Can we use Tile Matrix from ArchR as input? Thanks!
Yes, any binary (or close-to-binary) matrix can be used as input. We don’t currently have a direct interface with ArchR projects, but you can extract the matrix using getMatrixFromProject
.
We recommend that you filter the matrix to remove tiles that are very rarely observed (e.g detected in fewer than 1% of cells) since those generally increase the time to train the model without adding useful information.
Thanks for your quick answer and suggestions, Tal!
Hi Tal, I tried peakVI using tiled matrix from Arch and it worked. I have a question on pvi.differential_accessibility. For my dataset of 51319 × 27838, one versus all others have only <20 chromosome regions using is_da_fdr. I wonder if I can use a different criteria and select a bit more chromosome regions as da in one versus all others. Thanks, Tao
Hi Tao,
Glad it worked! For the differential accessibility, two things I would try:
- Try running the analysis in
vanilla
mode (mode="vanilla"
). This basically set the minimal difference required to 0, which might help the analysis recover more significant results. - If the “all” group in a “vs-all” comparison contains many cells that are highly similar to cells in the target group (for instance, a cluster that is very similar to the target cluster), this can reduce the ability of the analysis to accurately identify differential features. This If possible - try comparing against “all” except highly similar clusters.