What a sequencer actually gives you

What a sequencer gives you

Not a genome —a pile of fragments

Nothing reads a chromosome end to end. You get millions of short pieces, each with a confidence score, and the genome is something you reconstruct.

Reads, not sequences

Fragments of a few hundred bases, or tens of thousands on a long-read platform. Never a whole chromosome.

Every base has a score

A Phred quality per position. Q30 means one wrong call in a thousand, and that is the usual target.

Redundancy is the method

You sequence the same region many times over, and confidence comes from the reads agreeing.

The generations

Three technologies,three trade-offs

Read length against accuracy against cost. Every sequencing platform sits somewhere on that triangle, and none of them wins outright.

SangerLong, extremely accurate, one fragment at a time. Still the reference standard for validating a single locus.

Short read — IlluminaHundreds of millions of 100–300 base reads at very high accuracy. Cheapest per base, by a wide margin.

Long read — PacBio, NanoporeTens of kilobases per read. Resolves repeats and structural variants that short reads simply cannot span.

The practical consequence is that long and short reads are complements, not competitors. A common assembly today uses long reads for structure and short reads to polish the base-level accuracy, which is why the two file types keep turning up in the same project directory.

Two routes

Align to a reference,or assemble from nothing

Everything downstream forks here, and the fork is decided by whether a good reference genome exists for your organism.

Alignment

Place each read against a known genome with BWA or minimap2, then call the differences. Fast and cheap.

De novo assembly

Reconstruct the genome from overlaps alone. Memory-hungry, slow, and the only route without a reference.

Reference bias

Alignment can only find what the reference lets you see. Novel insertions are invisible to it by construction.

Alignment is cheap, well-tooled and answers most questions about a well-studied species. Assembly is expensive in both compute and memory, and it is the only option when the reference does not exist — or when the thing you are looking for is precisely what the reference does not contain.

Coverage, then the pipeline

Coverage first,then four settled steps

Depth of coverage governs what you can conclude. After that, the shape of a reference-based project has been stable for a decade.

30×Whole genomethe usual target for variant calling

100×+Exome, or cancersmall targets, or rare subclones

UnevenAlwaysaverage depth hides thin regions

Sequencing is a sampling process, so coverage is uneven by nature — Poisson at best, and worse wherever GC content is extreme. Thirty-fold average depth on a human genome still leaves regions at five, which is why a caller reports low confidence somewhere in every run. Deciding the depth you need before you sequence is far cheaper than discovering it afterwards.

a short-read pipeline

$ fastqc -t 8 -o qc/ reads/*.fastq.gz

$ bwa mem -t 8 ref.fa r1.fq.gz r2.fq.gz > aln.sam

$ samtools sort -@ 8 -o aln.bam aln.sam && samtools index aln.bam

1 · Quality controlFastQC over the raw reads, then trimming if adapters or a bad tail show up. Before anything expensive.

2 · AlignBWA-MEM for short reads, minimap2 for long. The output is SAM, and it is always enormous.

3 · Sort, index, deduplicateSAMtools turns the aligner output into something that every downstream tool will accept.

4 · Call and lookA caller produces a VCF. Then open the interesting loci in IGV, because callers fail in known ways.

On a machine

FastQC isalready built

Nothing here needs compiling. Pick a machine, and these two commands put it on there.

hub://fastqc

$ dxflow workflow create --identity fastqc hub://fastqc

$ dxflow workflow start fastqc

Start a machine What FastQC needs

Read it, then run itPrepaid and by the hour, on a machine that is yours about a minute after you ask.

Start a machine All notes