Recode a coding sequence to remove every CpG dinucleotide while following your host's codon usage. Built for AAV and lentiviral transgenes, where CpG content drives innate sensing and silencing. Everything runs in your browser: your sequence never leaves your computer.
A coding sequence, 5′→3′, in frame: plain DNA or FASTA.
What Recode-It does. For each codon it collects the synonymous codons your codon table allows, drops any that contain CG, and drops any whose frequency falls below the rare-codon cutoff. One rule crosses codon boundaries: a codon ending in C cannot be followed by one starting with G, or the junction makes a new CpG. That leaves an enormous number of valid sequences — routinely more than 1080 for a 700 bp ORF — so rather than take the first one it finds, Recode-It counts the valid completions at every position and samples a path in proportion to codon usage, which keeps common codons common instead of merely legal.
A repair pass then removes forbidden restriction sites and over-long homopolymers by swapping individual codons, and reports anything it could not clear without breaking a hard constraint. The run is seeded from its barcode, so the same inputs and the same barcode always give the same sequence.
Limits. The main run screens for CpG, the sites you list and homopolymer runs, and nothing else. Splice donors and acceptors, branch points, polyadenylation signals, secondary structure, repeats and cryptic promoters are not considered here — the Further analysis panel screens some of them, but none of it replaces looking at the finished design. Codon tables are rounded reference values unless you paste your own.
Every constraint below is applied while the sequence is built, not patched in afterwards.
The same protein and the same codon table admit an enormous number of CpG-free sequences. This section counts them exactly, shows the GC content every one of them could have, and selects a small panel of defensible candidates to synthesise. Nothing here changes the run above or its barcode.
Columns are raw measurements, not a ranking: sort by whichever matters for your construct and host. Hard failures (residual CpG, a forbidden site, a polyadenylation signal, an over-long homopolymer) are removed from the pool rather than scored, so every row above has already passed them.
On the identity column. Panel members will align at roughly 80–95% whatever is done to them: only synonymous positions can differ, most of those differ by a single wobble base, and methionine and tryptophan cannot differ at all. High identity is the floor of the synonymous space, not a sign the panel has collapsed — judge the spread by the GC, CAI and changed columns instead. The rows are chosen to span the reachable design space: the exact CAI optimum, the fewest- and most-changed sequences, a ladder of exactly steered GC values, and the flattest GC profile found, with any remaining places filled from the Pareto front under a minimum-distance rule.
All possible sequences cannot be written out; there are far too many. This draws a reproducible random sample instead, seeded from the run barcode on a stream of its own.
Read these as candidates, not verdicts. CAI correlates only loosely with expression, and over-optimisation can reduce it; GC, repeat and motif measurements are screens, not guarantees. The panel is a set of sequences worth synthesising and testing side by side, chosen to span the available design space. Secondary structure, branch points, acceptor sites and host-specific regulatory elements are not screened here.
Changed codons are highlighted. Any codon still containing CG would be shown in red.
What each number means, and what a bad one looks like.
The chips, first. Protein identical is the one that must be green: the recoded sequence was translated and compared with your input protein. If it ever reads Protein CHANGED, do not use the sequence — that is a bug, and the run barcode identifies it. CpG-free means no CG dinucleotide survives anywhere, junctions included. Sites & runs flags constraints the repair pass could not satisfy; the note underneath says which, and where.
| Tile | What it means | What a bad one looks like |
|---|---|---|
| CpG dinucleotides | How many CG dinucleotides remain, against how many you started with. | Anything above zero with CpG elimination on. The sequence note will say why it could not be removed. |
| CpG obs/exp | Observed CpGs against the number expected from the C and G content alone. A standard way of reporting CpG suppression that does not depend on length. | Not applicable once the count is zero. On the input, above about 0.6 means a sequence that has not been CpG-depleted at all. |
| GC content | GC across the whole recoded ORF. | Below about 35% or above about 65% starts to cause synthesis problems. Watch the sliding-window plot too: a flat 50% overall can hide a 70% patch. |
| CAI vs table | Codon adaptation index against the table you chose — how closely the recoded sequence uses the codons that table prefers. | A large drop from the input is worth questioning, usually caused by a high rare-codon cutoff. A very high CAI is not automatically good: over-optimisation can reduce expression. |
| Codons changed | How much of the ORF moved. Removing CpGs from a CpG-rich sequence typically changes a quarter to a half of the codons. | Near zero when the input had many CpGs means the constraints blocked almost every swap. Check the cutoff. |
| Length | Unchanged by definition — recoding is synonymous. | Any change at all. Report it with the barcode. |
The plots. The GC trace shows input and recoded side by side in a 60 nt window; you are looking for the recoded line to stay inside a reasonable band rather than to sit at any particular value. The CpG track shows where the CpGs were and where they now are — the lower track should be empty.
The sequence view. Changed codons are highlighted. Any codon still containing CG would appear in red, which should never happen with CpG elimination on.
The exports. FASTA for ordering, GenBank for Annotate DNA and most plasmid editors, and the run record, which is the only file that lets anyone reproduce the sequence from scratch. Keep the record with anything you order. There is more on barcodes, versions and citing a run in the help and FAQ.
For research use only. The recoded sequence is back-translated and compared with the input protein, but splice signals, polyadenylation signals, repeats and secondary structure are not screened. Verify critical designs before ordering synthesis.
Send the GenBank output to Annotate DNA to map the features, check your cloning sites in Cleave-It, and simulate the build in Clone-It. Each opens in a new tab with the recoded sequence already in its box.
Recode-It runs entirely in your browser: your sequence is not sent to our server or stored anywhere. Codon frequencies come from the table you choose; the built-in tables are rounded reference values. We count how often this page is opened, and nothing else.