search

dpeerlab / ENVI

MIT License
34 stars 3 forks source link

Issue with latent space embedding with new version of ENVI #65

Open rmathieu25 opened 3 weeks ago

rmathieu25 commented 3 weeks ago

Hello,

Thank you for developing ENVI, I think it's a really promising tool!

Nevertheless, I have tried out the first version of ENVI with Tensorflow, and the UMAP in the latent space looked quite good (see image attached). However, when plotting the UMAP in the latent space with exactly the same dataset (single-cell RNA-seq and MERSCOPE), the UMAP in the latent space is not really convincing, as the cell-types are barely overlapping.

What do you think could be causing the issue? ENVI_version_comparison

Thank you very much!

ccruizm commented 2 weeks ago

Good day! I am having the same problem as @rmathieu25. The new version does not give overlapping dimensional reduction between scRNA and ST. Do you have any insights why this is occurring?

CleanShot 2024-06-21 at 11 26 32@2x

Thanks in advance!

ccruizm commented 2 days ago

In addition to this issue, I also noticed that the imputed genes are only a fraction of the genes in the SC reference. Do you know if this behavior is expected?