Can Ergen

icon company University of California Berkeley icon location Berkeley icon location Computational Biology PostDoc at UC Berkeley and Weizman Institute of Science Clinician Scientist at University Hamburg

Results 98 comments of Can Ergen

Issues installing recommended packages

Closed due to inactivity.

Fix for get_normalized_expression

Hi Laura, LGTM. Sorry for the late reply. I think it would make things clearer to define mu=rate within the reconstruction losses, so every reconstruction loss has a "mu". I...

Support for weighted loss either per-batch or per-cell?

Hi, thanks for your question. @martinkim0 we wanted to have a stratified validation/train splitter at some time point but never followed up on this. I guess this would yield a...

Support for weighted loss either per-batch or per-cell?

You can now used stratified train splitters in main (#2902, it will be released in scvi-tools 1.2). You can define a train set in which small cell-types are more abundant...

How to modify scANVI to handle continuous cell labels?

The structure of scANVI relies on categorical covariates specifically the z1-z2 loss. I would start with scVI and add the continuous classifier using something like MSE loss. You might want...

Expected value argument (Parameter of shape (7150,)) to be within the support (Real()) of the distribution Normal(loc: torch.Size([7150]), scale: torch.Size([7150])), but found invalid values: Parameter containing: tensor([nan, nan, nan, ..

Hi, this is usually a problem with very low count input spots. It makes sense to remove these low count input spots before running the spatial model (below 20 counts...

Expected value argument (Parameter of shape (7150,)) to be within the support (Real()) of the distribution Normal(loc: torch.Size([7150]), scale: torch.Size([7150])), but found invalid values: Parameter containing: tensor([nan, nan, nan, ..

It's very hard to get rid of this behaviour in DestVI as mentioned above. In our hands these spots are usually anyhow that low in quality. We are working on...

Expected value argument (Parameter of shape (7150,)) to be within the support (Real()) of the distribution Normal(loc: torch.Size([7150]), scale: torch.Size([7150])), but found invalid values: Parameter containing: tensor([nan, nan, nan, ..

One is an updated DestVI designed for data like Curio that are closer to single-cell resolution. For StereoSEQ, we don't use any single-cell data but only spatial structure. I feel...

Expected value argument (Parameter of shape (7150,)) to be within the support (Real()) of the distribution Normal(loc: torch.Size([7150]), scale: torch.Size([7150])), but found invalid values: Parameter containing: tensor([nan, nan, nan, ..

@parkjosh-broadinstitute this is usually dataset dependent. We find that using a Softplus activation function for the variance stabilizes this behavior. Pass `var_activation=torch.nn.functional.softplus` when you set up your initial scVI model....

Expected value argument (Parameter of shape (7150,)) to be within the support (Real()) of the distribution Normal(loc: torch.Size([7150]), scale: torch.Size([7150])), but found invalid values: Parameter containing: tensor([nan, nan, nan, ..

@Rafael-Silva-Oliveira the updated DestVI is working under: https://github.com/scverse/scvi-tools/tree/can_destvi_v2. It will. be released latter this year.