Sequencing and variants - FASTQ to a filtered call set
Lesson 3 of 6 · Your instrument, end to end
1 · Learn the move · Agents + multi-step
PARALLEL TRACK - complete only if pl1 pointed you here. Sequencing is the size-defeats-reading pipeline: a FASTQ is millions of reads, a BAM tens of gigabytes, so the model reasons over summaries while scripts do the moving. FastQC and MultiQC summaries stand in for the FASTQ; the BAM is known through samtools flagstat and coverage; the VCF through bcftools stats; then normalization and filtering with every threshold stated. Three traps own the discipline: reference build - GRCh37 and GRCh38 are not interchangeable and nothing in the file warns you; multi-allelics must be normalized before counting; and BED is 0-based half-open while VCF is 1-based - the off-by-one survives every eyeball check. Two absolute boundaries: human-subject sequence data needs your institution's policy and IRB terms before anything pastes - flag and refer. And a variant list is never clinical guidance.
Author my variant pipeline as staged steps with a checkpoint after each: (1) FastQC/MultiQC on [FASTQ set], model sees the summary only; (2) alignment to [reference, build stated] producing sorted BAM; checkpoint = flagstat, mapped percentage vs my floor of [X]; (3) calling with [caller]; checkpoint = bcftools stats; (4) bcftools norm for multi-allelics, then filters, every threshold printed; (5) final call-set summary with all decisions logged. My data is [human / non-human] - if human, the first step is confirming my institution's policy covers this workflow, and nothing pastes into chat regardless.
2 · Your turn. You write the prompt
Forty exome FASTQs from a mouse cross just landed from the core, and your last attempt at this died in a swamp of half-remembered samtools flags. Build the staged pipeline where every stage hands the model a summary, every threshold is printed, and the whole run reconstructs from the log.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.