therefore sensitive to outliers. Instead, correlation based clustering (`distance = 1 – correlation(m)`) is more common for RNA-seq data, where spearman's rank can be used to more aggressively suppress outliers. We observed some genes, relatively rich in zero counts, of which the correlation to all other genes are somewhat lower, but the correlations consistently went in the same direction as other genes within a cluster. Since the directions of the correlation are consistent with other genes but the data didn't seem powerful enough, we took the correlation of the correlation as the distance metric: `distance = 1 – correlation(correlation(m))`. This distance metric was clustered hierarchically using the "ward.D2"" method, showing neat natural clusters.
# Installation
You can install *recursiveCorPlot* from Github using:
```{r, eval = FALSE}
devtools::install_github("yhoogstrate/recursiveCorPlot")
```
# Citing recursiveCorPlot
Please cite this paper when using *recursiveCorPlot* for your publications:
```
Youri Hoogstrate, Kaspar Draaisma, Santoesha A. Ghisai, Levi van Hijfte, Nastaran Barin, Iris de Heer, Wouter Coppieters, Thierry P.P. van den Bosch, Anne Bolleboom, Zhenyu Gao, Arnaud J.P.E. Vincent, Latifa Karim, Manon Deckers, Martin J.B. Taphoorn, Melissa Kerkhof, Astrid Weyerbrock, Marc Sanson, Ann Hoeben, Slávka Lukacova, Giuseppe Lombardi, Sieger Leenstra, Monique Hanse, Ruth E.M. Fleischeuer, Colin Watts, Nicos Angelopoulos, Thierry Gorlia, Vassilis Golfinopoulos, Vincent Bours, Martin J. van den Bent, Pierre A. Robe, Pim J. French,
Transcriptome analysis reveals tumor microenvironment changes in glioblastoma,
Cancer Cell,
2023,
ISSN 1535-6108,
https://doi.org/10.1016/j.ccell.2023.02.019
```
## Usage
| Example with G-SAM DE Genes: |
|---------------------------------------------|
| `data('G.SAM.corrected.DE.genes.VST', package = 'recursiveCorPlot')` |
| 
|
| Above: recursive correlation based clustering |
| |
| 
|
| Above: regular 1 - correlation based clustering |
| |
| 
|
| Above: scaled Euclidean distance based clustering |