Next-generation sequencing (NGS) reads millions of DNA fragments at the same time, rather than one fragment at a time like older methods. This massively parallel approach is what makes it fast enough and cheap enough to sequence an entire human genome in days. The process has several distinct stages: preparing the DNA, amplifying it, reading the sequence one base at a time, and then using software to assemble all the fragments into a meaningful result.
How NGS Differs From Sanger Sequencing
Traditional Sanger sequencing reads a single DNA fragment per reaction. That works fine if you’re looking at one gene or confirming a known mutation, but it becomes painfully slow and expensive when you need to examine hundreds or thousands of genes. A typical Sanger run produces 50 to 100 reads per sample.
NGS flips this by sequencing millions of fragments simultaneously in a single run, generating tens to hundreds of thousands of reads per sample. That difference in throughput is what opened the door to whole-genome sequencing, large cancer gene panels, and population-scale studies. It also means NGS can sequence up to 1,000 times more genes in a single assay than Sanger, making it far more cost-effective for anything beyond about 20 targets.
Step 1: Library Preparation
Before any sequencing happens, the DNA sample has to be broken into small, uniform pieces. This fragmentation can be done mechanically (using sound waves) or chemically (using enzymes). The ideal fragment size depends on the specific sequencing platform and application, but it typically falls in the range of a few hundred base pairs.
Once fragmented, short synthetic DNA sequences called adapters are attached to both ends of every fragment. These adapters serve as handles: they let the fragments stick to the sequencing surface and provide a starting point for the sequencing reaction. On Illumina platforms, these are known as P5 and P7 adapters. The adapters are platform-specific, so an Illumina library can’t be run on a different manufacturer’s machine without re-preparation. The finished library is a collection of DNA fragments, all roughly the same length, each with adapters on both ends.
Step 2: Clonal Amplification
A single DNA molecule produces too faint a signal for the sequencer to detect. To solve this, each fragment is copied thousands of times before sequencing begins, creating dense clusters of identical molecules.
Illumina uses a process called bridge amplification. Individual DNA fragments bind to primers attached to a glass surface called a flow cell. Each bound fragment bends over and attaches to a neighboring primer, forming a bridge. The bridge is then copied, the strands separate, and the process repeats. After many cycles, each original fragment has generated a tight cluster of identical copies, all anchored to the same small spot on the flow cell.
Other platforms use emulsion PCR instead. Here, individual DNA fragments are attached to tiny beads and sealed inside microscopic oil droplets, each droplet acting as its own miniature test tube. After around 120 PCR cycles, each bead is coated with thousands of clonal copies of the original fragment. The key requirement for both methods is the same: every cluster or bead must contain copies of just one original molecule, keeping the signal clean and unambiguous.
Step 3: Sequencing by Synthesis
This is where the actual reading happens. On Illumina platforms, the most widely used approach, sequencing works by building a complementary strand one base at a time and watching which base gets added at each step.
All four DNA building blocks (A, T, C, G) are washed over the flow cell at once, each labeled with a different fluorescent color. Each building block also carries a chemical blocker that prevents more than one base from being added per cycle. After a base is incorporated into the growing strand, a camera captures the color of the fluorescent signal at every cluster on the flow cell. Then the fluorescent label and blocker are chemically removed, and the next cycle begins.
Because all four bases are present during every cycle, there’s natural competition for the binding site. The correct base wins out based on normal DNA chemistry, which minimizes errors from biased incorporation. This cycle of “add one base, photograph, remove the label” repeats hundreds of times, building up the sequence read one letter at a time.
Step 4: Base Calling and Quality Scoring
The raw output from the sequencer isn’t text. It’s millions of images, each showing the fluorescent signals from every cluster on the flow cell. Software converts these images into actual DNA sequences through a process called base calling.
For Illumina, base calling works by aligning each image to a template of cluster positions, extracting the fluorescence intensity at each cluster, correcting for signal overlap between neighboring clusters, and assigning a base to each position. Each base also gets a quality score, known as a Q score, representing the probability that the call is correct. A Q30 score, a common industry benchmark, means there’s a 1-in-1,000 chance that a given base call is wrong. On newer Illumina chemistry, around 71% to 87% of bases meet or exceed this threshold.
The output at this stage is typically stored in FASTQ files, which contain both the sequence of each read and the quality score for every base.
Step 5: Alignment and Variant Calling
Millions of short reads are useless on their own. They need to be assembled into something interpretable, and that’s where bioinformatics takes over.
First, each read is aligned to a reference genome. A mapping algorithm finds the most likely position in the reference that matches the read, tolerating a small number of mismatches so that genuine differences between the sample and the reference aren’t thrown out. This alignment step produces SAM or BAM files, which store each read along with its position in the reference genome.
Next comes variant calling: identifying the positions where the sample’s DNA differs from the reference. Some variant callers work by simply counting how many high-quality reads disagree with the reference at a given position. More sophisticated tools use statistical methods, including Bayesian probability or machine learning, weighing factors like base quality and mapping confidence to distinguish real variants from sequencing errors. The final list of variants is stored in a VCF (Variant Call Format) file, which records each variant’s position, the reference base, and the alternative base found in the sample.
Where NGS Is Used in Medicine
NGS has become essential in clinical settings where multiple genes need to be examined at once. Cancer diagnosis is a prime example: tumors often carry mutations across many genes simultaneously, and traditional single-gene tests can’t efficiently cover that ground. NGS panels can screen hundreds of cancer-related genes in a single run, guiding treatment decisions about targeted therapies.
In clinical genetics, NGS is used more often for inherited (constitutional) genetic diseases than for cancer. It’s especially valuable when a genetic condition is suspected but no specific mutation has been identified through other testing. In those cases, whole-genome or whole-exome sequencing can uncover disease-causing mutations that would otherwise go undetected.
Long-Read Sequencing: The Next Step
Standard NGS platforms like Illumina produce short reads, typically a few hundred bases long. This works well for counting sequences, measuring gene activity, and identifying small mutations. But short reads struggle with repetitive regions of the genome and large structural changes like big insertions, deletions, or rearrangements.
Third-generation platforms from PacBio and Oxford Nanopore take a fundamentally different approach. PacBio’s technology immobilizes individual DNA molecules in tiny wells and watches a polymerase incorporate bases in real time, reading tens of thousands of bases in a single continuous stretch. These platforms skip the amplification step entirely, which eliminates the copying errors that PCR can introduce. It also means they can detect chemical modifications to DNA, like methylation, directly during sequencing, something short-read platforms can’t do without additional preparation steps.
Short-read and long-read technologies aren’t competing so much as complementary. Short reads remain the workhorse for applications where high accuracy and massive throughput matter most, while long reads excel at resolving complex structural features and assembling genomes from scratch.