CAD for DNA
and mRNA.
Assemble a construct from parametric parts rather than typing characters. Nucleora re-checks it against 23 design constraints on every edit, folds and simulates it, and writes a release record that records every check, waiver and source.
BUILD SPECIFICATION — 001
Every figure below is counted from this build. The command that produced it is printed underneath.
Nucleora 0.11.0/api/* paths.WHAT IT DOES — 002
A construct is a design, not a string of characters.
Nucleora holds a DNA or mRNA construct as a parametric document: promoter, 5′ UTR, coding sequence, 3′ UTR, poly(A) tract, backbone, each with its own parameters and its own source.
The nucleotide sequence is a rendering of that document. Change the poly(A) target length and the sequence regenerates. Swap the host and the coding sequence is re-optimised against that host's codon table. The two stay in sync in both directions — edit the sequence and the part boundaries follow.
On every edit the constraint engine re-evaluates all 23 rules against the current document and returns one of four verdicts for each — PASS, WARN, FAIL or NA — with the evidence that produced it. Sixteen rules carry hard severity: an unwaived hard failure blocks the build, and the document reads blocked until every one of them is explicitly waived. Seven carry guideline severity: they are reported but do not block. Every waiver is recorded against the document revision.
RNA secondary structure is folded through ViennaRNA on Nucleora's servers. In-vitro transcription is simulated against the template to produce a yield estimate and a manufacturability grade. The result is a release record: the document revision, every rule verdict, every waiver, every simulation output, and the bill of materials, exportable as GenBank, FASTA, SBOL 2 or SBOL 3.
THE WORKBENCH — 003
38 workspace views. One document model.
Cloning, codon optimisation, CRISPR design, gel and sequencing QC, phylogenetics, protocols and release — all acting on the same construct.
HOME — Dashboard and project overview
DESIGN STUDIO — Cloning, codon optimisation, mutagenesis
SEQUENCE EDITOR — mRNA linear and IVT plasmid circular views, with auto-annotation
ANALYSIS & QC — Align sequencing results to the design, classify every variant
CONSTRAINT ENGINE — 004
The complete rule set. All 23 of them.
Not a sample. This is every rule registered in engine/cad_constraints.py, with the severity each one carries. An unwaived hard failure blocks the build; a guideline failure is reported but does not block. Each rule returns its evidence and a citation alongside its PASS / WARN / FAIL / NA verdict.
| Rule id | Title | Severity |
|---|---|---|
| kozak_context | Kozak context must precede the start codon | HARD |
| bacterial_rbs_present | Bacterial CDS must be preceded by a Shine-Dalgarno RBS | HARD |
| no_internal_sd | No internal ribosome-binding (SD-like) site inside a bacterial CDS | GUIDELINE |
| frame_preserved | Reading frame preserved across every junction | HARD |
| no_premature_stop | No premature stop codon inside the coding sequence | HARD |
| gc_window | GC content within a synthesizable range | GUIDELINE |
| no_strategy_site_collision | No restriction-site collision with the chosen cloning strategy | HARD |
| no_cryptic_polya | No cryptic poly(A) signal inside the coding sequence | GUIDELINE |
| packaging_capacity | Viral payload within the vector's packaging capacity | HARD |
| promoter_host_match | Promoter host must match the codon-optimization host | HARD |
| terminator_present | Transcription terminator / poly(A) signal present and downstream | HARD |
| methylation_blocking | Dam/Dcm methylation must not block the strategy's restriction enzyme | GUIDELINE |
| insert_no_premature_stop | Insert CDS must contain no premature in-frame stop (checked before assembly) | HARD |
| insert_gc_window | Insert GC should sit inside the synthesis vendor's window (before assembly) | GUIDELINE |
| backbone_marker_present | Backbone must declare a selection marker (checked at selection time) | HARD |
| codon_host_coherence | Codon table used should match the host the backbone declares (before assembly) | GUIDELINE |
| mrna_utr5_uorf_burden | 5'UTR should carry no upstream AUG that competes with the main start codon | HARD |
| mrna_polya_tail_length | Encoded poly(A) tract should sit inside the design window | HARD |
| mrna_cap_compatibility | Transcript 5' end must be compatible with the selected capping chemistry | HARD |
| mrna_dsrna_propensity | Transcript should not show a high dsRNA-forming propensity index | GUIDELINE |
| mrna_runoff_template_integrity | IVT template must support run-off transcription to a defined 3' end | HARD |
| gateway_site_integrity | Gateway vectors must carry their recombination sites (attP + ccdB / attL / attR + ccdB) | HARD |
| lentiviral_elements_present | Lentiviral transfer vectors must carry LTRs and the packaging signal (cPPT / WPRE raise titre and expression) | HARD |
Severity counts: 16 hard, 7 guideline. Rule ids and titles are read verbatim from the registry — nothing on this list is illustrative.
Rules that depend on an optional component (RNA folding, for example) return NA rather than PASS when that component is absent.
PROVENANCE — 005
517 calculations, each one labelled.
Every quantity Nucleora computes is registered with a formula, a code reference, a source list and a status. The status says what kind of thing the number is — a published method, a physical definition, a library lookup, a deterministic computation, or a heuristic.
239 of the 517 carry the heuristic flag. That is 46% of everything the application calculates, and the application says so at the point of use: a heuristic result carries a disclosure in the interface naming what it approximates and why.
Source: data/provenance_registry.json, last updated 2026-08-14. All 517 entries carry a non-empty sources list.
Seven segments, 517 entries, no remainder. The registry uses 12 distinct status values; the six largest are named above, and the remaining 15 entries carry compound labels — PUBLISHED-inspired / HEURISTIC (10), LITERATURE (1), PUBLISHED-inspired / ANCHOR (1), ANCHOR / HEURISTIC (1), HEURISTIC / ANCHOR (1) and LIBRARY / HEURISTIC (1).
The separate is_heuristic boolean is true for 239 entries. 207 of those carry status HEURISTIC outright; the other 32 carry a different status — ANCHOR (10), PUBLISHED-inspired / HEURISTIC (10), PHYSICS_DEFINITION (4), LIBRARY (2), PUBLISHED (2) and four one-off compound labels — but are flagged heuristic because the formula behind them is approximate.
PLATFORM — 006
Nothing to install. Nothing left on your laptop.
Nucleora runs as a hosted service. You work in a browser or the desktop app, and the computation happens on Nucleora's servers.
Hosted, not installed
Open app.nucleora.org in a current browser. The 422-module engine, the codon tables and the constraint registry run server-side, so every tester is on the same build and there is nothing to update.
The device holds nothing
The desktop app is a window onto the same service. It stores no password, no session token and no copy of your sequences or projects. Sequences you compute are sent to Nucleora's servers; the Privacy page says where.
Every number is attributable
All 517 registered calculations expose their formula, code reference and sources from inside the interface. The 239 heuristics are labelled as heuristics where they are used.
Full workbench during testing
Closed testing, access on request. All 38 workspace views are available to testers — there is no reduced tier. Academic and commercial use are both in scope.
REFERENCE DATA — 007
39 codon tables. 194,152 coding sequences.
Tallied directly from RefSeq CDS records, not modelled or interpolated. Each table reports the number of coding sequences behind it, so you can see which ones are thin before you rely on them.
| Organism | CDS tallied |
|---|---|
| Arabidopsis thaliana | 27,203 |
| Danio rerio | 24,939 |
| Rattus norvegicus | 22,804 |
| Mus musculus | 22,121 |
| Cricetulus griseus | 20,316 |
| Homo sapiens | 20,015 |
| Caenorhabditis elegans | 19,963 |
| Drosophila melanogaster | 13,882 |
| Saccharomyces cerevisiae | 5,955 |
| Komagataella phaffii | 5,032 |
| Escherichia coli str. K-12 substr. MG1655 | 4,286 |
| Mycobacterium bovis | 1,229 |
| Mycobacterium tuberculosis | 472 |
| Acinonyx jubatus | 250 |
| Anolis carolinensis | 250 |
| Bos taurus | 250 |
| Elephas maximus | 250 |
| Gorilla gorilla | 250 |
| Gymnogyps californianus | 250 |
| Ovis aries | 250 |
| Panthera leo | 250 |
| Phascolarctos cinereus | 250 |
| Sus scrofa | 250 |
| Tursiops truncatus | 250 |
| Vulpes vulpes | 250 |
| Equus caballus | 249 |
| Panthera tigris | 249 |
| Canis lupus familiaris | 248 |
| Felis catus | 248 |
| Loxodonta africana | 248 |
| Macaca mulatta | 248 |
| Mustela putorius furo | 248 |
| Ursus maritimus | 248 |
| Ailuropoda melanoleuca | 247 |
| Ambystoma mexicanum | 247 |
| Pan troglodytes | 241 |
| Ceratotherium simum | 201 |
| Giraffa camelopardalis | 7 |
| Procyon lotor | 6 |
Two of the 39 tables are built from fewer than ten coding sequences — Giraffa camelopardalis at 7 and Procyon lotor at 6. A table that thin is not a reliable usage distribution, and Nucleora shows the CDS count next to the table rather than hiding it.
Eleven of the tables, including human, mouse, rat, CHO, zebrafish, yeast and E. coli, are tallied from a whole RefSeq genome assembly, and each records its assembly accession. Where a species has no table, Nucleora names the proxy it falls back to instead of substituting one silently.
Alongside the 39 usage tables, the build ships 4 codon-pair-score tables (human, mouse, yeast, chicken) derived from 40–60 reference sequences each, plus one placeholder file that is empty and is not used.
Browse the codon usage data →IN-BROWSER PREVIEW — 008
Paste a sequence. See a subset of the checks.
This runs in your browser on this page. It is a small preview of composition and sequence checks — not the 23-rule constraint engine, which requires the application.
Runs entirely in your browser; nothing is sent anywhere. Input is capped at 2,000 nt.
LIMITS — 009
What this is not.
Design software can be wrong in ways that only show up at the bench or in a subject. These are the boundaries of what Nucleora can tell you.
Not a medical device
Nucleora is research and design software. It is not a diagnostic, not a therapeutic, and not a component of one. No output of this software is clinical advice or a clinical decision.
No regulatory clearance or certification
Nucleora holds no FDA clearance or approval, no CE marking, no ISO certification and no SOC 2 attestation. It has not been validated against any of them. The application implements an e-signature and chain-of-custody workflow modelled on 21 CFR Part 11 §11.50 and §11.200; implementing a workflow is not the same as being a validated system, and binding it to your organisation's identity provider and validating it remains your responsibility.
It does not assess biosafety
Nucleora performs no biosafety, biosecurity, select-agent or dual-use screening. It will not tell you whether a construct is hazardous, whether a sequence is controlled, or whether the work is permitted at your institution. It does not substitute for synthesis-provider screening.
It does not predict behaviour in a living system
Off-target effects, immunogenicity, integration, toxicity, persistence and expression in a cell, animal or person are outside what this software models. Folding, yield and manufacturability figures describe the molecule, not the organism it is put into.
Every construct goes through your own IBC
A passing constraint check is a statement about a design document, not an authorisation. Review, approval and oversight of the work remain with your institutional biosafety committee and the applicable regulators in your jurisdiction.
46% of its calculations are heuristics
239 of the 517 registered calculations are labelled heuristic. They are useful for comparing options and ranking designs. They are not measurements, and they should not be the sole basis for an irreversible decision.
Nucleora is in closed testing at version 0.11.0. Interfaces, defaults and outputs are expected to change between releases.
If a figure on this site cannot be reproduced from the build, we would like to know: [email protected].
STATUS — 010
Where the build stands.
| Version | 0.11.0 — from the repository VERSION file |
|---|---|
| Stage | In development. Closed testing, access on request. No public release has shipped. |
| Platforms | A web browser, at app.nucleora.org. A desktop app is built but not yet code-signed, so it is not offered for download. |
| Folding engine | ViennaRNA, a declared dependency of the server build (requirements.txt). If it is unavailable, folding-dependent rules report NA rather than passing. |
| Export formats | GenBank, FASTA, SBOL 2, SBOL 3 |
| Benchmarks | Not published on this page. No timing, accuracy or throughput figure is claimed here because none has been measured under a stated protocol. |
QUESTIONS — 011
Common questions.
Nucleora is in closed testing. Access is granted on request to academic and commercial users. Testers get the full workbench — all 38 workspace views, no reduced tier and no feature gating. Use the form below and we will reach out when a seat is available.
To Nucleora's servers. Nucleora is a hosted service: assembly, constraint checking, codon optimisation, folding and IVT simulation run server-side, so the sequences you work on are sent over HTTPS to a Cloudflare gateway and on to Nucleora's computation service on Google Cloud. Projects you save are stored in Nucleora's database, linked to your account. The desktop app keeps no copy on your computer. The Privacy page sets out each step. The free calculators on the Tools pages are different: they compute in your browser and send the sequence nowhere.
A web browser and a tester account. The application runs at app.nucleora.org, so the operating system does not matter. A desktop app is built but not yet code-signed, so it is not offered for download, and no date is being given for it.
The underlying tables are tallied from RefSeq CDS records rather than modelled, and every table carries the number of coding sequences behind it: 194,152 in total across 39 species, ranging from 27,203 down to 6. Two tables are thin enough that we say so on this page. How well an optimised sequence expresses in your system is an empirical question this software does not answer; see the Methods page for how the tables are built.
Nucleora is not a validated system and holds no regulatory clearance or certification of any kind. It records an append-only chain of custody and supports an e-signature workflow modelled on 21 CFR Part 11, which may be useful as an input to a regulated process — but validating it for that use, binding signatures to your identity system, and meeting the applicable requirements are your organisation's responsibility, not something this software confers.
All 23 rules are listed by name and severity in the Constraint engine section above — the list is read from the registry, so it is the complete set rather than a selection. They cover translation initiation, reading-frame integrity, GC windows, restriction-site and methylation collisions, cryptic poly(A) signals, viral packaging capacity, promoter/host coherence, and mRNA-specific checks for uORF burden, poly(A) tract length, cap chemistry and dsRNA propensity. Each verdict comes with the evidence and citation behind it.
No. Nucleora performs no biosafety, biosecurity or dual-use screening, and it models nothing about how a construct behaves in a living system. It checks a design document against design rules. Biosafety review belongs to your institutional biosafety committee, and sequence screening belongs to your synthesis provider.