Check your probing data against a structure

For SHAPE or DMS data: does the structure a sequence folds to agree with your reactivities?

Ask: "Using rnafold, check my SHAPE profile against this sequence's predicted structure." and give the sequence and one reactivity per nucleotide (negative where there is no data).

The assistant calls fold_rna, keeps the dot-bracket structure (and notes the gap between the minimum-free-energy and ensemble energies: a large gap means the printed structure is a poor summary), then check_probing_agreement.

The input below is synthetic. scripts/make_guide_examples.py generates reactivities by adding seeded noise around the predicted pairing, then plants two conflicts (positions 6 and 34) so the tool has something to find. It is not measured data. Your own measurements replace it.

Output (examples/probing_agreement.json):

  • 63 positions had both a structural call and a reactivity outside the ambiguous band; 13 were ambiguous (between 0.4 and 0.85) and none had no data.
  • Position 6 is paired in the structure but reactive (1.2) in the data: "Those are the ones worth looking at first: reactivity is hard to explain away for a base that is paired."
  • Position 34 is unpaired in the structure but unreactive (0.02): "weaker evidence: stacking, a protein contact or burial all protect an unpaired base."

How to read it

  • It reports positions, not a number. By rule there is no merged agreement index: "A single number mixing a chemical measurement with a geometric one is not validated by anything."
  • "Reactivity is not pairing." A base can be unreactive because it is paired, stacked, protein-bound or buried, so a flagged position is a question about both your data and the structure.
  • A good next step is to sample the structures with your data as a constraint: rna_ensemble with reactivities set, beside the unconstrained ensemble, to see which states your data supports.

Tools used

fold_rna, check_probing_agreement, rna_ensemble (all free).

Last changed 2026-11-17