-
Notifications
You must be signed in to change notification settings - Fork 2
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
getting denoised counts #4
Comments
Hi, is the expression of these 4 genes significantly different from the remaining genes? May I see the histogram of the denoised gene expression data? Also it might be useful to see the umap visualization of the denoised counts. I think it might be a scaling issue. |
Hi, we fond the bug within the prediction function that caused the scaling issue, and fixed the code in the newest update. Please use the newest version of the code. |
Dear Peter, |
Oh you mean that the clusters are not separated in the umap visualization after denoising? It sounds more like an issue related to the training process than the scaling issue. I think the 4 gene is not the issue that causes the mixing of cell types. The model is not reconstructing/denoising the input successfully. Did you check the latent space and observe the cell type separation in the latent space? How about the reconstruction loss? If the cell types in the latent space are separated and reconstruction loss converges, then when you reconstruct the data it should at least show some cell type separation in the umap visualization. I see that you set the latent space to 15 because 8 doesn't work, this should not happen as 8 is generally more than enough to observe separation. Maybe you can try to set all the regularization to 0, using only the reconstruction loss for sanity check. It should definitely give you cell type separation for the denoised count. |
Dear Peter,
I am reporting an issue with the software. I've been running the workflow as suggested in the demo notebook and getting good results for the latent dimension representation, gene weights, and separation of the two condition values (treatment and tissue in my case). However, I'm encountering a problem when trying to get denoised counts.
Issue Description
When attempting to obtain denoised counts, I consistently end up with greater than 1 values for only a very few genes (4 out of 20,400). These 4 genes are also the most expressed genes in the matrix, while the rest of the values are all less than 1. I'm using raw counts as input and have tried several runs with different combinations, unfortunately without success.
Code and Setup
Here's my basic script for setting up and running the model:
Attempts to Get Denoised Counts
I've tried two approaches to get denoised counts:
1. Following the notebook:
2. Extracting from the original count matrix:
Both approaches result in only 4 genes having values greater than 1 in the denoised counts.
I would greatly appreciate any insights or suggestions on how to resolve this issue. Thank you for your time and assistance.
The text was updated successfully, but these errors were encountered: