Biocomputing and DNA computing

01Overview and value chain

Markers: [EC: research-stage, no dedicated conformity route | OECD: 1.6 Genomics and bioinformatics | Regulator: FDA (US), EMA (EU)]

Biocomputing covers two distinct propositions that share a substrate. The first is storage: DNA holds data at densities no magnetic or optical medium approaches and remains readable for centuries, so the engineering problem is not capacity but access — writing cheaply, finding one file among billions, and recovering the data despite a synthesis and sequencing process that corrupts it. Published benchmarking of six representative codecs shows the coding layer can tolerate error rates up to roughly 14% and sequence loss up to roughly 65% in isolation, which is the margin that makes an inherently lossy medium usable at all. The second proposition is computation: genetic circuits inside living cells performing logic, with 2026 work demonstrating three-input combinatorial gates, half adders, full adders and a dynamic multiplexer built from orthogonal trans-splicing AND gates and synthetic microRNAs — arithmetic executed by a cell rather than a transistor. Around both sits a smaller unconventional-computing tradition that treats slime mould and fungal mycelium as computing substrates. The honest framing for all of it in 2026 is that this is research and patent activity, not a product market: the named participants are corporate research labs, universities and platform companies filing patents, and no regulatory conformity route exists for a biological computer.

The key directions of biocomputing are:

1. DNA Data Storage and Random Access: encoding data in synthesised polynucleotides and retrieving specific files from a pool, with patent work on whole-pool amplification combined with in-sequencer random access so that access and read collapse into a single workflow step.
2. Error-Correction Coding for DNA Storage: the codec layer that makes a lossy medium reliable, benchmarked across six representative schemes tolerating error rates to roughly 14% and sequence loss to roughly 65% in isolation, with realistic deployment densities lower than the isolated figures suggest.
3. Genetic Circuit Computing: logic implemented inside mammalian cells using orthogonal trans-splicing AND gates, native-synthetic hybrid promoters and synthetic microRNAs for inhibitory logic, demonstrated in 2026 as three-input gates, half and full adders and a dynamic three-to-one multiplexer within few computational layers.
4. Unconventional and Biohybrid Computing: living systems as computing and sensing substrates — Physarum polycephalum cellular-automata models driving robot navigation, and fungal mycelial networks proposed as distributed sensing, self-healing and anomaly-detection layers.

Sectoral value chain

[Encoding & Design] ──> [Molecular Write] ──> [Storage/Execution] ──> [Molecular Read]
                                                        │
                                                (error correction)
                                                        │
                                                        ▼
[Recovered Data / Output] <─── [Decoding] <─────────────┘

Value chain levels

Cross-cutting technologies of the sector:

• Molecular Electronics Sensing: single-molecule electronic sensors using DNA reporter tags for multiplex genetic analysis, providing an electrical rather than optical readout of molecular events.
• High-Fidelity DNA Synthesis: the write step common to storage and circuit construction, where throughput, cost and fidelity jointly determine what is economically possible.
• Bio-Inspired and Cellular-Automata Algorithms: computational models drawn from organisms — notably Physarum polycephalum — used both to run biohybrid hardware and as algorithms in conventional robotics.

02US

The US contribution is split between a corporate research programme treating DNA as an archival storage tier and a platform company building molecular-electronic readout, with both visible primarily through patent filings rather than shipping products.

Microsoft DNA storage random access, Roswell molecular electronics, patent-stage maturity

• Microsoft (DNA data storage): a patent application by Microsoft Technology Licensing describes copying all polynucleotide-encoded data in a storage container while preserving random-access read capability, combining whole-pool amplification with in-sequencer random access so that access and read occur in a single workflow step — addressing the retrieval problem that separates a DNA archive from a DNA sample.
• Roswell Biotechnology (molecular electronics): the San Diego company’s filings cover molecular electronic sensors for multiplex genetic analysis using DNA reporter tags, and N-terminal multifunctional peptide conjugation with alpha-helical peptides for biosensing. The documents describe a technology platform rather than a finished product, with roots in provisional filings from 2020.
• Maturity, stated plainly: what is visible in 2026 sourcing for both is intellectual property and platform description, not revenue, deployment scale or funding disclosure — which is the accurate status of molecular computing and storage in the US commercial sector.

03CN

China’s activity is concentrated in university programmes that treat DNA information storage as an engineering pipeline to be optimised end to end, with recruitment aimed squarely at the synthesis and error-correction bottlenecks.

SJTU Bio-X DNA information storage, synthesis platform optimisation, encoding and error correction

• Shanghai Jiao Tong University (Bio-X Institute): the Shi Yongyong group recruited postdoctoral researchers in 2026 explicitly for DNA information storage and DNA synthesis platform optimisation, covering DNA encoding design, data writing, sequencing readout, error correction and information recovery as one connected workflow.
• Synthesis as the stated bottleneck: the same programme frames its objective as high-throughput, low-cost, high-fidelity DNA synthesis, with experimental design, process optimisation and performance evaluation named as the work — an accurate reflection that the write step, not the read step, currently limits the field.
• Disciplinary breadth of recruitment: candidates are sought from molecular biology, synthetic biology, chemical biology, biochemistry, nucleic-acid chemistry, bioengineering, bioinformatics, computational biology, information storage and sequencing technology, with DNA information storage and coding/error-correction algorithms named as preferred backgrounds — the interdisciplinary profile the field requires.

04EU

Europe supplies the theoretical layer that makes DNA storage trustworthy and the longest-running unconventional-computing tradition, both academic rather than commercial.

ETH Zurich codec benchmarking, UWE Bristol unconventional computing, biohybrid fungal systems

• ETH Zurich (error-correction benchmarking): a March 2026 Nature Communications study systematically benchmarked six representative codecs for sequence-based DNA data storage under both in-silico and in-vitro conditions, finding that in isolation codecs tolerate error rates up to roughly 14% and sequence loss up to roughly 65%, while realistic deployment implies lower usable storage densities than the isolated figures suggest.
• UWE Bristol (Unconventional Computing Laboratory): researchers including Genaro J. Martínez and Andrew Adamatzky built a vigilance robot, McIntosh I, whose navigation is driven by a bio-inspired algorithm based on a Physarum polycephalum complex cellular-automata model, with collaborators from Mexico’s Artificial Life and Robotics Lab and the Czech Unconventional Algorithms and Computation Lab.
• Fungal biohybrid systems: work from the same tradition proposes living fungal mycelial networks as biohybrid substrates for security and resilience — distributed sensing, self-healing materials and low-observability anomaly detection, leveraging decentralised control, embodied memory and autonomous repair for infrastructure protection and environmental monitoring.

05Leading companies and research institutes

06Tech stack and innovations

The stack is best read as two columns — storage and computation — resting on one shared physical layer, molecular synthesis and sequencing, with an unconventional-computing tradition running alongside.

1. DNA Storage: Write, Address, Read:
   • Data is encoded into constrained nucleotide sequences and synthesised; the write step is the cost and fidelity bottleneck, which is why the explicit research objective is high-throughput, low-cost, high-fidelity synthesis rather than higher density.
   • Retrieval is the differentiator between an archive and a sample: whole-pool amplification combined with in-sequencer random access preserves the ability to read a specific addressed subset while copying the pool, collapsing two operations into one.
2. Error-Correction Coding:
   • Benchmarking of six representative codecs under in-silico and in-vitro conditions shows tolerance of error rates up to roughly 14% and sequence loss up to roughly 65% when each is assessed in isolation.
   • The same work is explicit that realistic deployment implies lower usable storage densities than isolated codec performance suggests — the gap between a benchmark and a working archive.
3. Genetic Circuit Computing:
   • Mammalian gene circuits built from orthogonal trans-splicing AND gates, native-synthetic hybrid promoters and synthetic microRNAs implementing inhibitory logic operate within few computational layers, which is what makes them scalable.
   • Demonstrated circuits in 2026 include a three-input combinatorial logic gate, a half adder, a full adder and a dynamic three-to-one multiplexer — the same primitives as digital electronics, evaluated by cellular machinery.
4. Molecular Electronics and Biohybrid Substrates:
   • Molecular electronic sensors read single-molecule events electrically using DNA reporter tags for multiplex genetic analysis, bypassing optical detection entirely.
   • Biohybrid computing takes the organism as the machine: a Physarum polycephalum cellular-automata model drives robot navigation, and fungal mycelial networks are proposed as distributed sensing and self-healing layers with embodied memory and autonomous repair.

07Value chains and production pipelines

Industrial pipeline of a DNA information-storage campaign (research stage; no dedicated conformity route, FDA/EMA relevant only where biological outputs are clinical)

┌───────────────────────────┐      ┌───────────────────────────┐
│ 1. Codec & Sequence       │ ───> │ 2. DNA Synthesis (Write)  │
│    Design                 │      │                           │
└───────────────────────────┘      └───────────────────────────┘
                                                  │
                                                  ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 4. Random Access          │ <─── │ 3. Pool Storage           │
│    Retrieval              │      │                           │
└───────────────────────────┘      └───────────────────────────┘
               │
               ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 5. Sequencing Readout     │ ───> │ 6. Error Correction &     │
│                           │      │    Recovery               │
└───────────────────────────┘      └───────────────────────────┘

Stage 1: Codec and Sequence Design

The payload is mapped onto nucleotide sequences under biochemical constraints, and the error-correction scheme is selected before synthesis, because redundancy has to be designed into the codewords rather than added afterwards. The choice is consequential: benchmarked codecs differ in how much substitution error and outright sequence loss they survive.

Stage 2: DNA Synthesis (Write)

Designed sequences are synthesised into a physical pool. This is the field’s binding constraint — throughput, cost and fidelity together — and it is the stated target of dedicated research programmes rather than a solved commodity step.

Stage 3: Pool Storage

The synthesised pool is archived, where DNA’s density and longevity are the advantages that motivate the whole approach. In the computing branch, the equivalent stage is expression of the circuit in living cells, where the logic is evaluated by cellular machinery instead of being stored.

Stage 4: Random Access Retrieval

Locating one file among many without reading everything is what makes an archive practical. Patent work combines whole-pool amplification with in-sequencer random access, preserving addressed retrieval while the pool is copied and merging access with the read step.

Stage 5: Sequencing Readout

The addressed subset is sequenced, producing reads carrying substitutions, insertions, deletions and dropout. The raw readout is expected to be corrupted; the design assumption of the whole pipeline is that the medium is lossy and the coding layer compensates.

Stage 6: Error Correction and Recovery

The codec reconstructs the payload, tolerating error rates to roughly 14% and sequence loss to roughly 65% in isolated benchmarks. Realistic end-to-end deployment yields lower usable densities than those isolated figures imply, which is the honest measure of where DNA storage stands in 2026.