Nobody reads a genome from beginning to end — what a sequencing machine actually does

The picture most people carry is of a machine threading a chromosome through a reader, letter by letter, from one end to the other. Nothing about real sequencing looks like that. The genome is chopped into millions of short fragments, each fragment is read on its own with no record of where it came from, and the order is recovered afterwards by software comparing overlaps. What comes out of the instrument is not a book. It is confetti, and a great deal of what is confusing about genomics — why some variants are missed, why "the complete human genome" was announced twice twenty years apart, why your report has a category called uncertain — follows directly from that one design decision.

Why shred it at all

Because no chemistry can read a molecule two hundred million bases long without stopping. Every sequencing method has a read length — the number of bases it can call before accuracy collapses — and the largest is still thousands of times shorter than a single chromosome.

So the approach, called shotgun sequencing, works around the limit statistically. Take not one copy of the genome but millions of copies, from millions of cells. Break them all at random. Because the breaks fall in different places in different copies, the fragments overlap each other. Read enough of them and every position in the genome is covered by many fragments from many different starting points, and those overlaps are what let the original order be reconstructed.

That is why sequencing is quoted as coverage depth — "30×" means each base was, on average, present in thirty independent reads. Depth is not padding. It is how you tell a real variant from a chemistry error: a single read disagreeing with twenty-nine others is noise, while eight reads out of thirty carrying the same change is a signal worth acting on. Cancer sequencing often runs at several hundred times depth precisely because the mutation may exist in only a small fraction of the cells sampled.

Sanger: the method that still decides arguments

The original approach, published by Frederick Sanger's group in 1977, is beautifully indirect. Copy the DNA fragment with an enzyme, but poison the reaction slightly with dideoxynucleotides — chemically sabotaged letters that can be added to a growing strand but cannot be built on. Every time one is incorporated, that copy stops dead.

Run this on millions of copies and you get every possible truncation length: chains ending at position 1, at 2, at 3, and so on. Separate them by size, read off which terminating letter sits at the end of each length, and the sequence falls out. Modern versions use four fluorescent colours and capillary electrophoresis, and deliver 500 to 1,000 clean bases per run.

Sanger is slow and cannot scale to a genome, but it never went away. It remains the routine confirmation step: when a clinical laboratory finds something consequential in a high-throughput run, it very often re-checks that single site by Sanger before anyone acts on it.

What a modern flow cell is doing

The instruments that made genomes cheap use sequencing by synthesis, run in massive parallel. Fragments are fixed to a glass flow cell and locally amplified, so each spot becomes a cluster of identical copies bright enough to photograph. Then the machine cycles: flood the cell with the four bases, each carrying a fluorescent tag and a chemical block that permits exactly one addition; wash; photograph the entire surface; cleave the tags and blocks; repeat.

Every cycle adds one base to every cluster simultaneously, and the image records which colour appeared at each of hundreds of millions of positions. A hundred and fifty cycles later you have a hundred and fifty bases from every one of those clusters. The read is short — the chemistry loses phase after a few hundred cycles — but the parallelism is enormous, and that is the trade the whole field was built on.

Long-read platforms make the opposite trade. Nanopore sequencing drives a single DNA strand through a protein pore in a membrane and measures the tiny disruptions in ionic current as each stretch of bases passes through, calling the sequence from the current trace. PacBio reads a circular template repeatedly and takes the consensus. Both are less accurate per base than short reads were at their peak, and both routinely return reads tens of thousands of bases long. That length is not a luxury — it is the only way past the problem in the next section.

Short reads tell you what letters are present. Long reads tell you how the genome is arranged. Most of the things a clinical test still misses are arrangement, not letters.

Repeats, and the twenty-year gap

Reassembly is a jigsaw where many pieces are identical. Large stretches of the human genome are repetitive — the same motif tandemly duplicated thousands of times, especially around centromeres and telomeres. A 150-base read that lands entirely inside such a region matches ten thousand places equally well, so it carries no information about where it belongs.

This is why the Human Genome Project's "completion" in 2003 was not complete. Roughly 8 per cent of the sequence — the hardest, most repetitive 8 per cent — remained unresolved, and stayed that way for two decades. It was only in 2022 that the Telomere-to-Telomere consortium published a genuinely gapless human genome, and the thing that made it possible was long reads capable of spanning an entire repeat array and anchoring it to unique sequence at both ends.

There is a second, quieter limitation. Nearly all human sequencing is done by alignment: reads are mapped against a standard reference genome rather than assembled from scratch, because alignment is vastly cheaper. But the reference is itself a particular sequence from a small number of donors, and it under-represents most of the world's populations. Variation common in one ancestry and absent from the reference aligns poorly or not at all. That is why the Human Pangenome Reference Consortium moved in 2023 to a reference built from many diverse genomes instead of one — and why national efforts such as GenomeIndia, which set out to sequence ten thousand genomes across the country's population groups, and CSIR's earlier IndiGen programme matter scientifically rather than just symbolically. A variant database with almost no Indian genomes in it will keep returning "unknown" for Indian patients.

Why a clean report can still miss something

Three different things get called "getting your DNA tested", and they look at very different amounts of it.

A gene panel sequences a chosen set of genes very deeply — the right tool when the clinical question is specific. An exome covers the roughly 1 to 2 per cent of the genome that codes for protein, which is where most known disease-causing variants sit, and ignores the rest. A whole genome covers everything, including the regulatory sequence between genes that is increasingly where unexplained cases turn out to hide.

Even at whole-genome depth, three gaps persist. Short reads detect single-letter changes and small insertions well, but detect structural variants — large deletions, duplications, inversions — poorly, because those are arrangement problems. Consumer genotyping arrays are weaker still: they check a pre-selected list of positions and are blind by construction to anything not on the list. And a large fraction of what is found is classified as a variant of uncertain significance — the sequencing worked perfectly, and nobody yet knows what that change does. That is a gap in the literature, not in the machine, and it closes only as more people are sequenced and followed up.

Why it matters for students and researchers

Sequencing has stopped being the bottleneck. A genome that cost billions and took over a decade in the 1990s now costs a few hundred dollars and runs in a day or two, which means the hard part has moved decisively downstream — into alignment and assembly, variant calling, filtering, population reference design, and the statistics of deciding that a change is causal rather than merely present.

That shift is why genomics is now as much a computational discipline as a wet-lab one, and why the people in demand are those who can do both: run the chemistry and understand what the pipeline does to the data afterwards. Anyone entering bioengineering, molecular biology or biotechnology today will spend more career time on read alignment, coverage statistics and variant interpretation than on the sequencer itself — and the researchers who understand where the reads came from are the ones who catch the artefacts that a pipeline reports with a straight face.

Frequently asked questions

How does DNA sequencing actually work?

The genome is broken into millions of short overlapping fragments, each fragment is read independently by chemistry that identifies bases one at a time, and software reconstructs the original order from the overlaps. No instrument reads a chromosome continuously from end to end.

What is the difference between Sanger and next-generation sequencing?

Sanger sequencing reads one fragment at a time, delivering 500 to 1,000 highly accurate bases, and is still used to confirm individual findings. Next-generation sequencing reads hundreds of millions of short fragments simultaneously on a flow cell, which is what makes whole genomes affordable.

What does 30× coverage mean?

It means each position in the genome was covered by about thirty independent reads on average. Depth is how a real variant is distinguished from a random sequencing error, so higher depth gives more confidence — and cancer testing often uses several hundred times coverage.

Why are long reads important if short reads are more accurate?

Because repetitive regions defeat short reads entirely: a short read inside a long repeat matches thousands of locations equally well. Long reads span whole repeat arrays and detect large structural changes, which is how the first gapless human genome was finally completed in 2022.

Is a whole genome sequence better than a gene panel?

Not always. A panel sequences fewer genes at much greater depth and is often the better answer to a specific clinical question, while a whole genome gives breadth and catches regulatory variants. Breadth also produces far more findings of uncertain significance, which need expert interpretation.