Changelog

This version adds modified base support and a physics-based model to correct energy contributions depending on the concentration of mono-valent ions to our implementations. The new features are fully exposed through the command line tools and the API of RNAlib.

The new implementations for mono-valent ion concentration dependent energy corrections follows the lines of Einert and Netz, 2011, Theory for RNA folding, stretching, and melting including loops and salt., Biophys. J. 100, 2745-2753 (https://doi.org/10.1016/j.bpj.2011.04.038) and is described in Yao et al., 2023, Mono-valent salt corrections for RNA secondary structures in the ViennaRNA package, currently under revision.

The modified base support that is introduced with this release is based on the soft constraints framework and at this time integrates nearest neighbor parameters for the following non-standard nucleotides

• Inosine
• Pseudouridine
• m6A
• 7-deaza-adenosine (7DA)
• Purine (aka nebularine)
• Dihydrouridine

While the first five parameters sets have been determined experimentally through UV-melting, the last, dihydrouridine, has been predicted by us in-silico using the Rosetta/RECESS framework. A paper describing the details of the modified base implementation is in preparation and will be linked here as soon as it is published. See also the man pages of the respective programs implementing modified base support as well as the reference manual.

With this release, we started to automatically convert the documentation for the C-API to Python docstrings. As a consequence, we can now provide a Python API documentation at ReadTheDocs.

With this new release, we deactivated building the Python 2 interface by default. It is still part of the distribution tarball for those that require it but has to be activated via --with-python2 at configure time.

The list of all major changes in this version can be found below:

This version adds multi-strand RNA interaction support to our implementations, exposed to end-users in the form of the novel command line tool RNAmultifold and corresponding changes in the API of RNAlib.

The new implementations for RNA-RNA interaction can now handle multiple RNA input strands that shall form interaction complexes. For that, we implemented a generalization of the concatenation approach of RNAcofold following the ideas of Dirks et al. 2007, Thermodynamic Analysis of Interacting Nucleic Acid Strands, SIAM REVIEW Vol. 49, No. 1, pp. 65-88 https://doi.org/10.1137/060651100 and Lorenz et al., 2021, Efficient Algorithms for Co-Folding of Multiple RNAs, In: Ye X. et al. (eds) Biomedical Engineering Systems and Technologies. BIOSTEC 2020. Communications in Computer and Information Science, vol 1400. Springer, Cham. https://doi.org/10.1007/978-3-030-72379-8_10.

The major new features for multiple interacting RNAs are

• MFE prediction
• Partition function and base pair probability computation
• Equilibrium concentration computation
• Suboptimal structure enumeration within an energy band around the MFE
• Suboptimal structure enumeration ala Zuker 1989
• Free energy evaluation
• Due to the new implementations, the output order for ensemble free energies of interacting complexes slightly differ between RNAcofold and RNAmultifold with two input sequences. Apart from that, RNAmultifold can be considered a drop-in replacement for RNAcofold.

With this new release, we also adapted the covariance score computation for the prediction of consensus structures of multiple sequence alignments. In particular, the condition that decides which columns of the alignment are allowed to pair was updated to yield more sane results for edge-cases with many gaps and non-canonical base pairs (counter-examples). Therefore, comparing the output of RNAalifold against that of version 2.4.18 or less will most likely result in differences.

This version also makes Python 3 the default Python for the SWIG interface. The only difference you should see, however, is that the source tree changed from directories interfaces/Python and interfaces/Python3 to interfaces/Python and interfaces/Python2. Additionally, we add PDF versions of our contributors license agreement (CLA) to the source code tar-ball.

The list of all major changes in this version can be found below:

This version fixes two regressions in comparative structure prediction.

In particular, we adapt the computation of base pairing probabilities and Boltzmann sampling to the changes introduced with Version 2.4.4, and restore dot-bracket based hard constrained base pairs that got lost with Version 2.4.2.

This version introduces a correction with respect to the handling of dangling end contributions in exterior loops when computing the partition function for multiple sequence alignments!
Therefore, the ensemble free energies and equilibrium probabilities computed with programs such as RNAalifold and RNAplfold may slightly differ compared to previous versions.

The above correction can briefly be described as a harmonization of the partition function model compared to the MFE implementation, where end-gaps in the alignment can never contribute any stability to helices branching off the exterior loop.

This version adds hard/soft constraints support for our implementations of sliding-window structure prediction as used in the executable programs RNALfold and RNAplfold.

Together with the new feature of hard- and soft-constraints for sliding-window structure prediction we unified and corrected the implementations used in the tools RNALfold, RNALalifold, and RNAplfold.

The major changes are

• All the above tools now use the same definition of maximum base pair span.
  This means that for a maximum base pair span of L the recursions only consider base pairs (i, j) with (j - i + 1) ≤ L.
• Locally optimal structures that are part of another, larger locally optimal structure are filtered out.
  This mainly affects the output of RNALfold that sometimes did not filter properly. Note, however, that filtering is only performed for consecutive hits!. Thus, it may still happen that locally optimal substructures appear in the output.
• Locally optimal structure must not be branched in the exterior loop.
  This rule was not considered in the output of RNALalifold and thus has a large impact on the hits obtained from that tool.

Therefore, comparing the output of the above tools against that of version 2.3.5 or less will most likely result in differences.

This version also adds a new software sub-package, RNAlocmin. Detailed information on this tool can be obtained here. Additionally, we also add a PDF version of the RNA Bioinformatics tutorial as available on our homepage to the default instalation target.

The list of all major changes in this version can be found below:

This version introduces the first transitions towards a more flexible, callback-function based RNA secondary structure decomposition.

The so-called unstructured domains feature extends the RNA folding grammar with alternative handling of self-enclosed unpaired stretches in an RNA structure. This enables one, for instance, to model protein binding to single stranded sequence motifs, while addressing the competition between base pairing and binding affinity of the ligand. The entire extension is not only available using our RNAlib C-library, but also through the Perl and Python scripting interfaces. The program RNAfold already includes a convenience method to add unstructured domain sequence motifs and their corresponding binding free energy through easy-to-use command files.

For further details, we refer to our publication "RNA folding with hard and soft constraints", the section on Unstructured Domains and Domain extensions commands of the RNAlib reference manual, and the manpage of RNAfold.

This version introduces tons of new features. First of all, the new hard- and soft-constraints allow for easy manipulation of the prediction recursions. For instance, SHAPE reactivity data can be directly converted to soft-constraints and readily used in MFE, and partition function computations. Ligand binding to hairpins and interior loops with specific sequence/structure motifs, e.g. accessible through RNAfold, is another feature that became possible due to the new constraints framework.

Most of the new features we've implemented into RNAlib require specific new functions that, partially, replace already existing ones. Therefore, with this version, we start introducing the new v3.0 API, that will eventually replace the one currently existing. All functions of the new API are now prefixed with vrna_ that not only makes them easily recognizable, but also provides some sort of namespacing that helps in the development of third-party programs linking agains RNAlib.
Through usage of a generic data structure for all secondary structure prediction algorithms, most functions that implement a special algorithm, such as MFE, or partition function, are now polymorphic. Thus, there is only one single function in the interface that can deal with multiple variations of input, such as single sequences or sequence alignments.
However, we will maintain complete backward compatibility until v3 of the ViennaRNA Package.

We also started to make the new API accessible through the Perl and Python interface, where we introduce a new object-oriented way to use the RNAlib.

Please see the RNAlib Reference Manual, and the individual manpages of the executable programs in the ViennaRNA Package for further details on new features, and feature changes.

This version fixes a major bug in the energy evaluation of exterior loops and multibranch loops!

In previous versions of the ViennaRNA Package starting with version 2.0, the energy contributions for dangling ends and terminal mismatches of exterior- and multibranch loops were not applied correctly. This problem appeared due to a defective parser that converts the Turner 2004 energy parameters into ViennaRNA code. Therefore, if you compare predictions, e.g. from RNAfold 2.1.9, with those of versions prior to 2.1.9, they may differ. In general, the differences in predicted structures and energies should be very small. However, it may happen that results are totally different.

To assess the impact of the fixed energy parameter usage we re-ran all benchmarks as applied for the original publication of The ViennaRNA Package 2.0. The resulting performance measures suggest, that the impact is actually very small, although predictions seem to be slightly better with version 2.1.9.
See the ViennaRNA Package 2.0 - Performance page for details.

The ViennaRNA Package was recently extended to take a certain class of higher-order structure motifs, so called G-quadruplexes, into account, too.

This is the recursion scheme that we use to predict these structure motifs:

G-quadruplex predictions are marked by the character "+" in the extended dot-parenthesis notation output. Thus, RNAfold output may look like

Here, the four consecutive stretches of "+" mark the G-G interactions within the G-quadruplex. On calculation of base pairing probabilities, the dot-plot output of RNAfold contains the Boltzmann probabilities of a G-quadruplex delimited by sequence position i and j, depicted by a green triangle, as well as the individual G-G interaction probabilities displayed like canonical base pair interactions.
Calculation of the base pair probabilities for the above example then results in a dot-plot like this

References for this new feature: