Which strategy should I use?

For a synthesis order, most-frequent or usage-weighted: the first gives each residue its host's single most-used codon, which produces runs of one codon per family, while the second draws from the real frequency distribution and looks more like a natural gene. For a primer or probe against a conserved motif, use the degenerate strategy — it returns one IUPAC-coded codon covering every codon for the residue, which is not a single sequence at all.

Does it optimize the sequence?

No. There is no restriction-site avoidance, GC targeting or repeat removal here — that is constraint-based codon optimization, and it already ships as the Codon Optimizer. The intended flow is to back-translate, open the product as a sequence, then optimize it under constraints.

Where do the codon frequencies come from?

A live count over NCBI RefSeq coding sequences, the Kazusa mirror, or a bundled reference host — you choose the source, the organism and the counting basis, and the readout says which one produced the numbers on screen.

Is a sampled back-translation reproducible?

Yes. The sampler takes a seed, so the same seed returns the same sequence and a different seed returns a different one.

Design

Back-translate: protein → DNA, codon-usage weighted or degenerate

Take a protein and get the DNA that would encode it, weighted by a real codon-usage table rather than by the first codon in a list. Four strategies answer four different questions — the host's most frequent codon per residue, a draw from the family's real frequency distribution, one IUPAC-degenerate codon covering the whole family, or the table's first codon — and every residue's choice is shown with its share of its synonymous family. The product opens as a GenBank record carrying the strategy, genetic code and table it came from.

Open Back-translate: protein → DNA, codon-usage weighted or degenerate Open the Protein tools sample

Real CDS plasmids (GFP · TetA/Bla · ADH1 · a His6-TEV expression vector), twelve UniProt proteins, one with unknown residues, bacteriorhodopsin for membrane topology and preproinsulin for signal peptides, for the domain scanner, protein analysis, disorder, protease digest and back-translation

What it does

01Four back-translation strategies, none of which subsumes another: the host's most frequent codon per residue (EMBOSS backtranseq), sampling from the family's real frequency distribution, one IUPAC-degenerate codon covering every codon for the residue (EMBOSS backtranambig), and the table's first codon (DnaChisel reverse_translate)
02Codon weights from a live NCBI RefSeq count, the Kazusa mirror or a bundled reference host — the same table registry the Codon Optimizer uses, so a table fetched there is the table used here
03Genetic code selection, across the codes with a real ported backtranslation table
04A random seed, so a sampled back-translation is reproducible and re-rollable
05Every residue's chosen codon with its share of its synonymous family, and the whole family table with the choice marked
06Residues the genetic code has no codon for are emitted as NNN and named, rather than silently mistranslated
07Runs on the canvas selection or a feature, and clicking a residue selects the codons it produced
08The product opens as a real GenBank record with a CDS feature and a provenance line naming the strategy, genetic code, table and organism

What it runs on

The three table-driven and degenerate strategies are a port of EMBOSS 6.6.0 backtranseq and backtranambig (ajax/core/ajcod.c), golden-validated byte-for-byte against those routines compiled from the EMBOSS source over 54 cases covering three genetic codes, two usage tables and six proteins; upstream's disagreements between the two programs are reproduced deliberately. The fourth strategy is DnaChisel's reverse_translate, asserted identical to the pre-existing port. Every degenerate codon is cross-checked against the editor's own IUPAC lookup. The panel runs on @ports/codon, @ports/optimize.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

Questions

Which strategy should I use?: For a synthesis order, most-frequent or usage-weighted: the first gives each residue its host's single most-used codon, which produces runs of one codon per family, while the second draws from the real frequency distribution and looks more like a natural gene. For a primer or probe against a conserved motif, use the degenerate strategy — it returns one IUPAC-coded codon covering every codon for the residue, which is not a single sequence at all.
Does it optimize the sequence?: No. There is no restriction-site avoidance, GC targeting or repeat removal here — that is constraint-based codon optimization, and it already ships as the Codon Optimizer. The intended flow is to back-translate, open the product as a sequence, then optimize it under constraints.
Where do the codon frequencies come from?: A live count over NCBI RefSeq coding sequences, the Kazusa mirror, or a bundled reference host — you choose the source, the organism and the counting basis, and the readout says which one produced the numbers on screen.
Is a sampled back-translation reproducible?: Yes. The sampler takes a seed, so the same seed returns the same sequence and a different seed returns a different one.

Also searched as

back translate protein to DNA · reverse translate protein sequence · protein to DNA sequence converter · amino acid sequence to DNA online · EMBOSS backtranseq online · EMBOSS backtranambig online · degenerate codons from a protein sequence · IUPAC degenerate primer from conserved motif · most frequent codon back translation · codon usage weighted back translation · SnapGene back translate alternative · protein to gene sequence for synthesis