DomainCuts
Big proteins are built from smaller domains — compact folded units strung along the chain. Domain databases carve the chain into these domains automatically: TED (2024) across the AlphaFold Database, alongside the long-running CATH classification. This game asks you to do it: first carve a chain into its domains, then judge how much to trust where those domains sit relative to each other. The twist: a model can be crisp (high confidence) in every domain and still be unsure of the arrangement — only the PAE shows that. Everything runs locally; nothing about you is collected.
This is the tool. The carve is scored against the domains TED found in each AlphaFold model; the trust answer is computed live from the model’s AlphaFold PAE matrix, fetched in your browser. Every protein and its domains are listed below (works without JavaScript).
Loading the game… If it does not start, the full domain guide below still works.
How to play. Press Start. Move the cursor (slider or ←/→) and press Drop cut (Enter) to place the domain boundaries, then Score the carve. Next, judge the domains’ relative placement — pick the pair the model is least sure about (or Confident/Uncertain for a two-domain protein) with the number keys — and the live PAE settles it. Press Enter or Next to continue.
Why the cut points matter
In Src and ABL1 the SH3 and SH2 domains clamp onto the back of the kinase domain to hold it switched off. In chronic myeloid leukaemia the BCR-ABL1 fusion loses ABL1's own N-terminal cap, part of that brake, so the kinase stays on.
Two different questions: pLDDT vs PAE
pLDDT asks “how sure is the model of the local shape here?” — it is high inside a well-folded domain. PAE (predicted aligned error) asks a different question: “if I line up on residue i, how sure am I of where residue j is?” Between two domains joined by a flexible hinge, PAE is large even when both domains are crisp. That gap — confident domains, uncertain arrangement — is the whole point of this game.
The proteins and their domains
Answer key — open it after you play
| Protein | Length | TED domains | Note |
|---|---|---|---|
| c-Src kinase P12931 A signalling kinase built from an SH3, an SH2 and a kinase domain after a floppy N-terminal tail. TED counts the kinase domain's two lobes as two domains. The domains clamp together to switch the kinase off — a famous regulated arrangement. | 536 aa cuts count ± 27 | SH3 (87–147, consensus high) · SH2 (157–246, consensus high) · Kinase N-lobe (261–343, consensus medium) · Kinase C-lobe (348–524, consensus medium) | The N-terminal tail (SH4 + unique region) is disordered, not a folded domain. TED puts residues 1–86, 247–260, 525–536 in no domain. |
| Abl1 tyrosine kinase P00519 The kinase behind chronic myeloid leukaemia (the drug target of imatinib). Same SH3–SH2–kinase cassette as Src, then a very long C-terminal region. TED joins its SH3 and SH2 into one domain and counts the kinase domain's two lobes as two. | 1130 aa cuts count ± 57 | SH3 + SH2 (one TED domain) (64–215, consensus medium) · Kinase N-lobe (236–317, consensus medium) · Kinase C-lobe (321–510, consensus medium) · F-actin-binding domain (1028–1098 + 1110–1128, consensus high) | The long C-terminal region is not one compact domain: largely flexible, but carrying binding motifs, a DNA-binding region, nuclear localisation signals and a folded F-actin-binding domain at the very end. TED puts residues 1–63, 216–235, 511–1027 in no domain. |
| Calmodulin P0DP23 A calcium sensor with two lobes joined by a single long helix. The linker is so flexible that AlphaFold is unsure how the two lobes sit relative to each other — even though each lobe is crisp. | 149 aa cuts count ± 7 | N-lobe (1–78, consensus high) · C-lobe (94–146, consensus high) | The two lobes are connected by a flexible central helix — the textbook flexible inter-domain linker. TED puts residues 79–93 in no domain. |
Methods & about
Models and their PAE matrices are fetched live from the AlphaFold Database; the trust score is the mean inter-domain PAE computed in your browser (the same quantity AlphaFold’s own PAE plot shows): under 10 Å reads as confident, over 15 Å as uncertain (teaching cut-offs, the same pair HingeHunt and TwoIslands use). The domains are TED’s (The Encyclopedia of Domains: API 0.1.0, on AlphaFold DB v4 models, read 2026-10-22), a computer consensus of three methods, not an experiment, built on the AlphaFold models themselves; each domain shows its consensus level. A cut counts when it falls between two TED domains, give or take a tolerance scaled to the chain (its length divided by 20, at least 3 residues). Fibronectin (FN1), suggested in our concept census, is a beads-on-a-string of ~29 repeated modules — not a clean discrete carve — so calmodulin (two lobes on a flexible linker) stands in as the gold example of confident-domains-but-uncertain- arrangement. The names and notes are written by hand and science-checked; the ranges come from TED. No accounts, no storage, no servers of ours.
Want to see PAE for any structure yourself? Open the free AlphaFold viewer. Everything we make is free and non-profit; learn more on the home page.