Predicted gene expression is off

[Jorges1000]

Hi, @jianhuupenn,

I was able to successfully run the tutorial to completion, thanks for a great program. When using our own data, I was able to get the contours for one of the tissue sections in the image:

However, after applying Tesla inputation, the predicted gene expression is completely off:

Wondering if there is an easy explanation and fix for this?

Many thanks!

[jianhuupenn]

Thanks for your interest in our method. It's the first time I've seen such an issue. The gene expression prediction is incorrect. Since TESLA performs imputation based on original expression, the enhanced pattern should preserve the original pattern. Here are a few potential reasons you can check:

1. The original and enhanced gene expressions may not be from the same gene. I would double-check the variable name columns in ADDdata; there may be some mismatches.
2. Try subsetting the gene expression data to only include spots from the identified bottom-left region to check if this issue is due to multiple tissues on one slide.

[Jorges1000]

Thanks for your quick reply. The gene names are correct:

>>> counts.var.loc['MUC5AC']
genenames    MUC5AC
Name: MUC5AC, dtype: object

I subsetted the object to just the one tissue, and also flip the x and y, but it seems that Tesla created more or less linear patterns for every gene with poor correlation with histology or original expression: