Home / Synthetic genomics
Synthetic genomics,
in plain English
Biology has learned to write DNA, not only read it. This page explains what that actually means today, what artificial intelligence contributes, where micro-robots really stand — and, just as carefully, what has not happened yet.
Read, edit, write
Three different abilities get mixed together in coverage of this field, and they are not equally developed.
- Reading — sequencing DNA. Routine, fast and inexpensive. A genome that once took an international consortium years is now ordinary laboratory work.
- Editing — changing letters in a genome that already exists. Precise, widely used, and the basis of approved medical treatments.
- Writing — manufacturing long stretches of DNA to a design. Possible, improving, and still slow and expensive at genome scale. This is the bottleneck, and it is what "synthetic genomics" mostly means.
What a "synthetic genome" is
A synthetic genome is a genome whose DNA was manufactured rather than inherited. In every case so far, that DNA is then placed inside an existing cell, which supplies the machinery to read it.
That distinction matters. The cell is not built; it is borrowed. Nobody has made a living cell from non-living chemistry, and the sequences written so far have been based on natural genomes rather than invented from nothing.
So "creating artificial life" is not a description of any result to date, however often the phrase appears in headlines.
How the field got here
A short milestone timeline. Each of these was a large, multi-year effort by a substantial team.
-
2010
A genome assembled from manufactured DNA runs a cell
A team at the J. Craig Venter Institute produced a bacterium controlled by a chemically assembled copy of a natural genome. The organism was not new; the demonstration that a genome could be written and booted up was.
-
2016
The minimal cell
The same group cut a bacterial genome down to fewer than 500 genes — about the smallest that still grows and divides. Roughly a third of the surviving genes had no known function. That result is quoted constantly, and rightly: it measures how much of the irreducible core of a living thing remains unexplained.
-
2019 onward
Recoded organisms
The genetic code — which DNA triplets mean which amino acid — turns out not to be fixed. Bacteria have been built that use a reduced set of codons, freeing others for new purposes. Work in this direction has continued, including strains reported in 2025 using a substantially compressed code. Beyond the basic science, recoded organisms can be made resistant to viruses that would otherwise ruin an industrial culture.
-
2025
Synthetic yeast, after two decades
The international Sc2.0 project set out to rewrite all sixteen chromosomes of baker's yeast, with laboratories on several continents each taking a chromosome. The final chromosome was published in early 2025. Consolidating the pieces into a single strain has been the long tail of the project — a reminder that in biology, integration is usually harder than construction.
-
2024
Protein design wins a Nobel Prize
The Nobel Prize in Chemistry recognised computational protein design alongside the prediction of protein structure from sequence. Proteins that have no natural ancestor are now designed, manufactured and tested routinely — the clearest example of AI-assisted design producing molecules that work.
-
2026
Sequence models applied to whole genomes
Models trained on genomic sequence at scale moved from preprint to peer-reviewed publication. In the same year, researchers reported using such models to design genomes of bacteriophages — viruses that infect bacteria — based on a well-studied laboratory virus. A small fraction of the designs were functional. The work was published alongside an explicit discussion of safeguards, which is now a normal and necessary part of this field.
What AI actually contributes
Three distinct capabilities, at three very different stages of maturity.
Predicting structure Established
Given a protein sequence, predict the shape it folds into. Accurate enough for many proteins that predicted structures are now a normal starting point for research rather than a novelty.
Designing proteins Established
Describe the shape or function you want and get back candidate sequences. Designed proteins are made and tested in laboratories; many fail, some work, and the ones that work are confirmed at the bench — never in the model.
Genome-scale sequence models Early
Trained on very large collections of DNA, these learn what is typical: which arrangements recur in living organisms and which never appear. Useful for ranking candidates and flagging unusual variants. They do not know what a cell is, and a high score is a hypothesis, not a result.
Designing a whole organism Not achieved
No AI system has designed a novel free-living organism, and none is close. The gap is not mainly computational: we cannot yet predict what a genome will do without building it and observing the result.
AI and micro-robots
Probably the most over-illustrated topic in science coverage. Here is what has actually been built.
Xenobots and anthrobots Demonstrated
Small clusters of living cells — first from frog embryos in 2020, later from human airway tissue in 2023 — that move under their own power and organise themselves. Computer search was used to explore which arrangements would behave which way before anything was built.
No genes were added or removed. The surprise is biological: ordinary cells, taken out of their usual context, do more than their role in a body would suggest. They are visible objects, not nanoscale ones, they have no control system, and they are not treatments.
DNA origami Demonstrated
DNA can be folded into defined shapes — a technique dating to the mid-2000s — including containers that open in response to a molecular signal. A widely discussed 2018 study described such structures delivering a clotting agent to tumours in mice.
These are real, reproducible results in laboratories and animals. None of them is an approved treatment.
Biohybrid microrobots Early
A motile cell or bacterium harnessed to a synthetic carrier, often steered with magnetic fields. Active research with genuine progress in the laboratory.
The unglamorous problems are the unsolved ones: arriving where you intended, being cleared from the body safely afterwards, and manufacturing consistently at scale.
Nanorobots in your bloodstream Not a thing yet
Despite decades of illustrations, no nanorobot or microrobot of this kind is in routine clinical use. Treat any article implying otherwise as a description of laboratory work, an animal study, or an artist's impression.
Taking the risks seriously
A field that learns to write biology has to think about what gets written. Two threads are worth knowing about, because both are being discussed openly by the researchers doing the work rather than only by outside critics.
Screening what gets manufactured. Most laboratories do not make their own long DNA; they order it. That makes commercial synthesis providers a natural checkpoint, and the industry operates screening of orders and customers, reinforced by government frameworks in recent years. As design tools get better at proposing sequences, this manufacturing checkpoint matters more, not less.
Lines drawn in advance. In late 2024 a large group of scientists published a detailed warning against pursuing so-called mirror life — organisms built from molecules of the opposite handedness to all known life — on the grounds that such organisms could be unusually difficult for natural immune systems to recognise. What makes that document notable is its timing: it was written well before anyone could build such a thing, by people close enough to the science to see it coming.
Gataca works on data analysis, not on building organisms. We follow this discussion because anyone working in genomics should, and because the governance questions are as much a part of the field now as the technical ones.
Common misconceptions
"Scientists created artificial life."
No synthetic genome has been placed anywhere but inside an existing cell, which supplies the machinery to read it. Nobody has made a living cell from non-living chemistry.
"The minimal cell was designed from scratch."
It was a natural genome cut down until further cuts killed the cell — subtraction, not design. And about a third of what survived has no known function.
"AI designed a new organism."
AI designs proteins that work, and ranks candidate sequences. Reported genome-design work has been on viruses that infect bacteria, based on a known laboratory virus, with most designs non-functional. A free-living organism is a different order of problem.
"Xenobots are tiny robots."
They are clusters of unmodified living cells, large enough to see, with no control system and no programming in the ordinary sense.
"Nanobots are already used in medicine."
They are not. Laboratory and animal results exist; routine clinical use does not.
"Writing DNA is basically solved, it is just expensive."
Cost is only part of it. Assembling long, accurate sequences and getting a cell to accept and run them remains difficult, and each genome-scale project so far has taken years.
Where Gataca fits
We are not building organisms. We build the analysis layer around genomics work: pipelines, variant interpretation, transcriptomics and custom tools, plus six free browser-based tools for everyday sequence work.
That perspective is why this page is written the way it is. Every result above was settled by an experiment, and the value of a computational prediction is set entirely by the quality of the validation behind it. If you are working in this area and need the data side handled carefully, tell us about the project.
We also write about this every week in The Gataca Review — one short, plain-English post, with the same care about separating what exists from what does not.
Working on something in this space?
Tell us about your project and we'll get back to you within two business days.