DrTransformer -- heuristic cotranscriptional folding.

DrTransformer (short for "DNA-to-RNA transformer") is a program for heuristic and deterministic cotranscriptional folding simulations of RNA molecules. The code of this project is available under MIT license, however this software depends on the ViennaRNA package which is available through the ViennaRNA license.

Installation

If you already have the Python bindings of the ViennaRNA package installed, then the latest stable release of DrTransformer can be installed from PyPI:

  ~$ pip install drtransformer

DrTransformer can also be installed with bioconda to resolve the ViennaRNA dependency automatically. First, make sure bioconda is set up properly with:

  ~$ conda config --add channels defaults
  ~$ conda config --add channels bioconda
  ~$ conda config --add channels conda-forge
  ~$ conda config --set channel_priority strict

Second, install or update your DrTransformer installation.

  ~$ conda install drtransformer

Testing/Contributing

To install the latest development version of DrTransformer, clone the repository and run:

  ~$ pip install .[dev]

Use the following command to run all present unittests:

  ~$ python -m pytest 

Please provide unittests if you are submitting a pull request with a new feature.

Usage

Until further documentation is available, please use the --help options of the command line executables:

  ~$ DrTransformer --help
  ~$ DrPlotter --help

An example cotranscriptional folding simulation

We show simulations of three sequences designed by Xayaphoummine et al. (2006). Briefly, two sequences are composed of the same palindromic subsequences (A, B, C, D) in forward and reverse order (ABCD and DCBA); the third sequence (DCMA) has a point mutation which changes B to M. The experiment demonstrates how the order of helix formation determines which structures are formed at the end of transcription, an effect that cannot be observed with a thermodynamic equilibrium prediction, because the free energies of, for example, the helices A:B and B:A are almost the same due to their palindromic subsequences. The three input files ABCD.fa, DCBA.fa and DCMA.fa contain a fasta header and the respective sequence from the original publication. Those files can be found in the subfolder examples/.

  ~$ cat ABCD.fa | DrTransformer --name ABCD --o-prune 0.01 --logfile 

This command line call of DrTransformer produces two files:

• ABCD.log contains a human-readable summary of the cotranscriptional folding process.
• ABCD.drf contains the details of the cotranscriptional folding simulation in the DrForna file format.

Structure-based data analysis

DrPlotter supports different types of visual analysis for the .drf file format. The following command line call reads the previously generated file ABCD.drf and produces a plot called ABCD.png.

  ~$ cat ABCD.drf | DrPlotter --name ABCD --format png

The legend of ABCD.png must be interpreted in combination with the ABCD.log file. Note that the structure IDs from your newly generated files might not match the ones shown here. For example, to see which structures are shown at the simulation of nucleotide 73, read the log file entries for this transcript length:

73    1 .(..(((((((((((((((....)))))))))))))))..).(((((((((.......)))))))))...... -42.60 +[0.0213 -> 0.9876] ID = 24
73    2 ....(((((((((((((((....))))))))))))))).(..(((((((((....)).)))))))..)..... -39.90 -[0.9787 -> 0.0124] ID = 25

The logfile lists two structures (in order of their free energy), it shows their occupancy at the start of the simulation and at the end of a simulation in square brackets, and it provides the ID to follow a specific structure through the transcription process (+/- indicate a change in occpancy). The IDs are used as labels in the plot ABCD.png.

Motif-based data analysis

Instead of following specific structures, it is often more helpful to visualize when specific helical motifs are formed in the ensemble. Generally, we refer to a helix formed from sequences A and B as A:B, etc. All potential helices plotted here are provided in dot-bracket notation in the files ABCD.motifs, DCBA.motifs and DCMA.motifs.

  ~$ cat ABCD.drf | DrPlotter --name ABCD-motifs --molecule ABCD --format png --motiffile ABCD.motifs --motifs A:B C:D A:D B:C
  ~$ cat DCBA.drf | DrPlotter --name DCBA-motifs --molecule DCBA --format png --motiffile DCBA.motifs --motifs B:A D:C D:A C:B
  ~$ cat DCMA.drf | DrPlotter --name DCMA-motifs --molecule DCMA --format png --motiffile DCMA.motifs --motifs M:A D:C D:A C:M

ABCD forms only structures A:B and C:D but not A:D and B:C. Also, helix C:D is not formed "immediately", because there is a competing structure which is cotranscriptionally favored (see ID 25 from the previous anlysis).

DCBA forms structures with all motifs. The helical structures C:B and D:A dominate with more than 90%, the helices D:C and B:A are below 10% of the population. Eventually, D:C and B:A will be dominant, but not on the time scale simulated here. (Can you repeat the analysis to see how much time it needs until D:C and B:A dominate the ensemble?)

As shown in the publication, a single point mutation (from DCBA to DCMA) is sufficient to drastically shift occupancy of helices: M:A and D:C are more occupied at the end of transcription than D:A and C:M.

Tips and tricks

• The header of the logfile contains all relevant DrTransformer parameters that generated the file.
• You can use the parameter --plot-minh to group similar structures (separated by energy barriers < plot-minh) together. In contrast to the --t-fast parameter, this will not affect the accuracy of the model.
• Use --pause-sites to see the effects of pausing at specific nucleotides on cotranscriptional folding.
• Motifs for DrPlotter can also contain 'x' in the dot-bracket notation for must be unpaired.

Version

v0.12 -- perparing for official release

• changed --t-lin, --t-log defaults and fixed --t-lin=1, --t-log=1
• fixed potential issues with --t-end = --t-ext
• adapted README example to publication

v0.11 -- using lonely base-pairs

• removed the --noLP default (added parameter setting)
• added profiling option for runtime optimization
• using --cg-auto default paramter
• using k0=1e5, t-ext=0.04 default parameter
• added new visulization types and fixed motif file input
• added epsilon to t-fast sanity check

v0.10 -- moved to beta status (first official release)

• changes in parameter defaults
• bugfix in linalg
• new DrPlotter simulation layout and motif plotting
• repaired code to enable plotting including pause sites

v0.9 -- standalone package (no official release)

• extraction from the [ribolands] package to a standalone Python package.
• using scipy and numpy for matrix exponentials (instead of [treekin])
• implemented lookahead to skip pruning of potentially relevant future structures

Cite

Stefan Badelt, Ronny Lorenz, Ivo L Hofacker: DrTransformer: heuristic cotranscriptional RNA folding using the nearest neighbor energy model, Bioinformatics, Volume 39, Issue 1, January 2023, https://doi.org/10.1093/bioinformatics/btad034