Industries / Frontier science / Synthetic biology
DNA synthesis
This work is making DNA to order: writing short pieces with chemistry or enzymes, joining them into genes and longer constructs, and checking that what comes back matches what was asked for.
7
Milestones achieved, from 1970 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1970
A gene made from chemicals
A team of thirteen chemists reported the total synthesis of the gene for an alanine transfer RNA from yeast, built by joining short pieces made in flasks. Until then genes could only be copied from living material.
Lessonthe author list and the years of work show what writing one short gene cost before any of it was automated.
Credit
K. L. Agarwal, H. Büchi, M. H. Caruthers, H. G. Khorana and colleagues, in Nature
2004
Genes assembled from a chip
Thousands of short pieces were made on light-programmed microfluidic chips, released, copied and then sorted by sticking them to tagged partner pieces, which cut synthesis errors ninefold. The team assembled all 21 genes for the proteins of one part of the E. coli ribosome.
Credit
Jingdong Tian, Xiaolian Gao and George Church, in Nature
2009
One-pot joining of fragments
Three enzymes in a single tube at one temperature trim the ends of overlapping DNA pieces, let them stick together and seal the joins, with no need for cut sites in the sequence. The authors reported joining pieces into molecules of several hundred thousand bases.
Credit
Daniel G. Gibson, Hamilton O. Smith and colleagues at the J. Craig Venter Institute, in Nature Methods
2009
A gene foundry shuts down
Codon Devices, founded in 2004 to sell synthetic genes and design services, stopped operating after its board voted in late March 2009; the Boston Herald reported the closure and the company made no announcement.
Lessonthe analyst Rob Carlson, who was interviewed for the Nature piece and puts the company's funding at the order of USD 50 million, wrote that the possible margin on synthetic genes was falling fast while the value sat in the design work customers would not hand over, so a firm selling both had nothing left to charge for.
Credit
Codon Devices; reported by the Boston Herald and by a Nature news piece
2018
An enzyme writes the DNA
Each copy of the enzyme terminal deoxynucleotidyl transferase was tied to one DNA building block it could add, so a strand grows one base per round and then releases. The authors give this as a way past the roughly 200-base limit of the standard chemistry and its hazardous waste.
Credit
Sebastian Palluk, Daniel H. Arlow and colleagues at the Joint BioEnergy Institute, with Jay Keasling, in Nature Biotechnology
2023
A bench machine checked outside
A US Army DEVCOM Chemical Biological Center team ran the first commercial benchtop enzymatic synthesiser, DNA Script's Syntax-100, which makes 96 pieces of 60 bases in 13 hours, and used its output to assemble a working green fluorescent protein gene. Of the clones they sequenced, 24% were correct from the machine's 60-base pieces, against 27% for bought 60-base pieces and 40% for bought 90-base pieces; the machine's pieces arrived in half the time and left far less hazardous waste.
Lessonon their figures, buying the finished gene for about USD 280 still cost less than making the pieces either way, so the machine bought time rather than money.
Credit
Rebecca Simmons, Daniel McDonald and Jennifer Robinett, in Frontiers in Bioengineering and Biotechnology
2026
Assembly separated from sequence
Every earlier assembly method steers the joins using sequence that stays in the finished construct, so hard sequences cannot be changed to make assembly easier. Sidewinder puts the instructions in removable three-way junctions instead, and the authors report a 40-piece assembly, sequences with high GC content and long repeats, several genes built in one reaction, a library of 442,368 variants, and a wrong-join rate of about one in a million.
Credit
Noah E. Robinson, Kaihang Wang and colleagues at the California Institute of Technology, in Nature, 21 January 2026
Watch for
A tenfold better junction
Sidewinder built 320 DNA molecules at once from a single pool of short pieces, with a wrong-join rate of one in ten million, using a new design program called PyWinder.
Achieved whenthe promised preprint appears and a group with no stake in Genyro or the Wang laboratory reports the same rate.
Claimed by
Genyro, the San Diego company that licensed the method from Caltech, May 2026
Closed DNA on the bench
A kit makes Touchlight's doggybone DNA, a closed loop of double-stranded DNA with no bacterial sequence in it, by enzymes on a bench in a single day, as a replacement for growing plasmid DNA in bacteria.
Achieved whena laboratory outside the two companies publishes yields and purity from the kit, or a filing shows it in routine production use.
Claimed by
New England Biolabs with Touchlight, March 2026
Open
01
An error table for synthesis
One public table of how often synthetic DNA comes back correct. Each row carries the method (bought chemical pieces, chip pieces, enzymatic pieces, with or without error correction), the piece length, the year, the measure used (errors per thousand bases, per-position error rate, or the share of sequenced clones that are perfect), how many pieces or clones were measured, who ran the test, and the source. Rows where the test was run by a company selling one of the products are marked as such, and so are rows where the other suppliers are not named.
The briefCredit · unclaimed02
Oligo design tools, compared
A public test that takes assembly designs already printed in papers' supplementary tables, feeds the same target sequences to each open design tool, and reports how often a tool reproduces the published split into pieces and how its predicted melting temperatures compare. Use only targets from papers on reporter genes, enzymes and model organisms.
The briefCredit · unclaimed
Open 01 · the brief
An error table for synthesis
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
One public table of how often synthetic DNA comes back correct. Each row carries the method (bought chemical pieces, chip pieces, enzymatic pieces, with or without error correction), the piece length, the year, the measure used (errors per thousand bases, per-position error rate, or the share of sequenced clones that are perfect), how many pieces or clones were measured, who ran the test, and the source. Rows where the test was run by a company selling one of the products are marked as such, and so are rows where the other suppliers are not named.
Why it is still open
The comparisons across suppliers that exist are run by sellers and name nobody. DNA Script compared 60- and 120-base pieces from four commercial suppliers with its own machine's output, measured errors by Illumina sequencing, and reported mean purity for the 60-base pieces of 74.4%, 77.0%, 64.0% and 71.1% for suppliers it called A, B, C and D. Integrated DNA Technologies' white paper compares its own gene fragments with fragments from a few other synthesis providers, also unnamed. The 2023 DEVCOM team had to produce its own figures, and reports that the machine's maker had published a different number for bought chemical pieces without saying where those pieces came from.
Built with what exists
The Europe PMC and Crossref search interfaces for papers; suppliers' own application notes and white papers; a current AI model that reads PDFs and fills a fixed schema, the kind of document work covered in the Note "Seven more flagged"; Python and pandas; a dated release on Zenodo.
First step
Extract the rows of the DEVCOM paper's own comparison table and of the DNA Script note, add the ten papers they cite, and publish the schema and the first CSV.
Done looks like
At least 150 rows from at least 60 sources, published as a CSV with the extraction code; a second person re-checks 20 random rows and the disagreement rate is published; one chart of perfect-clone share by method and year, with seller-run rows marked.
Open 02 · the brief
Oligo design tools, compared
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test that takes assembly designs already printed in papers' supplementary tables, feeds the same target sequences to each open design tool, and reports how often a tool reproduces the published split into pieces and how its predicted melting temperatures compare. Use only targets from papers on reporter genes, enzymes and model organisms.
Why it is still open
The comparisons that exist are run by each tool's own authors. The CertPrime paper, in Computational and Structural Biotechnology Journal in 2025, benchmarks itself against earlier methods on one complex sequence and states that current tools are limited on long sequences, length control and spurious pairing. No shared, rerunnable test set exists. Agents are now scored on a neighbouring task: the Terminal-Bench suite carries a task that asks a model to design PCR primers for a Golden Gate assembly and checks the result against length, melting-temperature and cut-site rules, which scores a design against rules rather than against designs that were published and built.
Built with what exists
Primerize, pydna and the public CertPrime web tool; supplementary tables from open-access papers; Python; GitHub Actions to rerun the test when a tool changes; Zenodo for releases.
First step
Collect ten open-access papers that print their oligo tables, run Primerize on their targets, and publish the first comparison for those ten.
Done looks like
60 published designs run through every tool, a per-tool tally of exact reproductions and near misses, and a published list of targets that no tool reproduces.