First set up your Zorn/Bascet working directory as before. If you wish to run these steps on a SLURM cluster, see separate vignette and adapt accordingly.
library(Zorn)
bascet_runner.default <- LocalRunner(direct = TRUE, show_script=TRUE)
bascetRoot <- "/home/yours/an_empty_workdirectory"The first step in data processing is take the raw FASTQ files and identify which read belongs to which cell. Zorn can automatically figure out which files to use for what:
rawmeta <- DetectRawFileMeta("/home/yours/directory_with_raw_fastq")TODO explain this data frame
You can now parse the reads the figure out which cell they belong to, and any UMI information. The only thing you additionally need to specify is what type of data you have, where “atrandi_wgs” is for the Atrandi SPC-based MDA protocol for WGS.
BascetGetRaw(
bascetRoot,
rawmeta,
chemistry="atrandi_wgs"
)Once barcodes have been processed, you can perform knee plot analysis, which is the first important quality control.
### Decide cells to include
h <- ReadHistogram(bascetRoot,"debarcoded")
PlotHistogram(h)This is just a histogram of reads per barcode. It however tells you * how many cells you have * how much free-floating RNA/DNA you have in your background (empty droplets/SPCs) * how distinct cells are from background
It makes sense to exclude barcodes if the number of counts are too low. You can exclude them also at a later stage, but removing the worst offenders already will speed up computation. Here is one example for getting the names and number of cells having at least 100 reads:
includeCells <- h$cellid[h$count>100]
length(includeCells)Finally, before you are ready to run all other commands, you need to make sure that cells are sharded. This is primarily relevant if you have multiple FASTQ-files for your library (i.e. you sequenced your library on multiple lanes). Sharding sorts cells such that the reads for one cell are in one single shard. This is essential for parallel processing.
At this step you can also choose to create multiple shards. This is only relevant if you wish to parallelize the workflow using SLURM, as the number of shards decides how many nodes you can use in parallel. If you run everything on your own computer then we advise only creating one shard, as there is no benefit to having multiple and it is more likely to cause mistakes.
In this step we also filter out barcodes with too low read counts (optional).
BascetShardify(
bascetRoot,
includeCells = includeCells,
num_output_shards = 1
)Additional trimming
Bascet only does rudimentary trimming. We recommend trimming the reads further. To do this, first convert your data to FASTQ format:
### Get reads in fastq format
BascetMapTransform(
bascetRoot,
inputName="filtered", #default; can omit
outputName="asfq",
out_format="R1.fq.gz"
)Now use FASTP to perform the trimming:
### Get reads in fastq format for BWA
BascetRunFASTP(
bascetRoot,
numLocalThreads=10,
inputName="asfq",
outputName="fastp"
)If your plan is to perform alignment, then the resulting fastp FASTQ files can be used as input (don’t forget to specify inputName=“fastp”!). If you want to instead get a TIRP file, you can convert it back:
BascetMapTransform(
bascetRoot,
"fastp",
"new_filtered",
out_format="tirp.gz"
)You now have a TIRP file again, equivalent to the first shardified output. Don’t forget to specify inputName=“new_filtered” to later commands, as the default is to use “filtered”