cozygene/bisque

leave-one-out cross validation

empiricalbayes opened this issue · 2 comments

empiricalbayes commented

Hello,

I wonder if you could provide the codes you used for leave-one-out cross validation in your simulations.

Thanks.

brandonjew commented

Hi @empiricalbayes, sure thing. Please note that this is an R script, not a TXT file, so it should be renamed (made TXT to comply with attachment requirements). This script is meant to be run from the command line with a single positional argument that is an integer between 1 and 5. This argument specifies the scaling factor to simulate (one of [0, 5, 10, 15, 20]). Note that this script takes quite a long time to run with CIBERSORT and bseq-sc comparisons. If commented out, Bisque and MuSiC should complete fairly quickly.

Arguments that should be noted/edited:

  • cibersort.path is the path to the CIBERSORT.R file (retrieved by following https://shenorrlab.github.io/bseqsc/vignettes/pages/installation.html)
  • seurat.rds is an RDS file containing the single-cell/single-nuc object processed by Seurat
  • seurat.marker.rds is an RDS file containing the output of Seurat::FindAllMarkers()
  • out.prefix is the prefix for all cell composition output CSV files (it will output the ground truth as {prefix}.loo.true.csv and a file for each method, scaling factor, and replication).

Here is the script: simulation_analysis.txt

Let me know if you run into any issues!

Thanks,
Brandon

  • 👍1