next-generation sequencing
/ NGS /
Sanger sequencing reads one DNA template at a time. Reading a whole human genome that way took an international consortium more than a decade and billions of dollars. The obvious question was: what if you could read not one template, but millions of templates at the same time, in parallel? That idea — massively parallel sequencing — is what 'next-generation sequencing' means, and it collapsed the cost and time of reading DNA by orders of magnitude.
Rather than one careful reaction in one tube, NGS chops the DNA into a huge number of short fragments, attaches them to a surface or beads, and reads all of them simultaneously — typically by sequencing by synthesis, where each added base flashes a detectable signal. A single run produces hundreds of millions to billions of short reads at once. The catch is that each individual read is short (often 50 to 300 bases) and somewhat error-prone, so the power comes not from any one read but from sheer numbers: the same region is read many times over, and the consensus of all those reads is highly reliable. Software then stitches or maps the flood of short reads back into genes and genomes.
NGS turned molecular biology into a data science. The first human genome cost on the order of billions of dollars; NGS brought a human genome down toward the cost of a routine medical test, opening up sequencing whole genomes, just the protein-coding exome, transcriptomes (RNA-seq), tumour DNA, microbial communities, and more. The honest framing: NGS is a triumph of scale and parallelism, not of read quality — it trades short, error-prone individual reads for overwhelming throughput, and it leans heavily on computation to turn raw reads into answers.
An NGS run sequences a person's genome by reading roughly a billion short fragments at once, each only about 150 bases long. No single read covers a whole gene, but because every position is read 30 or more times over, software can assemble a confident, complete genome from the overlap.
Power from millions of parallel short reads, not one perfect read.
NGS reads are individually short and error-prone — accuracy comes from reading each spot many times (coverage), not from any single read, and the raw output is useless without heavy computational assembly or mapping.