Unofficial Julia interface to the LinearFold suite of programs for RNA secondary structure prediction. Please cite the applicable original LinearFold, LinearPartition, LinearSampling, or LinearTurboFold publications if you use this library.
The LinearFold, LinearPartition, LinearSampling, and LinearTurboFold programs are supported at the moment. This library calls the binary executables of these programs directly and parses their output.
The name LinearFold derives from the O(n)
running time (where n
is
the sequence length) of calculating an approximate solution, compared
to the typical cubic O(n^3)
running time for the exact solution.
This speedup is achieved by recasting the normal dynamic programming
algorithms to work on the sequence lefttoright and by using a beam
search approximation. Most algorithms also have a linear or quadratic
time dependence on the beam size used. Please refer to the LinearFold
publications for further details.
] add LinearFold
using LinearFold, Unitful

model=:vienna
: energy model to be used. Valid options are:vienna
and:contrafold
. Default is:vienna
. 
beamsize=100
: size used for beam search approximation. Larger numbers trade longer computation time for more precise answers. Default is100
. 
constraints
: structural constraints of the predicted structure. A string consisting of the characters '?', '.', '(', ')', corresponding to positions that have unspecified base pairing, unpaired, or basepairing specified by matching parentheses. 
is_sharpturn=false
: enable sharp turns in predictions. Default isfalse
. 
verbose=false
: output extra information from the program runs to stdout. Default isfalse
.
Uses the LinearFold program to predict the MFE and MFE structure.
# mfe(seq; model, beamsize, constraints, is_sharpturn, verbose)
mfe("GGGAAACCC") # => (1.2 kcal mol^1, "(((...)))")
mfe("GGGAAACCC"; constraints="?(.????)?") # => (0.9 kcal mol^1, "((.....))")
mfe("GGGAAACCC"; model=:contrafold) # => (0.09 kcal mol^1, ".........")
Uses the LinearPartition program to calculate base pair probabilities.
# bpp(seq; model, beamsize, bpp_cutoff, is_sharpturn, verbose)
bpp("GGGAAACCC") # => (1.62 kcal mol^1, sparse(...))
bpp("GGGAAACCC"; bpp_cutoff=0.1)
Uses the LinearPartition program to predict possibly pseudoknotted secondary structures with a beam search approximation to the ThreshKnot algorithm.
Because the predicted structures can contain pseudoknots, the structure is returned as a list of integers which indicate the basepairing partner of the current index.
# threshknot(seq; model, beamsize, threshold, is_sharpturn, verbose)
threshknot("GGGAAACCC") # => (1.62 kcal mol^1, [9, 8, 7, 0, 0, 0, 3, 2, 1])
threshknot("GGGAAACCC"; threshold=0.2)
Uses the LinearSampling program to return num_samples
secondary
structures sampled according to their Boltzmann probability for an RNA
sequence.
# sample_structures(seq; beamsize, num_samples, is_nonsaving, is_sharpturn, verbose)
sample_structures("GGGAAACCC") # => [ "((....)).", ... ]
sample_structures("GGGAAACCC"; num_samples=100)
Uses the LinearTurboFold program to simultaneously align and fold multiple RNA sequences.
# turbofold(sequences; beamsize_hmm, beamsize_cky, iterations,
# threshknot_min_helix_len,
# threshknot_iterations, threshknot_threshold,
# verbose)
turbofold(["GGGAAACCC", "GGCCAAAUGGCCA"])
Uses the LinearPartition program.
# mea(seq; model, beamsize, gamma, is_sharpturn, verbose)
mea("GGGAAACCC") # => (1.62 kcal mol^1, "(((...)))")
mea("GGGAAACCC"; gamma=0.5) # => (1.62 kcal mol^1, ".(.....).")
Uses the LinearFold program.
# zuker_subopt(seq; model, beamsize, delta, is_sharpturn, verbose)
zuker_subopt("GCGCGAAAAAACCCCCCC") # => [ (2.9 kcal mol^1, "....(........)...."), ... ]
zuker_subopt("GCGCGAAAAAACCCCCCC"; delta=4.0u"kcal/mol")
Uses the eval mode of the LinearFold program.
# energy(seq, structure; model, is_sharpturn, verbose)
energy("GGGAAACCC", "(((...)))") # => 1.2 kcal mol^1
Uses the LinearPartition program.
# partfn(seq; model, beamsize, is_sharpturn, verbose)
partfn("GGGAAACCC") # => 1.62 kcal mol^1