Consider using UMAP embedding method

lgatto / pRoloc

A unifying bioinformatics framework for organelle proteomics

http://lgatto.github.io/pRoloc/

15 stars 14 forks source link

Consider using UMAP embedding method #111

Open lgatto opened 6 years ago

lgatto commented 6 years ago

See https://github.com/ropenscilabs/umapr

ococrook commented 5 years ago

Following up on this a newer more reliable verion is available here: https://github.com/tkonopka/umap

lmsimp commented 5 years ago

umap is now available in CRAN

## Install and load umap package
install.packages("umap")
library("umap")

## Load example dataset
library("pRolocdata")
data("hyperLOPIT2015")

## create UMAP visualisation
hl.umap <- umap(exprs(hyperLOPITU2OS2018))

## Pass projections matrix to plot2D 
xx <- (hl.umap$layout)
plot2D(xx, fcol = "markers", method = "none", 
       methargs = list(hyperLOPITU2OS2018), 
       main = "UMAP - hyperLOPIT2018")

lgatto commented 5 years ago

@ococrook - any opinion on these different R packages:

ropenscilabs/umapr.
tkonopka/umap, including R and python implementations - this package also allows to project data onto an existing embedding, which is a useful feature.
jlmelville/uwot.

ococrook commented 5 years ago

umap seems the most reliable - already on CRAN etc - well tested. Likely to be well developed in the future. The code looks better written as well.

lgatto commented 4 years ago

I'm closing this issue but happy to get a PR that add UMAP.

lmsimp commented 3 weeks ago

I have written code to add UMAP as an option to plot2D and tested some default parameters that work well with quantitative proteomics data. This uses the umap function from the umap package in CRAN. which Olly suggested several years ago. I still think this looks like the best option.

I was thinking it would be nice to have a vignette dedicated to plot2D and how to customise plots. Will open a separate issue for this to discuss.

Plan to submit a PR for UMAP shortly if you are still happy for a PR. Will also wait to hear from you regarding #152

CC @Charl-Hutchings