ViennaRNA 2.8.0 was released on 6 November 2026. Cross-linking experiments can now guide a structure prediction through interval-pair constraints, equilibrium probabilities for base pairs, unpaired stretches and whole helices are available over an entire sequence without a sliding window, and the parameter files for pseudouridine and N1-methylpseudouridine ship inside the library. The release also brings a new documentation site, design seeds for RNAinverse, and a set of fixes that are worth knowing about if you work with G-quadruplexes or modified bases.
ViennaRNA in practice, a series of tutorials. RNAfold tutorial · SHAPE and chemical probing data · RNAplfold and accessibility · 2.8.0 release note (this post)
ViennaRNA 2.8.0 was released on 6 November 2026. Twenty-six additions, and three of them change what kinds of experimental evidence a folding calculation can take and what questions you can ask of an ensemble. This is a short tour of what the release makes possible, written for people who use the package rather than build it.
Cross-linking experiments can now guide a prediction
Probing data have been usable in ViennaRNA for a decade, in the form of per-nucleotide reactivities from SHAPE or DMS. 2.8.0 adds a second kind of evidence: interval pairs from cross-linking experiments, selected with --sp-strategy=X in RNAfold. Rather than a number per position, the input is a pair of stretches that an experiment says were close enough to be joined, and the prediction is guided towards structures consistent with that.
For anyone sitting on data from a proximity-ligation method this is the first time the evidence goes into the folding itself rather than into a figure beside it.
Underneath it is a new constraint type of its own, interacting interval pair soft constraints, reachable from command files and from the library as well as through the RNAfold flag. Anything that produces evidence about two regions rather than one position can use the same machinery.
The structure probing input is also tidier than it was. --sp-strategy now selects how reactivities become energies, --sp-preprocess selects how the data are prepared first, and both take their parameters more flexibly on the command line. The three SHAPE methods the probing-data tutorial works through are still there and are now members of a family. One related change is worth having: fractional soft constraints are preserved for partition function calculations, where reactivity-derived pseudo-energies used to be rounded to integers.
Equilibrium probabilities over a whole sequence, and for whole helices
Three new functions, vrna_eqm_probs_bp(), vrna_eqm_probs_up() and vrna_eqm_probs_hx(), return equilibrium probabilities for base pairs, unpaired stretches and helices across an entire sequence, with a callback-based API for the same quantities. The release notes describe this as accessibility computations in global scope.
Two things here are new rather than rearranged. The first is the word global. The RNAplfold tutorial in this series computes accessibility in a sliding window, where the window size and the maximum base pair span are part of the answer and a region cut out of its context stops being the same object. These functions compute the quantity over the whole molecule with no window to choose. For a sequence short enough to fold as a whole that removes two parameters from the result.
The second is helix probabilities. Base pair probabilities and unpaired probabilities have been available for years. The probability that a helix is formed is a different question, and closer to how people actually reason about a structure than a dot plot of individual pairs. Having it as a primitive rather than something to assemble by hand removes a step from a lot of scripts.
These are library functions rather than new command-line flags, so they are for people writing against RNAlib or its Python bindings. The callback form is there because the quantities get large and collecting them all into memory is not always what you want.
Modified nucleotides are no longer a side quest
2.8.0 ships parameter files for pseudouridine and N1-methylpseudouridine, adds dihydrouridine parameters for pairs with G, adds G to the pairing partners of pseudouridine, and includes every modification parameter file inside RNAlib instead of leaving them as separate downloads.
The last part is the one that changes daily practice. A calculation on modified RNA used to begin with finding the right parameter file; it now begins with asking for it. For therapeutic mRNA the relevance is immediate, since a drug substance fully substituted with N1-methylpseudouridine, folded against uridine parameters, is a calculation about a molecule nobody will make.
Design, suboptimal folding, benchmarks
RNAinverse gains LinearBPDesign sequence seeds through a new --init-strategy option, which changes where an inverse folding run starts rather than how it searches. Wuchty-style RNAsubopt now supports hi-motifs. The structure benchmark in RNAfold and RNAalifold gained relaxed metrics, an option to ignore compatible base pairs, and a changed output format, which makes it more useful for comparing predictors honestly.
A documentation site worth reading
The documentation was rewritten into a stand-alone website, built with pydata-sphinx-theme, sphinx-design and sphinx-contributors. It now carries tutorials for eleven of the programs alongside the man pages and the API reference, which is a considerable step up from a manual.
One practical note: it moved every URL. The man pages that used to live at viennarna.readthedocs.io/en/latest/man/ are now at viennarna.readthedocs.io/latest/usage/man/, and the old paths return 404 rather than redirecting, so any bookmarks and any links in your own notes need updating. Every link in this series was changed for it.
What is different in the numbers
Twenty-eight fixes came with the release, and a handful of them change results rather than behaviour. If one of these describes your work, it is worth recomputing before citing an older number.
G-quadruplexes. The energy parameters had the wrong units, and two further G-quadruplex problems in Wuchty-style RNAsubopt were fixed in the same release.
Modified bases. The dangling-end and mismatch corrections were wrong, the correction for bulges of length one was wrong, and base pair hard constraints were applied incorrectly. RNAfold also differed from the equivalent API calls, because the executable was passing the unmodified sequence.
Wuchty-style suboptimal folding was wrong with G-quadruplexes, with circular RNAs, and for dangles=1 and dangles=3. Multibranch loop soft constraints in the same code path were fixed too.
Energy evaluation. The split of internal loops into stacks, bulges and real internal loops was corrected, which touches RNAeval and anything that scores a given structure. Separately, parameter retrieval after reloading parameters from a file returned the wrong data structures, so a pipeline that switches parameter sets mid-run was not necessarily using the set it asked for.
Two deprecations are worth planning for. Python 2 support is gone from the SWIG interface, so 2.7.2 is the last release that builds those bindings. All global variables are deprecated as well. They still work today, and code that sets dangles or temperature through a global rather than through a model details structure should be migrated while the deprecation is only a warning.
Worth the upgrade
Take 2.8.0. The cross-linking strategy is the headline if you have interval data, the global equilibrium probabilities are the one to look at if you work through the library, and the modification parameters matter to anyone whose RNA is not made of four standard nucleotides. Then ask whether you have used G-quadruplexes, modified bases, Wuchty-style RNAsubopt or RNAeval, and recompute those results if you have.
Sources: ViennaRNA 2.8.0 release notes · ViennaRNA documentation · RNAfold man page, structure probing options · changelog.