May 09 2026 How to interpret SHAPE and chemical probing data for RNA structure decisions

SHAPE and related chemical probing assays are often described as if they directly reveal RNA structure. That is too simple. These experiments measure nucleotide flexibility under a defined condition, and the structural interpretation remains indirect.

A reactive nucleotide is not automatically unpaired, and an unreactive nucleotide is not automatically base-paired in a particular helix. Tertiary contacts, protein binding, ligand occupancy, local stacking, and co-existing conformations can all affect the signal. The relevant question is therefore which structural interpretations remain plausible once the probing data are taken into account.

This distinction becomes most important when the computational model is ambiguous. If sequence-based folding yields several candidate structures within a narrow free-energy range, probing data can suppress weak alternatives and sharpen the analysis substantially. If the relevant uncertainty lies elsewhere, for example in folding kinetics, long-range interactions, or a protein-bound state, the same data may still be informative, but they will not repair an incomplete model.

SHAPE directed RNA folding with the ViennaRNA Package has remained relevant for exactly that reason. Reactivities can be translated into folding constraints in different ways, and the choice affects the result.

The same probing profile can influence a structure prediction more strongly or more weakly, depending on how the data are processed and which parameters are used. The Deigan et al. method is probably the most widely used approach for converting probing data into pseudo-energies, but its performance depends on the chosen parameters. Those parameters are not universal.

We discussed this broader framework in Predicting RNA structures from sequence and probing data. Thermodynamic RNA folding remains useful because it keeps alternative structures explicit. Probing data can then shift the balance between those alternatives toward what the molecule appears to do under the measured condition.

Two questions are often mixed together. One is whether probing data improve a structure prediction at all. I discuss that question in When SHAPE data actually improves RNA structure prediction. The other is what those data allow one to conclude. In practice, the second question is often the more important one.

DMS, SHAPE-MaP, DMS-MaPseq, and related workflows differ in chemistry and readout, but the interpretive issue is similar. They are most useful when they are used to test whether a proposed structural explanation remains credible once measured nucleotide flexibility is taken into account.

This becomes particularly relevant when RNA structure predictions are combined with automated pipelines, large benchmarks, and more complex prediction workflows. A good score on a benchmark does not remove the need to interpret probing data carefully. Conversely, probing data do not remove the need to ask whether the computational model misses the relevant mechanism.

Probing data are most helpful when the structural question is well posed, the experiment matches that question, and the computational interpretation remains conservative about what is still unresolved. Otherwise the analysis can start to look more definitive than it is.

That is often the point at which groups benefit from a technical review or a focused training session. The issue is usually whether the current design, probing setup, and computational interpretation are coherent enough to support the next experimental decision. That is exactly the kind of question I take up in design reviews, workshops, and advisory work.