// Example project

tRNA-Phe: does G10A change the fold?

An example project built from real tool output on a public sequence. The notes show the kind of thing you or your assistant might write; the fold below ran on six engines.

1 fold · 3 notes

Question: does G10A change the predicted structure of yeast tRNA-Phe (76 nt, a public sequence)? In the wild-type prediction G10 pairs with C25 and closes the D stem (positions 10-13 paired with 22-25).

noteadded here

Free 2D comparison, one energy model at 37 C. Minimum free energy -22.4 to -22.1 kcal/mol; ensemble free energy -24.13 to -23.93. The printed minimum-free-energy structure loses the D stem while the energy barely moves (0.3 kcal/mol). In 1,000 sampled structures the largest cluster holds 35.4% of the wild type's samples and 45.5% of G10A's. This is a calculation, not a measurement: no tertiary contacts, magnesium or modified bases.

notefiled by your assistant

done

fold of positions 1-28 (28 nt)

6 engines · they differ at 10 of 28 residues

foldfiled by your assistant

The fold in full

What ran

EngineStructuresTime
boltz2327 s
chai1353 s
opendde516 s
openfold3_apo930 s
rf3314 s
rhofold14 s

Listed alphabetically. Whole fold: 2.8 min.

Where the engines agree

18 of 28 residues have the same pairing state in every engine; 10 differ, in 6 regions: 1–2, 11, 14–15, 20–21, 24, 26–27.

How far apart they sit

EngineDistance to boltz2 (Å)
chai12.09
opendde1.53
openfold3_apo12.60
rf37.59
rhofold6.56

RMSD in angstroms after superposing each engine on boltz2, listed alphabetically: where each one sits relative to that single engine, and nothing about which is nearer the truth.

Was it in training?

Earliest close relative: 6TNA, deposited 1978 (100% identity). The closest match is 9YPY (2025).

  • boltz2: training cutoff 2023-06-01 · the earliest relative is older
  • chai1: training cutoff 2021-01-12 · the earliest relative is older
  • openfold3: training cutoff 2025-06-30 · the earliest relative is older
  • rf3: cutoff not published

A close relative (>=90% identity) deposited before an engine's training cutoff means its answer may be recall. The earliest one is what decides that, not the closest.

Read this with it

  • A summary: the full record (every engine's per-residue states, all structural checks, every close PDB match) is get_analysis(job_id).
  • Agreement between engines is not evidence of correctness: they can share a bias, and they can all be recalling a structure they were trained on. The novelty section says whether that applies here.
  • The platform offers a set of predictions, not a single answer: no engine is preferred, and nothing should be inferred from the order below, which is alphabetical.
  • Distances are to the named reference engine only. They show where engines sit relative to that one, not which is closer to the truth.

Folded the first 28 nucleotides (GCGGAUUUAGCUCAGUUGGGAGAGCGCC), which hold the D stem, on 6 engines. 18 of 28 residues have the same pairing state in all of them; 10 differ. The closest PDB match is a 2025 entry (9YPY), but the earliest close relative is 6TNA, deposited 1978, before the published training cutoff of boltz2, chai1, openfold3. Their agreement may be recall of a tRNA they have seen, not a prediction for this fragment.

notefiled by your assistant

Make one of your own

Sign up, then ask your assistant to study a sequence and tell it the project's name. Its folds and notes appear here as they finish.