Open abspangler13 opened 2 years ago
Hi Abby, yes I have seen this happen before in other data as well. In such cases, BayesSpace's model finds better partitions of the data using fewer clusters. The spatial smoothing BayesSpace applies penalizes neighboring spots with different cluster labels, so if you have two clusters (in your initialization) that are fairly similar, BayesSpace may combine them.
One way to get around this potentially is by reducing the value of the smoothing parameter gamma
. Trying a different initialization may also help here since you have a pretty large dataset.
Thank for the info Edward! Thanks for the confirmation that you've seen this before and for the info on gamma
.
I think that we'll be ok with k_obs <= k_input
. Abby, I think that you can close this issue.
Best, Leo
Hello,
I ran
BayesSpace
with k = 28 for ~113,000 spots. I've noticed that I only get 26 clusters because clusters 18 and 21 are missing. Is this to be expected? Is it possible,BayesSpace
is merging small clusters?Thanks,
Abby cc @lcolladotor
R Session Information