Closed mcsimenc closed 1 year ago
In the example you gave, the error is because scSHC
expects a matrix of non-negative integer counts. In principle, rounding the normalized data shouldn't cause an error in the code (one quick check would be to make sure that after rounding, you truly do have all non-negative integer counts), but I wouldn't recommend this strategy regardless because our approach is intended to operate on the raw counts without any kind of normalization, integration, or other processing. Instead, in this case, you should just use the raw counts (e.g. in the Seurat object, this should be slot = counts
).
Thank you for clarifying @igrabski. Does your model account for differences in sequencing depth among barcodes? This, as well as multiplet abundance, affects results from other clustering procedures.
Thanks for your question -- yes, our procedure accounts for sequencing depth, which is why it is not necessary to normalize beforehand.
That's great. What are your thoughts on multiplet inclusion? It seems possible to me that any multiplets present could affect which genes are chosen as highly variable. Is there a way to have scSHC
use specific genes?
Yes, that's true, multiplet inclusion could skew your results. Currently, we use an automated gene selection procedure, but I would recommend removing columns correspond to suspected multiplets (as determined by any procedure) prior to running our approach.
Greetings!
I have been unable to run
scSHC
, getting different errors depending on whether I use thebatch
option. I have a Seurat analysis of snRNA-seq data.scSHC
worked using the example sparse matrixcounts
fromcounts.rda
.The scSHC documentation says
data
should be a raw counts matrix or Matrix. I am trying normalized data but also tried using the matrix with values rounded up and saw the same error.Any idea what might be causing this?
Thanks!