1

When I run the same R code in my local computer RStudio (R 4.0.2) and on Code Ocean R 4.0.3, I have two different UMAP visualization results and they are mirrored

[enter image description here]

enter image description here

I use Seurat 3.2.0 version in both environments and particularly for umap visualization, here is the line:

DimPlot(MU197PDXThgFiltered, reduction = "umap", label = TRUE, label.size = 6, pt.size = 0.5)

I can't give a reproducible example, but maybe someone faced this issue before?

Yulia Kentieva
  • 641
  • 4
  • 13
  • Did you set a seed (e.g. `set.seed(123)` or `seed.use = 123` within the RunUMAP function) before running the `Dimplot()` function? – jared_mamrot Apr 15 '21 at 03:27
  • yes, I used set.seed(100) before clustering. Still almost mirrored results – Yulia Kentieva Apr 15 '21 at 03:52
  • Yeah, they aren't exactly mirrored - the clusters are slightly different e.g. 6, 0, 8 - I think the best way to get an answer on 'why' they're different is to raise an issue on github (https://github.com/satijalab/seurat/issues) or contact someone from Rahul's lab to see if they can provide advice. Also might be worth asking on https://bioinformatics.stackexchange.com/ where you might find others with the same issue – jared_mamrot Apr 15 '21 at 03:52
  • thank you. I found a comment from them that UMAP can differ depending on OS – Yulia Kentieva Apr 15 '21 at 04:31

1 Answers1

3

I found this on Seurat github:

The exact location of points on a UMAP plot can chance across different computers and OSs. We do our best to minimize any randomness to the procedure by fixing the random seed, but some fluctuation across systems is inevitable, and nothing to worry about

Other than point locations on UMAP, gene expressions in clusters and cell numbers in them are identical

Yulia Kentieva
  • 641
  • 4
  • 13