| Q | p_error | accuracy |
|---|---|---|
| 10 | 1e-01 | 90.0000% |
| 20 | 1e-02 | 99.0000% |
| 30 | 1e-03 | 99.9000% |
| 40 | 1e-04 | 99.9900% |
MLB7052.LT: Modern Methods in Molecular Ecology
TalTech / UT Institute of Technology
| Q | p_error | accuracy |
|---|---|---|
| 10 | 1e-01 | 90.0000% |
| 20 | 1e-02 | 99.0000% |
| 30 | 1e-03 | 99.9000% |
| 40 | 1e-04 | 99.9900% |
Q30 — the usual quality threshold — means one wrong base in a thousand. At 300 bp per read that is still, on average, 0.3 errors per read.
Multiplexing lets a large number of libraries be pooled and sequenced together in a single run. Each library carries an index (barcode); after sequencing, reads are demultiplexed back to their sample and aligned.
Fragment → adapters → clonal amplification → immobilise on a surface → cycle chemistry with imaging → base calls → assembly or mapping.
Cycles of dTTP, dCTP, dGTP — but dATPαS instead of dATP, because luciferase would use ATP itself.
PPi released on incorporation → APS → ATP → luciferase → light, read by a CCD camera. Apyrase degrades unincorporated nucleotides between cycles.
Märkus
454 gave 16S researchers their first community-scale datasets. A great deal of early microbial-ecology literature carries its homopolymer errors as OTU inflation.
Fluorescent-labelled DNA octamers, a di-base colour code, and five universal sequencing primers offset by one position.
Every base is interrogated twice, in two different ligation rounds — which is what gave SOLiD its inherent error correction.
Cheaper to produce and more competitively priced than 454.
Cost of a human genome: $10M in 2007 → $100k in 2009 → $1000 in 2014.
Sequencing by synthesis, on a flow cell:
Each cluster is ~1000 identical copies — enough photons to see one base.
Reversible terminators carry a dye and a 3′ blocking group. Extend one base, image, cleave the dye and the block, repeat.
4-channel systems use four dyes; 2-channel systems encode the four bases as red, green, both, or neither.
Several million clusters are amplified at distinguishable locations, imaged by camera, and turned into per-cluster base calls in a text file.
The insert size and the inner distance are the numbers your assembler and your amplicon merger both need.
Library preparation ends in circularisation: the insert plus barcode is closed into a single-stranded circle with a splint oligo.
Rolling circle amplification copies the circle hundreds of times into one long concatemer that folds into a DNA nanoball — many copies, one molecule of origin.
Nanoballs are loaded onto a patterned array flow cell — one ball per spot, by electrostatic interaction — and read by cPAS (combinatorial probe anchor synthesis).
The Helicos single-molecule device:
2010 — Pacific Biosciences real-time sequencing. A single polymerase sits at the bottom of a zero-mode waveguide; phospholinked nucleotides fluoresce as they are incorporated, and the pulse pattern is the sequence.
Because the same circular template is read repeatedly, random error averages out — the basis of HiFi reads.
The ZSG process lets electron microscopes “see” DNA, labelled with unique contrast-atom labels.
Biological nanopores for DNA sequencing were first proposed in the 1990s and made commercially available in May 2014.
The first commercial nanopore sequencing device:
Nõuanne
Portability changes study design, not just logistics: sequencing at the sampling site removes the cold chain and the shipping permit from the critical path.
The signal is not one base at a time — several bases sit in the constriction together, so basecalling is a model over a k-mer of current levels.
Because basecalling is a model, old data can be recalled with a better model years later.
Flongle · MinION · SmidgION · GridION · PromethION
Pore chemistry matters as much as the device: R9.4.1 reads through a single constriction, R10 through two — better homopolymer accuracy.
Three compartments — common reservoir, bulk buffer, buffer in the wells — shipped with storage buffer in all three, so the running buffer can be changed without changing the electrochemistry.
| Technology | Reads per run | Read length |
|---|---|---|
| Sanger (96-well plate) | 96 | 1,000 |
| Roche 454 (FLX) | 700k | 700 |
| SOLiD (5500 W) | 266M | 100 |
| Illumina (HiSeq) | 312M | 300 |
| PacBio (Sequel) | 370k | 20,000 |
| Nanopore (MinION) | 2M | 1,000–100,000 |
Nobody is best at both axes — which is the whole argument for hybrid approaches.
Human Genome Project · second generation · third generation — the same problem, three amounts of cloning.
Combining the strengths of different technologies:
Used for de novo genome assembly and for (meta)transcriptomics, where long reads resolve isoforms that short reads only imply.
15 Escherichia coli and 10 Staphylococcus aureus isolates from livestock and food; GeneJET genomic DNA purification.
Short reads — NextSeq (Illumina), paired-end 2 × 150 bp, 1.9–9.1 M reads per genome, Q > 20, fastp for QC and adapter removal, SPAdes de novo assembly.
Long reads — up to 269 kbp, combined with the short-read assembly.
The pore reads the first part of a molecule, the software decides whether it is in the region of interest, and if not the voltage is reversed to eject the strand and free the pore. Enrichment without any wet-lab capture step.
DNA, RNA, proteins, metabolites — potential, expression, and activity, each with its own sequencing-based readout.
The bottleneck moved from the bench to the pipeline: QC, assembly, annotation, taxonomic and functional profiling, and validation — each a set of choices that change the answer.
RADseq is Lecture 2’s RFLP logic with a sequencer at the end: digest, size-select, sequence around the cut sites, call SNPs at thousands of loci.
A SNP is a substitution of a single nucleotide at a specific position in the genome that is present in a sufficiently large fraction of the population — 1% or more. Rarer than that, and we call it a mutation.
SNPs fall in noncoding or coding regions; coding SNPs are synonymous or non-synonymous, and non-synonymous ones missense or nonsense.
Barcoded gel beads in droplets give every cell its own barcode, so one library holds the transcriptional profile of thousands of individual cells rather than their average.
Amplification. Second-generation platforms need clonal copies of each fragment to produce enough signal; third-generation platforms read a single molecule, which removes amplification bias and allows very long reads.
The light signal is proportional to the number of identical bases incorporated in one flow, so a run of, say, seven A’s has to be distinguished from six or eight by intensity alone.
A 1 in 1000 probability that the base call is wrong — 99.9% accuracy. Q = −10·log₁₀(P_error).
Long reads span repeats and structural variants so contigs join up; short reads have far lower per-base error, so they polish the consensus. Neither alone gives a contiguous and accurate assembly.
Unmethylated cytosines are deaminated to uracil and then read as thymine, while methylated cytosines are protected and still read as C. Every C remaining in the read was methylated.
Tähtis
Peeter Laas · peeter.laas@taltech.ee · peeter.laas@ut.ee
Next: from reads to inference — quality control, assembly, and the analyses these platforms actually feed.
Fall 2026