RBS Designer — Ribosome Binding Site Calculator & 5' UTR Design
Translation initiation rate at each start codon; 5' UTR design for a target expression level.
Predicted, not measuredSpearman ρ = 0.39 on two 5' UTR datasets it was not fitted to
- How good is it?
- Good at telling a strong RBS from a weak one, and not good enough to order two similar ones: across fourteen measurement systems it ranked a pair correctly 53% of the time when the two predictions were less than 1.2-fold apart — a coin toss — against 87% above 100-fold, with individual systems varying widely either side. The scale is arbitrary, so compare candidates from one run rather than reading an expression level off the number.
- Only valid for:
- translation INITIATION only, in E. coli-like Gram-negative hosts (the model is parameterized on the E. coli anti-Shine-Dalgarno sequence). Rankings within one construct context; the absolute value has no units and no meaning.
Paste a bacterial mRNA and get the predicted translation initiation rate at every start codon in it, with the free-energy breakdown behind each number: Shine-Dalgarno hybridization to the 16S rRNA, the cost of unfolding the structure over the site, the SD-to-start spacing penalty, the standby-site term and initiator-tRNA binding. Give this ribosome binding site calculator a coding sequence instead and it builds a spread of Shine-Dalgarno cores and spacings, scores each one against your own CDS, since the rate depends on how the RBS folds with that specific CDS and cannot be read off a parts table, and ranks them. Add a target rate to rank by closeness rather than raw strength, or your current 5' UTR for a baseline and a fold-change per candidate.
Base-pairing over the ribosome binding site
A rate above is computed from the site itself, so a site that scores well can still be folded shut. This finds every stretch of the same transcript that is reverse-complementary to the ribosome's footprint, with the length and melting temperature of each duplex.
Every stretch of the transcript that is reverse-complementary to the ribosome's footprint, with the Tm of each duplex. Base-pairing arithmetic, not a fold.
▸Compare the ligand-bound and ligand-free states
Folds the transcript twice: once freely, and once with the aptamer held out of the structure as a bound ligand would hold it. Reports the effect on the ribosome binding site. Enter the aptamer's coordinates; no sequence feature identifies one.
Design, edit, and verify complete constructsIt opens on the translation of this sequence, in all six frames.
How to use the RBS Designer tool
- 1Pick a mode: "Predict rate" to score an existing mRNA, or "Design an RBS" to generate and rank candidates for a CDS.
- 2Paste your sequence: an mRNA (5' UTR plus the start of the CDS) to predict, or a CDS beginning at its own start codon to design. DNA and RNA spellings are both accepted.
- 3Optionally add your current 5' UTR for a baseline fold-change, a target rate to design toward, your promoter's real transcribed leader, or a non-E. coli anti-Shine-Dalgarno sequence.
- 4Read the ranked candidates and the ΔG breakdown, and copy the top candidate's full 5' UTR to clone.
FAQ
Which model produces these numbers?
OSTIR, with free energies from the ViennaRNA Package. Its coefficients were re-fitted against ViennaRNA's energy parameters, so its values are not interchangeable with numbers from RBS Calculator v2; do not mix the two in one comparison.
What are the units of the predicted rate?
There are none; the scale is arbitrary. The ratio between two predictions from the same run is the meaningful quantity ("this candidate is about 8x the one I have now"); the absolute number is not a protein concentration and cannot be converted into one.
How accurate is it?
Accurate enough to tell a strong RBS from a weak one, and not accurate enough to order two similar ones. Measured over fourteen published libraries, hosts and reporters spanning six bacterial species, and stated against the ratio of the two predicted rates, the stronger candidate is ranked first 53% of the time when the two predictions are less than 1.2-fold apart (50% is a coin toss, so a gap that small carries essentially no information), 64% at 2–3-fold, and 87% above 100-fold. Individual systems vary widely around those averages, so read them as a guide to when a gap stops being trustworthy rather than as odds on your own pair. In practice: trust a large gap, treat a small one as unresolved, and build and measure a design series whose members sit within a few fold of each other.
How are the design candidates generated?
By combining a spread of Shine-Dalgarno cores, from full complementarity to the anti-SD down to minimal, with SD-to-start spacings across the biologically relevant range, then scoring every combination in the context of your actual CDS. No strength is asserted for any candidate sequence in advance. The spacer is poly-A so that varying the spacing does not also introduce new secondary structure; check the returned sequence if your cloning strategy needs a particular site inside the UTR.
Why does it ask for a 5' leader, and what happens without one?
The standby-site term depends on the sequence upstream of the RBS, so the model needs some 5' context. If you supply none, a 20 nt unstructured poly-A leader is assumed and the result says so. For a construct-specific number, paste the real transcribed leader your promoter produces.
Does it work for organisms other than E. coli?
Partially. The model is parameterized on E. coli, but you can supply a different anti-Shine-Dalgarno sequence (the 16S rRNA 3' end) for your host, which is the single largest host-specific term. Everything else in the model stays E. coli-derived, so treat a non-E. coli prediction as a ranking heuristic rather than a calibrated rate.
What does this NOT tell me?
It models translation initiation only. It says nothing about transcription, elongation, mRNA stability, protein folding or toxicity, codon usage, or whether the designed UTR introduces a restriction site, cryptic promoter or out-of-frame upstream start codon in your final construct; check the returned sequence for those separately. The Construct QC Linter covers several of them.
Method and validation
- Model:
- Mechanistic — a physical model with fitted constants.
- Fitted on:
- A thermodynamic model whose coefficients OSTIR refitted against ViennaRNA's Turner 2004 energies over the 132 measurements of Salis et al. 2009 (Nat Biotechnol 27:946).
- Validation:
- Spearman ρ = 0.39 against measured expression on two 5' UTR datasets it was not fitted to (Gilliot & Gorochowski, Nucleic Acids Res 2024;52(13):e58); reproduced in this repo at ρ = 0.509 on an independent leave-one-context-out split over 194,636 measurements, and at a mean ρ = 0.62 over fifteen further measurement systems it was not fitted to — six species, four reporters, three readouts — on which every learned model tried here scored BELOW it (scripts/rbs-eval) — above the published figure because 65,841 rows of that compilation record no measurement at all (fluorescence mean and s.d. both exactly 0) and are excluded here rather than scored as the weakest expression observed. The widely quoted 53% within 2-fold / 91% within 10-fold are calibration residuals on the fitting set, not held-out validation.