Golden Gate Overhang Fidelity Calculator - Score Your Own Junction Set

From the published Potapov/Pryor ligation-count datasets - real measurements of how often T4 ligase joins each 4-base overhang to each other overhang, made by high-throughput sequencing of ligation products. SeqBench ships a permissively-licensed mirror of that data maintained by the Edinburgh Genome Foundry, not the AGPL-3.0 potapovneb repositories. The exact dataset and its citation are shown under every result.

Because they are not computed the same way, and we say so rather than implying parity. The underlying ligation counts are the same published measurements, but the formula NEB and Potapov use to turn those counts into an aggregate fidelity percentage for a named set is not disclosed anywhere accessible, and re-deriving their published figures from the same counts did not reproduce them. What you get here is SeqBench's own transparent methodology, stated on the page: a junction's correct-partner signal divided by that signal plus every cross-reaction with another overhang in your set. Use it to see which pairs in your own set are risky and how risky relative to each other - not as a number to quote against a vendor's.

The one matching the enzyme you will actually digest with, if a dataset exists for it; otherwise generic T4. This matters more than any other choice on the page. The same overhang set can score very differently between datasets, because one-pot conditions with a specific Type IIS enzyme are not the same reaction as generic T4 ligation, and treating a generic-T4 figure as if it described your BsaI-HFv2 one-pot reaction is the main way to be confidently wrong here.

Because fidelity here is defined relative to the set. The denominator is the correct signal plus the cross-reactions with the other overhangs in the same tube, so a new overhang that cross-reacts with an existing one lowers that junction's score even though neither sequence changed. That is the honest behaviour for a one-pot reaction - an overhang is only as specific as its competition - and it is why you should re-score after every substitution rather than assuming the rest of the set held still.

Look at the risky-pairs list rather than the single overhang, because a low score always comes from a specific competitor. Replace one member of the pair - usually the one that is easier to change in your part design - and re-score. Some overhangs are intrinsically poor partners regardless of set, and some are poor only against a particular neighbour; the per-overhang table's worst-partner column tells you which case you are in.

That varies enormously and it is shown deliberately. The table prints the raw correct and cross-talk counts next to each percentage, because a fidelity computed from a few counts is far less certain than one computed from thousands. A handful of overhangs in the 256-member matrix have very thin diagonals, and a set containing one of those can have its aggregate figure dominated by a single sparsely-measured cell.

No, and the comparison is there to make that concrete rather than to crown a winner. A published set was optimised for a particular size and enzyme; scored against a different dataset, or truncated to the subset you actually need, it can do worse than a set you picked yourself. Both sets are scored identically here so the comparison is fair, but neither figure is a guarantee about your reaction.

The overhangs you paste are scored server-side and nothing is persisted. golden_gate_fidelity is also callable from the REST API and the MCP server, and the response carries the dataset list and the named sets so a script or an agent can enumerate the options without hard-coding them.