Julia interface to ViennaRNA for RNA structure prediction and analysis
Author marcom
19 Stars
Updated Last
1 Year Ago
Started In
October 2021

Julia interface to the ViennaRNA library

Unofficial Julia interface to the ViennaRNA library for RNA secondary structure prediction and analysis. Please cite the original ViennaRNA publications if you use this library.


Install ViennaRNA from the Julia package REPL, which can be accessed by pressing ] from the Julia REPL:

add ViennaRNA


using ViennaRNA, Unitful

The Unitful library is needed to be able to specify units with @u_str, e.g. 4u"kcal/mol" or 37u"°C". You can get the degree symbol ° by typing \degree and pressing the TAB key in the REPL or in an editor with Julia syntax support.

The original C API functions can be found in the submodule ViennaRNA.LibRNA. Most functions can be called with a String containing the RNA sequence instead of a FoldCompound, e.g. mfe("GGGAAACCC").


A FoldCompound encapsulates nucleic acid strands and model details, such as energy parameters, temperature, etc.

fc = FoldCompound("GGGGGAAAAACCCCCC";
                  options=[:mfe, :pf],

Important keyword arguments

  • options is a subset of [:default, :eval_only, :hybrid, :mfe, :pf, :window].

  • temperature is used to rescale the free energies with the formula ΔG = ΔH - TΔS (the energy parameter sets contain enthalpy and entropy contributions). The default is 37u"°C"

Model details (additional keyword arguments):

  • circular: determines if the RNA strand is circular, i.e. the 5'-end and 3'-end are covalently bonded. Default is false.
  • dangles: how to treat dangling base pairs in multiloops and the exterior loop. Can be 0, 1, 2, or 3. See ViennaRNA docs for details. Default is 2.
  • gquadruplex: allow G-quadruplexes in predictions. Default is false.
  • log_ML: use logarithmic energy model for multiloops. Default is false.
  • max_bp_span: maximum number of bases over which a basepair can span. Default value is -1 (which means unlimited).
  • min_loop_length: the minimum size of a loop (without the closing base pair). Default is 3.
  • no_GU_basepairs: disallow G-U basepairs. Default is false.
  • no_GU_closure: disallow G-U basepairs as closing pairs for loops. Default is false.
  • no_lonely_pairs: disallow isolated base pairs. Default is false.
  • special_hairpins: use special hairpin energies for certain tri-, tetra- and hexloops. Default is true.
  • uniq_ML: use unique decomposition for multiloops, needed for sample_structures and subopt. Default is false.
  • window_size: window size to be used for local calculations performed in a window moving over the sequence. This value is ignored unless the :window option is set in the FoldCompound options. The default value for window_size is -1.

Changing the energy parameter set

ViennaRNA stores energy parameters in global variables after loading them from a file. Each time a new FoldCompound is created, the parameters are copied from the global variables and saved inside the FoldCompound.

The global variables storing energy parameters can be changed by calling a function specific to each parameter set, or via a Symbol with ViennaRNA.params_load(:RNA_Turner1999). Subsequent calls to FoldCompound will use the new parameters and store a copy of the parameters in the newly created FoldCompound.

The default energy set loaded on startup is :RNA_Turner2004.

ViennaRNA.params_load_defaults()  # default is :RNA_Turner2004

# Options are
#   :DNA_Mathews1999, :DNA_Mathews2004,
#   :RNA_Andronescu2007, :RNA_Langdon2018,
#   :RNA_Turner1999, :RNA_Turner2004

Multiple strands

Mutiple strands can be given by separating them with an &, e.g. FoldCompound("GGGG&CCCC").

Comparative folding with an MSA (alifold)

Pass multiple sequences to the FoldCompound constructor for comparative mode (alifold): FoldCompound(["GG-GAAAACCCC", "GCCGAAA-CGGC"]).

It is currently not possible to have multiple strands in alifold mode.

Minimum free energy structure (MFE)

# please excuse the excess precision printed when displaying -9.4 kcal/mol
mfe(fc)  # => ("(((((.....))))).", -9.399999618530273 kcal mol^-1)

Partition function

partfn(fc)  # => ("(((((.....})))),", -9.81672180213034 kcal mol^-1)

Free energy change of folding into a structure

energy(fc, "((((.......)))).")  # => -6.199999809265137 kcal mol^-1

Basepair probabilities

bpp(fc)  # => 16×16 Matrix{Float64}

Boltzmann probability of a structure

prob_of_structure(fc, "(((((.....))))).")  # => 0.5085737925408758

Ensemble defect

ensemble_defect(fc, "(((((.....))))).")  # => 0.33085374128228884

Sample structures (probabilistic / stochastic backtrack)

Sample from Boltzmann ensemble of secondary structures

sample_structures(fc)                         # => [ "((((......)))).." ]
sample_structures(fc; options=:nonredundant,
                      num_samples=20)         # => 20-element Vector{String}

Suboptimal structures

All suboptimal structures with energies delta above the MFE structure

subopt(fc; delta=4u"kcal/mol")  # => Vector{Tuple{String, Quantity}}

Suboptimal structures with the method of Zuker

subopt_zuker(fc)  # => Vector{Tuple{String, Quantity}}

Sliding window prediction of MFE

mfe_window saves all the results in a Vector.

seq = "G"^50 * "A"^4 * "C"^50
mfe_window(seq; window_size=30)
fc = FoldCompound(seq; options=[:default, :window], window_size=30)
mfe_window(fc)  # => Vector{ResultWindowMFE}

mfe_window_channel returns a Channel that can be used to iteratively process the results.

seq = "G"^50 * "A"^4 * "C"^50
chan = mfe_window_channel(seq; window_size=30)
fc = FoldCompound(seq; options=[:default, :window], window_size=30)
chan = mfe_window_channel(fc)

Neighboring structures

Move set to reach neighboring structures of a given structure

neighbors(fc, Pairtable("((.....))"))  # => Vector{Vector{Tuple{Int,Int}}}

Basepair distance between secondary structures

bp_distance("....", "(())")  # => 2

Tree edit distance between secondary structures

tree_edit_dist("(..)", "....")  # => 4.0f0

Mean basepair distance

Mean basepair distance of all structures to each other, weighted by the structure's Boltzmann probabilities

mean_bp_distance(fc)  # => 5.266430215905888

Centroid structure

Centroid structure of ensemble: structure with smallest sum of base-pair distances weighted by Boltzmann probabilities:

centroid(fc)  # => ("(((((.....))))).", 4.799131457924728)

Maximum expected accuracy (MEA) structure

The gamma parameter trades off specificity (low gamma) and sensitivity (high gamma).

mea(fc; gamma=1.0)  # => ("(((((.....))))).", 10.706348f0)

Heat capacity calculation

# starting temperature, end temperature, temperature increment
heat_capacity(fc, 10u"°C", 60u"°C")  # => Vector{Tuple{Quantity,Quantity}}
heat_capacity(fc, 10u"°C", 60u"°C", 1u"°C"; mpoints=5)

Plotting structures

# plot coordinates of a secondary structure, returns two arrays with
# x and y coordinates
plot_coords("(((...)))")  # => Tuple{Float32[], Float32[]}

See PlotRNA.jl for more secondary structure plotting functionality.

Inverse folding / sequence design

inverse_fold("AAAAAAA", "((...))")     a# => ("GCAAAGC", 2.0f0)
inverse_pf_fold("AAAAAAA", "((...))")  # => ("GCCAAGC", 2.0244526863098145 kcal mol^-1)

Seeding the random number generator


Reducing memory usage

When creating many FoldCompounds, running finalize manually will avoid excessive memory buildup.

for i = 1:100_000
    fc = FoldCompound("ACGU")
    # do something with fc

Related Julia packages

Used By Packages