Closes #104. This PR tests the use of the cr-like resolution in comparison to the cr-like-em resolution mode in alevin-fry using similar per cell QC metrics discussed previously in benchmarking.
The prep for this notebook included updating docker containers for alevin-fry and salmon to use alevin-fry v0.4.0 and salmon 1.5.1. Two single cell (SCPCR000126 and SCPCR000127) and two single nuclei (SCPCR000220 and SCPCR000221) samples were run using splici with selective alignment, and either cr-like or cr-like-em with USA mode. The difference between the two modes is that the cr-like-em mode resolves multi-mapped reads using an expectation maximization algorithm, while cr-like throws them out. I am comparing these samples to the same samples run through cellranger. The dataframes used as input for this script were generated using the benchmarking_generate_qc_df.R script.
From this analysis, cr-like looks as it did previously with similar numbers to cellranger when looking at shared cells and cr-like-em reporting slightly higher numbers. Unless there is a reason not to want to resolve multi-mapped reads, I think we could use cr-like-em moving forward.
Closes #104. This PR tests the use of the
cr-like
resolution in comparison to thecr-like-em
resolution mode in alevin-fry using similar per cell QC metrics discussed previously in benchmarking.The prep for this notebook included updating docker containers for alevin-fry and salmon to use alevin-fry v0.4.0 and salmon 1.5.1. Two single cell (SCPCR000126 and SCPCR000127) and two single nuclei (SCPCR000220 and SCPCR000221) samples were run using
splici
with selective alignment, and eithercr-like
orcr-like-em
with USA mode. The difference between the two modes is that thecr-like-em
mode resolves multi-mapped reads using an expectation maximization algorithm, whilecr-like
throws them out. I am comparing these samples to the same samples run through cellranger. The dataframes used as input for this script were generated using thebenchmarking_generate_qc_df.R
script.From this analysis,
cr-like
looks as it did previously with similar numbers to cellranger when looking at shared cells andcr-like-em
reporting slightly higher numbers. Unless there is a reason not to want to resolve multi-mapped reads, I think we could usecr-like-em
moving forward.