Hi!
Just wondering, is there any (rough estimate is fine) guideline for minimum cluster size that will still produce generally clear and reliable results?
Just to narrow down if there’s any other clusters I should look at a bit more carefully after using SCVI for annotation.
Unfortunately there’s not a straightforward answer to this question. Manual clustering/annotation is a bit more of an art than a science. You might receive more relevant responses on Scanpy’s forum, as this problem is not unique to any of the models in scvi-tools.
Speaking from rule of thumb experience, if you got >100 cells in a group where you are performing comparisons it should be fine. Sometimes even as low as 20 or 50, but 100 is good.