search

KrishnaswamyLab / MAGIC

MAGIC (Markov Affinity-based Graph Imputation of Cells), is a method for imputing missing values restoring structure of large biological datasets.
GNU General Public License v2.0
347 stars 96 forks source link

Use MAGIC on SEURAT integrated DATA #179

Closed DomenicoSkyWalker89 closed 4 years ago

DomenicoSkyWalker89 commented 4 years ago

Hi Scott, i have tre different data set (T0 ,T1, T2) and i used to analize this data the suggested SEURAT workflow (https://satijalab.org/seurat/v3.1/immune_alignment.html). I would use MAGIC to denoise my data, in which step of workflow su suggest to use magic? Because after integration i have two assay (RNA and integrated) and i don't understand which one to use or if there is something wrong on what i doing.

kind regard, Domenico

scottgigante commented 4 years ago

Hi @DomenicoSkyWalker89 ,

Typically we use MAGIC as the input to things like plotting gene-gene relationships and colouring dimensionality reductions by gene expression, as well as (with some care) ranking gene-gene interactions, differential expression etc. We don't typically use MAGIC as the input to dimensionality reduction.

I've tried running the tutorial you linked to, and I noticed that if you run MAGIC on the "integrated" assay, there are essentially no differences between the treatment and control; the integration is removing the biological effect. I would probably try running the entire workflow, including MAGIC, on the uncorrected data (i.e. the "RNA" assay) and seeing how things look, especially how the dimensionality reduction looks. 

# load data
...
immune.combined <- RunUMAP(immune.combined)
library(Rmagic)
DefaultAssay(immune.combined) <- "RNA"
immune.combined <- magic(immune.combined)
DefaultAssay(immune.combined) <- "MAGIC_RNA"
# do the rest of your analysis
...

If it looks like you have a strong batch effect, you could try correcting it (with Seurat or otherwise) and then running MAGIC, but be warned that at least in my experience this has a high likelihood of removing some biology too.

# load data
...
immune.combined <- RunUMAP(immune.combined)
library(Rmagic)
DefaultAssay(immune.combined) <- "integrated"
immune.combined <- magic(immune.combined)
DefaultAssay(immune.combined) <- "MAGIC_integrated"
# do the rest of your analysis
...
DomenicoSkyWalker89 commented 4 years ago

Hi @scottgigante , Thanks a lot. I will try what you suggest!

Best, Domenico

zqun1 commented 3 years ago

@DomenicoSkyWalker89 Hi, I am facing the same issue. Any experience to share? Thanks!

DomenicoSkyWalker89 commented 3 years ago

Hi @zqun1, I performed the denoising on RNA assay as suggested by Scott and all ran perfectly. I did not noticed improvement of the data and for that i did not included this step within my final workflow(https://www.nature.com/articles/s41467-021-21928-4). I also used saverX with the same results.

Best, Domenico

zqun1 commented 3 years ago

@DomenicoSkyWalker89 I see. May I know what you used it for and how you decide there was no improvement? By the way, your linked paper is very interesting. Can you email me for further discussion qun.zeng@kuleuven.be ?

I am trying MAGIC to reduce dimensionality. I have integrated datasets. Without MAGIC, my clusters were very diffused on UMAP. I am trying MAGIC on "integrated" assay and interestingly, I saw the clusters were less diffused! But I am not sure if I am doing the correct thing as Scott said: How do you think @scottgigante ?

We don't typically use MAGIC as the input to dimensionality reduction.

DomenicoSkyWalker89 commented 3 years ago

Well, first I plotted the heatmap before and after denoising and then the distribution of single gene expression like showed in this vignette(http://htmlpreview.github.io/?https://github.com/KrishnaswamyLab/MAGIC/blob/master/Rmagic/inst/examples/bonemarrow_tutorial.html). Then i also tested if the DE genes were improved or not.

Can you share the results of UMAP embedded with clusters before and after MAGIC? Have you take into account cell cycle or mitocondrial genes etc.. as other source of variability?

ps email me for further discussion at 206560@studenti.unimore.it

Best, Domenico