Commands#

Five commands move reads, and three read no reads at all. That count is fixed: a sixth pipeline command needs a failing benchmark that the existing five cannot pass, because every flag and every stage is a thing that can disagree with another one.

The pipeline#

migec checkout  reads.fq.gz --bc-pattern '^NNNNNNNN' -o co/   # find and cut out the barcode
migec refine    co/S1.fq.gz                          -o ref/  # fix the errors IN the barcode
migec assemble  ref/S1.fq.gz                         -o asm/  # collapse each molecule

command	what it does	the number it decides
suggest	reads the barcode layout off the data	the pattern you paste into `checkout`
checkout	finds the pattern, cuts the barcode out, demultiplexes	how many reads carry a usable barcode
refine	corrects errors in the barcode itself, calls cells	how many molecules there were
assemble	collapses each molecule’s reads into one consensus	the sequence, and the quality it is allowed to claim
subsample	keeps every read of a fraction of the barcodes	how many molecules a smaller copy of the library holds

suggest and subsample are outside the pipeline in different directions: one runs before it to tell you what to type, the other cuts a library down to something a laptop can iterate on.

Reading the output#

command	what it does	reads no reads because
plot	draws twenty QC panels from the TSVs the stages wrote	it computes nothing; a figure that cannot be redrawn from a committed table will eventually disagree with the report
`sheet`	writes or explains a barcode table, lists the presets	it only ever touches the layout
`info`	prints what a `.mig` bucket or a tagged FASTQ contains	it reads headers

Every stage writes plain TSV next to its output and a .json summary beside that, so the figures and any downstream script see the same numbers the report printed.