Long-read sequencing

Short-read sequencing produces very accurate fragments of a hundred or so bases. Its limitation is not accuracy but ambiguity: a read shorter than a repeated element cannot be placed uniquely, so any feature longer than the read is inferred rather than observed. Long-read platforms trade per-base accuracy for reads spanning kilobases to megabases, and that trade is what resolves the ambiguity.

Two different physical measurements

Nanopore sequencing drives a single DNA strand through a protein pore under an applied voltage and records the ionic current, which is modulated by the several bases occupying the constriction at any moment. A basecaller — in practice a neural network — converts the current trace back into sequence. Because the signal depends on a window of bases, homopolymer runs are the classic difficulty: five identical bases and six give nearly the same current, so the errors are characteristically insertions and deletions in low-complexity sequence. Read length is limited chiefly by how long a molecule survives extraction and loading intact.

PacBio’s single-molecule real-time sequencing instead watches a polymerase incorporate fluorescently labelled nucleotides in a zero-mode waveguide, a well small enough that only the enzyme’s active site is illuminated. Individual incorporation events are noisy, but the template is circularised, so the polymerase reads the same molecule repeatedly. Averaging those passes yields circular consensus reads whose accuracy approaches that of short reads while the read remains long — accuracy here is bought with polymerase passes rather than with read length.

Raw versus consensus accuracy

The distinction matters when reading a specification. Raw per-read accuracy describes a single molecule; consensus accuracy describes the result after combining evidence, either multiple passes over one molecule or coverage from many molecules over one locus. Both platforms now report consensus figures competitive with short reads at reasonable depth. Where raw accuracy still governs is single-molecule work — detecting a variant present in a small fraction of molecules, or resolving something seen only once.

What length buys

Repeats: assemblies stop fragmenting when reads exceed the repeat length, which is how complete telomere-to-telomere human assembly became possible, including segmental duplications and centromeric arrays that short reads simply could not traverse.

Structural variation: insertions, deletions, inversions and translocations of kilobase scale are read directly within a single molecule rather than inferred from anomalous read-pair distances, which is why long reads find substantially more structural variants per genome than short-read pipelines.

Phasing: a read spanning two heterozygous sites shows directly which alleles are on the same chromosome, so haplotypes are observed rather than statistically imputed from population data.

Modified bases: both platforms sense base modification without bisulfite treatment — nanopore because methylation alters the current signal, PacBio because it alters polymerase kinetics — so methylation is read in the same run as sequence, on the same molecule, with phase information attached.

What limits it

Long reads require long intact molecules, so extraction becomes the sensitive step; degraded clinical or environmental material caps read length regardless of platform. Depth costs more per base than short reads, and the analysis stack is different rather than merely retuned. The reasonable summary is that long reads answer questions of structure and phase, and short reads remain efficient for counting.