A cell built from scratch
Nobody has yet assembled a living, dividing cell from non-living molecular parts.
Open in the interactive tree →Today's 'synthetic cells' are top-down: a chemically written genome is transplanted into a hollowed-out natural bacterium. A bottom-up cell would be built from lipids, DNA, ribosomes and enzymes mixed in a bubble that grows, copies itself and divides.
As of October 2026
JCVI-syn3.0 (2016) runs on only 473 genes and JCVI-syn3A (2017) on 493, but both were made by transplanting a synthetic genome into an existing Mycoplasma cell. In syn3.0, 149 genes had no known function, and 91 genes in syn3A were still unexplained in 2019. The March 2026 whole-cell simulation of syn3A shows how well a minimal cell can now be modelled, but no lab has built a self-reproducing cell from purified parts.
What is missing
- Function of the remaining 'essential' genes of unknown role
- Self-assembly of a working ribosome and protein-making machinery from parts
- Coupled genome copying, membrane growth and division inside a compartment
- Ways for the cell to evolve and repair itself
- Design tools that predict whether a parts list will work
Becomes possible once solved
- Designer microbes for medicines and materials
- A test of what life minimally requires
- A route to artificial life in the lab
Open steps
- Genes of unknown role High AI leverageRoughly 90 genes of the minimal cell syn3A are essential, yet their function was still unexplained in 2019.
- Designing a working minimal genome Medium AI leverageA design tool that predicts whether a parts list or redesigned genome will boot a living cell; the AI-designed genomes shown so far are tiny viral ones.
- Building a ribosome from parts Low AI leverageSelf-assembly of a working ribosome and protein-making machinery from purified parts, so a cell can rebuild its own components.
- Copy, grow and divide together Low AI leverageCouple genome copying, membrane growth and division inside one compartment so that the cell can reproduce.
- Self-repair and evolvability Low AI leverageGive a built cell ways to repair damage and to evolve, which needs error-correcting copying and a stable genome.
Where AI could help
Medium AI leverage. AI can design genomes and explain unknown genes, but assembling a working cell from parts still depends on slow wet-lab work.
- Genome language models that propose minimal or redesigned genomes to test
- Structure and function prediction to explain essential genes of unknown role
- Design tools that predict whether a parts list of enzymes and lipids will self-assemble
- Whole-cell simulations as a digital twin to test designs before building them
Shown so far
- In September 2025 a Stanford, Arc Institute and Memorial Sloan Kettering team reported in a bioRxiv preprint that the genome language models Evo 1 and Evo 2 designed bacteriophage genomes, with 16 viable phages in lab tests. source
- In early 2021 the Journal of Proteome Research reported a deep-learning structure pipeline that assigned functions to many previously unannotated proteins of the JCVI-syn3 minimal genome. source