The Evolution of DNA Sequencing: Chemical Kinetics, Signal Processing, and Multi-Platform Strategic Selection

Meta Intent: A deep technical resource for scientists who need to compare sequencing platforms through chemistry, signal formation, error behavior, and run-efficiency logic.

DNA sequencing is often introduced as a timeline of platforms. That view is too flat for real project design. A sequencing system is not just a machine that reads bases. It is a chain of physical events. Molecules are captured, amplified or isolated, interrogated by light or current, converted into signal, and then decoded under noise. Read length, throughput, error type, and cost-per-base all emerge from that chain.

That is why the most useful question is no longer “What is DNA sequencing?” The better question is “What physically limits each platform?” In short-read systems, the answer usually starts with synchrony, cluster behavior, and cyclic chemistry. In long-read systems, it starts with single-molecule detection, signal modeling, and correction strategy. In both cases, sequencing performance is shaped long before downstream analysis begins.

This article follows that logic. It starts with second-generation sequencing by synthesis, because short-read sequencing still sets the benchmark for dense parallel data generation. It then builds toward third-generation platforms, where signals are read from single molecules through optical confinement or ionic-current disruption. Finally, it turns to library complexity and run design, because many expensive sequencing failures begin before cycle 1.

For researchers planning whole genome sequencing projects, the main lesson is simple: platform outputs make more sense when you understand the chemistry and signal model that created them.

Why second-generation sequencing still matters

Short-read sequencing remains dominant because it combines mature chemistry with extreme parallelization. The common summary is “high accuracy, short reads, high throughput.” That summary is correct, but it hides the deeper reason for success. Sequencing by synthesis solved a coordination problem. Instead of reading one molecule at a time, it reads huge numbers of spatially separated features in repeated chemical cycles. Each cycle is designed to add one base, capture one signal, and reset for the next round.

That architecture matters because it makes the decoding problem easier. The instrument is not trying to infer an entire sequence from one continuous analog trace. It is asking a simpler question at each step: which base was incorporated in this cycle at this feature? That design is powerful. It is also fragile in one specific way. The platform depends on synchrony. If part of a cluster falls behind and another part moves ahead, the signal stops representing one clean molecular state.

This is the central tradeoff in SBS. It wins by imposing chemical order. It loses quality when that order begins to decay.

Sequencing by synthesis begins before sequencing

Many overviews begin the SBS story at nucleotide incorporation. In practice, the story starts earlier, with cluster generation on the flow cell. If cluster generation is unstable, later chemistry can perform well and the run can still underdeliver.

A flow cell is not just a passive surface. It is a spatial control system. Library fragments bind through adapter sequences and become the starting templates for local signal amplification. On non-patterned surfaces, bridge amplification creates clusters through repeated bending, hybridization, and extension. The fragment bridges to a nearby oligo, gets copied, denatured, and repeats the process until a dense clonal feature forms.

This design is elegant, but it depends on stochastic occupancy. If loading is too low, the flow cell wastes usable area. If loading is too high, neighboring clusters can overlap and become harder to resolve. In other words, cluster density is not just a throughput issue. It is also a signal-separation issue.

Patterned flow cells changed that geometry. Instead of allowing clusters to emerge wherever capture succeeds, they place nanowells at defined positions. Exclusion amplification, or ExAmp, then drives cluster formation at those predetermined sites. The gain is clear: more uniform feature spacing, better packing efficiency, stronger imaging registration, and higher total output. But the gain comes with a stricter demand on library quality. Patterned systems are less tolerant of cleanup problems, especially when free adapters or residual index-bearing species remain in the pool.

That is why library preparation cannot be treated as a warm-up step. A sequencing run is not only a readout problem. It is also a templating problem. If the pool contains adapter dimers, poorly normalized fragments, or a distorted size distribution, the instrument spends expensive chemistry cycles on avoidable artifacts. In sensitive multiplexed designs, that cost is not only lost throughput. It can also become sample misassignment risk.

For this reason, platform choice and library quality control should always be discussed together. A well-matched platform cannot rescue a poorly behaved library indefinitely.

Bridge amplification vs. ExAmp: the physics of cluster generation

Bridge amplification and ExAmp solve the same core problem in different ways. Both exist to create enough local copies from one starting fragment so that the resulting feature emits a strong, readable signal. The difference lies in how spatial exclusivity is enforced.

Bridge amplification relies on local surface proximity. A bound fragment bends toward a neighboring oligo, forms a bridge, and gets copied. The cycle repeats, and the cluster grows where it began. The advantage is conceptual simplicity. The weakness is geometric uncertainty. Cluster size and spacing emerge from local competition on an open surface.

ExAmp imposes more control. Patterned nanowells predefine where a cluster can form. The starting molecule seeds a known location, and amplification proceeds within that boundary. This improves packing density and reduces spatial ambiguity between neighboring features.

The practical consequence is important. Throughput is not just a property of optics. It is also a property of geometry. Patterning lets an instrument place more readable features in the same physical area. That lowers cost per informative base when the library is clean. But geometry does not erase chemistry. If the pool still contains problematic small fragments or free indexed species, a denser system can expose that weakness rather than hide it.

This tradeoff is often missed in broad platform comparisons. They present throughput as a line item on a spec sheet. In reality, throughput is a negotiated outcome between surface design, loading behavior, library cleanliness, and downstream signal separability.

Figure 1. Sequencing by synthesis as a gated cycle: a surface-bound cluster undergoes nucleotide incorporation, fluorescence imaging, cleavage/deblocking, and re-extension in repeated synchronized rounds. This matters because the same cycle design that enables high-density short-read output also creates a dependence on chemical synchrony over time.

Reversible terminator chemistry: the real engine of SBS

At the center of SBS is reversible terminator chemistry. This is the point where the platform stops being a generic polymerase assay and becomes a controlled measurement system. A nucleotide enters, is incorporated, emits an identifiable signal, and is then chemically reset so that the next base can be added in the next cycle.

That sequence sounds simple. In practice, it is a difficult balancing act. The nucleotide analog must still function well enough inside a polymerase active site. At the same time, it must carry a detectable fluorophore and a reversible blocking group. If the block is too stable, extension stalls. If it is too unstable, more than one base may be added in a single cycle. If fluorophore removal is incomplete, signal leaks into later cycles. If cleavage chemistry is too harsh, template integrity drops.

This is why SBS should be understood as kinetic governance. The platform is not asking each individual reaction to be perfect. It is asking millions of related reactions to stay synchronized enough that their combined fluorescence remains interpretable.

A useful way to picture one SBS cycle is as a four-step rhythm.

First, the nucleotide mix is presented to the cluster. Each template that is chemically ready can incorporate its complementary reversible terminator. Second, excitation light interrogates the newly incorporated fluorophore, and imaging maps the color to base identity. Third, cleavage chemistry removes the dye and blocking group. Fourth, the strand becomes extendable again, and the instrument advances to the next cycle.

This rhythm explains why small inefficiencies matter so much. Incomplete extension creates lagging molecules. Incomplete deblocking creates molecules that are chemically out of step with the main population. Residual fluorophore increases background. The damage is cumulative. The run does not fail because one cycle was imperfect. It fails because dozens or hundreds of slightly imperfect cycles leave a measurable memory in the cluster.

That is also why short-read accuracy should not be explained as a magical platform trait. It is the result of repeated chemical control. Accuracy is won early, then defended cycle by cycle.

Signal-to-noise in SBS is a population problem

The phrase “signal-to-noise” is often used as if it refers only to optics. In SBS, that is too narrow. Signal is not just emitted light. It is emitted light from a sufficiently synchronized population of clonally related templates. Noise includes optical background, residual fluorescence, channel cross-talk, local overlap, and most importantly, molecular disagreement within the cluster itself.

This distinction matters because SBS is an ensemble measurement. The instrument does not see each molecule independently. It sees the composite state of many related copies. Early in a good run, that is an advantage. Many templates are doing almost the same thing, so the signal is strong and easy to decode. Later in the run, the same feature becomes vulnerable. If a fraction of the templates drifts behind or ahead, the cluster begins to report a mixed state.

That is why later-cycle data can still look bright but become harder to interpret. Brightness and information purity are not the same. In SBS, coherence matters more than raw intensity once the run begins to age.

A good mental model is this: the platform is limited less by how much light it can collect than by how long it can preserve collective timing inside each cluster.

The phasing and pre-phasing problem

Phasing is the signature error dynamic of SBS. It describes the gradual loss of cycle synchrony among molecules that began as a clonal cluster. Some strands fail to extend in a given round and fall behind. Others move effectively ahead because the chemistry did not reset cleanly relative to the rest of the population. The exact mechanism can vary by chemistry, but the result is always the same: the cluster signal becomes a weighted mixture of different molecular states.

This is why phasing is a more useful concept than a vague phrase like “quality drop.” It explains what is physically happening. A late-cycle signal is not simply weaker. It is more ambiguous. Instead of representing one base at one cycle, it increasingly reflects a distribution of lagging, on-time, and leading strands.

Figure 2. Phasing and pre-phasing convert a synchronized clonal cluster into a mixed population signal, causing fluorescence purity and base-calling confidence to decay across cycles. This matters for platform choice because short-read performance depends not only on chemistry quality, but on how long useful synchrony can be preserved.

Why this matters for real project design

The biggest mistake in platform comparison is to judge technologies only at the output layer. Short reads are not useful merely because they are accurate and abundant. They are useful because dense, cycle-controlled fluorescence readout is a strong fit for many research questions that depend on per-base confidence and deep sampling.

That is why short-read platforms remain highly effective for many high-depth DNA studies, including whole exome sequencing designs where breadth, depth, and mature variant workflows matter more than uninterrupted molecule length.

At the same time, short reads carry structural limits that follow directly from their chemistry. They compress haplotypes. They break long-range context. They depend heavily on library quality and cluster behavior. They translate molecule diversity into cluster diversity, which means low-complexity inputs and biased fragment pools can waste a large fraction of theoretical throughput.

The correct conclusion is not that SBS is old, and not that long-read platforms are automatically better. The correct conclusion is that SBS is optimized for a specific class of inference problems. It is strongest when controlled cyclic chemistry and very high parallel density matter more than native molecule continuity.

That observation sets up the real transition to third-generation sequencing. Long-read systems do not merely read farther. They observe DNA through different physical principles. One watches polymerase activity in an optically confined nanostructure. The other measures ionic-current disruption as a strand passes through a pore. Those designs relax the cluster-based synchrony demands of SBS and replace them with different tradeoffs in processivity, raw error, and correction strategy.

Third-generation sequencing changes the measurement model

Second-generation platforms read many clonally amplified features in lockstep. Third-generation platforms relax that architecture. They move closer to the individual molecule. That shift changes everything. It changes how signal is generated, how errors arise, how consensus is built, and what kinds of biological structure become visible.

The usual summary is that third-generation sequencing produces longer reads. That is true, but it misses the main point. Long reads are not a bonus feature added onto the same detection model. They come from different sensing physics. PacBio SMRT sequencing detects polymerase activity in real time inside a highly confined optical environment. Nanopore sequencing detects changes in ionic current as a nucleic acid strand passes through a nanoscale pore. In one case, the signal is optical and enzyme-linked. In the other, it is electrical and translocation-linked. Read length follows from those architectures. So do their strengths and weaknesses.

This is why long-read platform selection should never begin with a slogan like “better for structural variants.” It should begin with a more basic question: what is the instrument actually measuring? If a project depends on continuity across repeats, complex rearrangements, haplotypes, full-length isoforms, or native modifications, then the value of a platform lies in how directly it preserves molecule-level context.

Zero-mode waveguides: why SMRT sequencing can watch one molecule at a time

PacBio SMRT sequencing solves a hard optical problem. In solution, fluorescent nucleotides are everywhere. If the instrument illuminated a normal reaction volume, background would overwhelm the single incorporation event that matters. The zero-mode waveguide, or ZMW, solves that by shrinking the illuminated volume to a sub-diffraction-scale observation zone. In effect, it creates a nanoscopic optical chamber where signal from the polymerase active site dominates while freely diffusing background is strongly suppressed.

That physical design is the enabling step. A polymerase is immobilized at the bottom of the ZMW. Fluorescently labeled nucleotides diffuse in and out. When the correct nucleotide is held transiently during incorporation, the fluorescent pulse becomes detectable. Once the label is released as part of the chemistry, synthesis continues. The instrument therefore records a time series of incorporation events from a single template rather than a cycle-gated ensemble image from a clonal cluster.

This creates a very different error landscape from SBS. There is no phasing problem in the same sense, because there is no need to keep a dense population synchronized. But the platform now lives or dies by polymerase processivity, pulse interpretation, and consensus strategy. A single pass through a long template can contain substantial raw error. The rescue comes from repeated observation. Circular consensus sequencing lets the polymerase traverse the same insert multiple times within a circularized template, and the resulting subreads can be collapsed into a much more accurate consensus read. Accuracy is not merely a property of a better polymerase. It is a property of repeated observation of the same molecule.

This is an important strategic point. PacBio’s value is not just “long and accurate.” It is “long enough to span complex structure, then accurate after internal consensus.” That is why it is strong in de novo assembly, repeat resolution, phased analysis, and full-length transcript characterization. It is also why insert design and molecule integrity matter so much. If the library does not preserve informative long fragments, the platform cannot recover that value later. Workflows such as Human Whole Genome PacBio SMRT Sequencing and Full-Length Transcripts Sequencing (Iso-Seq) benefit precisely because the sequencing physics and library architecture are aligned.

Nanopore sequencing: ionic current as a sequence-dependent signal

Nanopore sequencing approaches the same high-level goal from an entirely different direction. Instead of observing base incorporation, it measures how a nucleic acid strand perturbs ionic current while moving through a pore embedded in a membrane. Voltage drives ions through the pore. The strand partially blocks that flow. Different local sequence contexts alter the current in distinguishable ways. The instrument is therefore not imaging chemistry. It is reading a continuously changing electrical trace.

This is where many simplified explanations stop too early. The important detail is that the signal is not usually caused by one isolated base. The pore senses a local sequence window. In practice, several neighboring nucleotides contribute to each current state. That is why the system is often described through a kmer-to-current relationship. The basecaller is solving an inference problem: given a noisy time-series of current levels, which sequence path most likely produced it?

That measurement model has two major consequences.

First, nanopore sequencing can preserve very long molecular continuity because the system does not need repeated stop-and-image cycles. As long as the strand can be controlled and the signal remains interpretable, the read can continue. That is why the platform is attractive for ultra-long genomic reads, structural variant mapping, repeat traversal, full-length RNA, and scaffold-scale assembly.

Second, the same signal model creates a natural route to modification detection. If base modifications alter the local chemical environment inside the pore, they can alter the observed current state. That means canonical base identity and some epigenetic marks can be inferred from the same electrical measurement rather than from separate conversion chemistry. This principle is one reason nanopore sequencing is so attractive in methylation and direct RNA applications, although practical performance still depends heavily on basecaller models, training data, and context-specific calibration.

Figure 3. Third-generation sequencing uses two distinct sensing models: optical detection of real-time synthesis in a ZMW and electrical detection of ionic-current disruption during nanopore translocation. This matters because platform choice should begin with the signal model, not with read length alone.

How nanopore detects 5mC and 6mA in the same signal framework

Modification-aware sequencing is often described as an extra capability of nanopore platforms. A better view is that modification sensitivity falls directly out of the sensing mechanism. The nanopore does not read bases by fluorescent labels or conversion chemistry. It reads how the local molecular environment perturbs current. If a methyl group changes that environment in a reproducible way, then it becomes part of the current signature.

That does not mean modification calling is easy. It is not. The signal from a modified base is context dependent because the pore senses a short neighborhood, not one isolated base. A 5mC event in one kmer context can be easier to distinguish than the same modification in another. Basecalling and modified-base detection therefore depend on model training, pore chemistry, noise control, and downstream calibration. But the strategic advantage remains strong: the platform can preserve sequence and modification information on the same native molecule.

That capability changes experimental design. If methylation status is the main research variable, conversion-based workflows may still be preferred when very mature site-level quantification is needed. But native-molecule sequencing becomes compelling when long-range context, direct RNA, or modification co-occurrence on the same molecule matters just as much as site-level status.

Processivity versus raw accuracy: the real long-read tradeoff

Long-read discussions often collapse into one vague contrast: short reads are accurate, long reads are long. That is too crude to guide real projects. The more useful distinction is between processivity and raw accuracy.

Processivity asks how long the system can continue observing one molecule without losing useful signal. Raw accuracy asks how often a single observation is already correct before consensus or correction. These two variables interact, but they are not the same. A platform can be highly processive and still require heavy correction. A platform can be highly accurate per consensus read but still depend on multiple passes or stricter molecule quality.

PacBio addresses this tradeoff through repeated observation of the same insert. Nanopore addresses it through improvements in pore chemistry, motor control, basecalling models, and downstream consensus generation. In both cases, the raw signal is not the end product. It is the input to a correction pipeline. The strategic question is therefore not “Which raw read is best?” It is “Which correction path preserves the biological structure I care about?”

For example, hybrid assembly uses highly accurate short reads to polish a structurally informative long-read scaffold. This is attractive when the long read resolves repeats or large structural features, but the short read can still improve local base accuracy. Self-correction, by contrast, relies on redundancy within the long-read dataset itself. This is powerful when the project needs to stay within one long-read framework, such as haplotype-aware assembly, full-length transcript reconstruction, or native modification analysis, but it usually requires sufficient long-read depth to work well.

A useful rule is simple. Use hybrid correction when short-read polishing adds real value without destroying the signal you care about. Use self-correction when molecule continuity, native modifications, or full-length architecture are the primary outputs.

Library complexity is not a QC footnote

Many sequencing articles treat library QC as a short checklist: measure concentration, confirm fragment size, proceed. That is useful but incomplete. Library quality does not just affect whether the run works. It determines how much new information each additional read can still buy.

This is the core idea behind library complexity. A complex library contains many distinct molecules that can still be sampled as sequencing depth increases. A low-complexity library saturates early. Additional reads increasingly revisit molecules already seen. At that point, throughput remains high, but information yield flattens.

This matters because many run-design decisions are made as if depth and information scale together. They do not. Once duplication rises and the unique-molecule curve begins to plateau, the economic logic of the run changes. More data may still be generated, but less of it is biologically new. That is a serious problem in low-input, over-amplified, or compositionally narrow libraries.

The physics of fragment size distribution

Fragment size distribution is one of the quiet drivers of sequencing performance. It influences capture, amplification, loading behavior, cluster formation, translocation, and effective read utility. A library with a tight and appropriate distribution is easier to load and easier to interpret. A library with a very broad or distorted distribution can create multiple hidden penalties at once.

In short-read systems, fragment length affects cluster formation and spatial occupancy on the flow cell. Extremely short fragments can enrich adapter-related artifacts or waste informative capacity. Overly broad distributions can complicate loading balance and reduce uniformity across features. In some multiplexed designs, poor cleanup of short contaminating species can worsen misassignment risk by sustaining free adapter or index-containing molecules in the pool.

In long-read systems, the same issue appears differently. A degraded DNA population lowers the effective upper bound of read continuity before sequencing even begins. No platform can reconstruct an ultra-long molecule that was already fragmented during extraction. This is why long-read workflows place such heavy emphasis on high-molecular-weight input, gentle handling, and damage control. The platform is only as “long-read” as the molecules it actually receives.

Lander-Waterman logic and the saturation point of a sequencing run

The Lander-Waterman framework was developed for mapping and assembly, but its central intuition still helps modern sequencing design: coverage is probabilistic, and returns are not linear forever. As sequencing proceeds, the chance of discovering new unique fragments decreases because an increasing fraction of reads lands on molecules or regions already sampled.

In practical terms, this means every run has a saturation zone. Before that zone, more reads recover many new molecules or new loci. Beyond that zone, extra reads mostly deepen redundancy. The exact shape depends on library complexity, target space, bias, duplication, and project type. Exome capture behaves differently from amplicon sequencing. Single-cell libraries behave differently from bulk DNA. Metagenomic mixtures behave differently from clonal isolates. But the governing principle is the same: one should estimate whether the next tranche of sequencing is likely to buy discovery or mostly buy repetition.

This is where sequencing strategy becomes a sovereignty issue rather than a workflow issue. A lab that understands its saturation curve can decide whether to invest in more depth, better libraries, longer molecules, or a different platform. A lab that ignores complexity often overspends on the least effective variable. In many cases, improving library diversity or fragment integrity yields more new information than simply buying another lane.

What saturation means in run design

Three signals usually show that deeper sequencing is nearing diminishing returns. First, the unique-molecule yield starts to flatten even while total read count keeps rising. Second, duplicate burden rises faster than new informative coverage. Third, technical metrics may still look productive while biological novelty barely increases.

When those signals appear, the next best move is often not more depth. It is better library preparation, stricter size selection, improved input quality, or a platform change. In other words, saturation is not just a statistical outcome. It is a decision point.

Figure 4. Sequencing efficiency is bounded by library complexity: fragment-size control affects loading behavior, unique-molecule yield plateaus with depth, and platform choice should follow error mode and information return rather than read count alone. This matters because more reads do not always mean more usable biology.

Sanger, NGS, and TGS: compare failure modes, not just specs

A useful comparison table should explain why each platform wins, where it breaks, and what kind of correction it needs. That is more informative than a generic old-versus-new ranking.

Platform selection matrix

The deeper lesson is simple. A platform should be chosen for the structure of the question. Sanger is ideal when the unknown is small and the template is simple. SBS wins when massive parallel coverage and strong per-base accuracy dominate the value equation. Long-read TGS wins when the unknown is architectural, contiguous, epigenetically native, or difficult to reconstruct from fragments.

Choosing a platform by mechanism, not habit

The most reliable sequencing decisions come from matching the measurement model to the biological problem.

If the project needs dense counting, mature pipelines, and strong consensus across many short fragments, SBS remains hard to beat. If the project needs molecular continuity across repeats, structural rearrangements, or full-length transcripts, long-read platforms deserve priority. If the project needs both, hybrid designs are often more rational than platform loyalty.

That logic scales across project design. A study centered on compact variants may fit Targeted Region Sequencing or Whole Exome Sequencing. A study centered on full genome architecture may require Human Whole Genome PacBio SMRT Sequencing or Nanopore Target Sequencing. A study centered on native methylation or transcript structure may lean toward Nanopore RNA Methylation Sequencing Service or MeRIP Sequencing.

The point is not that one platform is the future and the others are legacy. The point is that sequencing technologies are specialized ways of turning molecules into signals. Once that is understood, platform selection becomes less ideological and more precise.

Conclusion

DNA sequencing has evolved from chain termination chemistry to massively parallel cyclic imaging and then to direct single-molecule sensing. But the most important evolution may be conceptual. The field has moved away from asking which platform is newest and toward asking which platform preserves the right information with the fewest hidden compromises.

SBS remains the master class in controlled cyclic chemistry. Its strength comes from synchrony, reversible terminator logic, and dense optical readout. SMRT sequencing shows how nanophotonic confinement and repeated observation can turn long single molecules into high-value consensus reads. Nanopore sequencing shows how ionic-current disruption can convert native nucleic acids into a continuous electrical signal that carries both sequence and modification information. Library complexity then sits above all of them as the quiet governor of sequencing economics.

The best next step is to define the hardest constraint in the project first. It may be repeat continuity, haplotype structure, native modifications, low-input complexity, or the need for dense short-read coverage. Once that constraint is clear, the chemistry, signal model, and correction path usually become much easier to choose.

FAQ

Mechanism & signal questions

What is the core difference between SBS and third-generation sequencing?
SBS reads synchronized clusters in repeated chemical cycles, while third-generation platforms read single molecules through optical or electrical signals.

Why does PacBio need circular consensus sequencing?
Because repeated observation of the same insert improves final accuracy and turns a noisier single pass into a stronger consensus read.

Why does nanopore signal depend on more than one base?
Because the pore senses a local sequence window, so each current state reflects the influence of several neighboring nucleotides.

Can nanopore sequencing read DNA modifications directly?
It can infer some modifications from altered current signatures on native molecules, though performance still depends on model quality and context.

Project design & platform selection questions

When are short reads still the best choice?
When dense coverage, mature analysis pipelines, and strong per-base confidence matter more than uninterrupted molecule length.

When are long reads worth the extra complexity?
When the project depends on repeat resolution, structural variation, haplotypes, full-length isoforms, or native molecular context.

When does deeper sequencing stop helping?
When unique-molecule yield flattens, duplicate burden rises, and extra reads add much less new information than expected.

When is a hybrid workflow better than a single-platform workflow?
When long reads provide the necessary structure and short reads can still add meaningful polishing or orthogonal support.

References

1. Rodriguez R, Krishnan Y. The chemistry of next-generation sequencing. Nature Biotechnology. 2023;41:1709-1715. DOI: 10.1038/s41587-023-01986-3
2. Wang Y, Zhao Y, Bollas A, Wang Y, Au KF. Nanopore sequencing technology, bioinformatics and applications. Nature Biotechnology. 2021;39:1348-1365. DOI: 10.1038/s41587-021-01108-x
3. Daley T, Smith AD. Predicting the molecular complexity of sequencing libraries. Nature Methods. 2013;10:325-327. DOI: 10.1038/nmeth.2375
4. Bailey TL IV, Ng K, Rusk N, et al. Towards inferring nanopore sequencing ionic currents from nucleotide chemical structures. Nature Communications. 2021;12:6545. DOI: 10.1038/s41467-021-26929-x
5. Lander ES, Waterman MS. Genomic mapping by fingerprinting random clones: a mathematical analysis. Genomics. 1988;2(3):231-239. DOI: 10.1016/0888-7543(88)90007-9
6. Wenger AM, Peluso P, Rowell WJ, et al. Accurate circular consensus long-read sequencing improves variant detection and assembly of a human genome. Nature Biotechnology. 2019;37:1155-1162. DOI: 10.1038/s41587-019-0217-9

*For Research Use Only. Not for use in diagnostic procedures.