search

gagneurlab / drop

Pipeline to find aberrant events in RNA-Seq data, useful for diagnosis of rare disorders
MIT License
129 stars 43 forks source link

add samples to model #436

Closed kdewispelaere closed 9 hours ago

kdewispelaere commented 1 year ago

Hi, This might be a naive question but I suppose it's not possible to add samples to an already built model? Thanks! (great package btw)

vyepez88 commented 1 year ago

Hi, not sure what you mean exactly. There is the possibility to add your samples to other count matrices.

kdewispelaere commented 1 year ago

So what I'm doing is I want to build a large model with drop. I'm currently using outrider and fraser in R seperately for testing. Say I built a large reference model already and the autoencoder steps have been done. If there is a new batch of counts, then I assume the existing calculated confounder corrections can't be adjusted to this new data but will have to be calculated from scratch based on the entire dataset, right?

vyepez88 commented 1 year ago

Right, with the current design, they'll have to be calculated from scratch. Even the optimal dimension of the latent space might be different.

kdewispelaere commented 1 year ago

Ah yes, ofcourse. What if the added samples would have to be masked anyway. Would it then be possible since these samples wouldn't be used during autoencoding anyways. Thank you for your time

vyepez88 commented 2 months ago

Hi, sorry for the super late reply. Unfortunately, we haven't had good experience with masking samples from the AE. The p-values of the genes of the masked sample are not normally distributed.