Living cells already measure their surroundings, combine chemical signals and change their behaviour. Engineers can borrow that machinery for computation by connecting biological components in deliberate ways. Recent work has made two techniques especially tangible: printing different layouts of engineered bacterial cells, and designing networks of regulatory genes in mammalian cells with a computational compiler.
These are different systems from the neural organoids and cultured-neuron devices often called biocomputers. The information here passes principally through chemistry and gene regulation. Signals may take minutes or hours to propagate. That can be perfectly acceptable when the eventual task is detecting a chemical, regulating a therapeutic response or operating in an environment that electronics cannot directly sense.
What does a cell compute?
A genetic regulatory circuit responds to inputs such as metabolites, small molecules or the activity of another gene. The input alters transcription, RNA stability, protein abundance or another cellular process. A response is read out through a reporter or a subsequent biochemical reaction. When researchers couple several such elements, an output can depend on multiple inputs, previous state or timed sequences.
Logical operations are one possibility, and graded analogue responses are another. A digital gate separates outcomes into categories. A mammalian regulatory network often gives a continuous, concentration-dependent response, making analogue computation a natural design target.
| Approach | Unit doing the work | Signals | Evidence in 2026 |
|---|---|---|---|
| Printed bacterial circuits | Engineered bacteria arranged in a layout | Small molecules pass between strains | Peer-reviewed demonstrations of a demultiplexer and adders |
| Mammalian-cell genetic circuits | RNA regulators and gene-expression machinery | Continuous input and output levels | Preprint reporting three generated circuits built in one pass |
| DNA molecular circuits | Designed DNA strands or complexes | Molecular binding and displacement | Peer-reviewed 2026 thermodynamic DNA computer |
Printed bacteria: programme the arrangement
Doosthosseini, Chen and Voigt developed five strains of Pantoea agglomerans that implement two transistor-like functions and three signal relays. An acoustic liquid handler placed the strains on a surface in particular geometries. Because communication between cells is chemical, changing their spatial arrangement changes the connections between operations.
The team constructed multi-input, multi-output circuits, a demultiplexer, a half-adder and a full-adder. The experiment was published in Nature Chemical Biology on 17 August 2026. It is a useful advance because the same set of strains served several circuit designs. The bacterial computing guide examines how that reconfigurability works and what it costs in time.
Mammalian cells: programme the target response
Jean Disset and colleagues published a September bioRxiv preprint on a biocompiler for mammalian gene circuits. Their system learns how measured regulatory building blocks behave, then searches for combinations that produce a requested analogue input-output function. The models imitate transcription, translation and RNA cleavage as reusable components.
The authors built three new circuit architectures proposed by the compiler and report that each produced its intended behaviour in a single design pass, without manual tuning after the design. That is encouraging, although three examples and one team's preprint do not establish general reliability across all cell types or tasks.
Can it scale into useful technology?
Scale does not simply mean adding more biological gates. Circuit elements compete for cellular resources, populations grow and change, and biochemical signals leak or diffuse differently as the environment changes. A useful cellular device needs stable behaviour over time, reliable readout and controls for unwanted interactions. Safety, containment and repeatability are central requirements if engineered organisms leave the research laboratory.
There are applications where these limitations may be manageable. Environmental sensors need to respond to chemicals, and diagnostic systems may only need a decision within a biologically meaningful time window. Programmable living materials could react directly to local conditions. Those are research directions, not products demonstrated by the two 2026 papers.
How this relates to the rest of biocomputing
The term biocomputer is used for several very different machines. Neural wetware processes rapidly changing electrical signals; printed cellular circuits route molecules; DNA systems compute through designed binding reactions. Gharios and colleagues' October 2026 review provides a wider perspective across nucleic acids, proteins, cells and materials. Our technology map keeps these branches distinct so claims about one do not migrate uncritically to the others.