The RF Fold module is designed to allow transcriptome-wide reconstruction of RNA structures, starting from XML files generated using the RF Norm tool. This tool can process a single, or an entire directory of XML files, and produces the inferred secondary structures (either in dot-bracket notation, or CT format) and their graphical representation (either in Postscript, or SVG format).
Folding inference can be performed using 2 different algorithms:

1. ViennaRNA
2. RNAstructure

Prediction can be performed either on the whole transcript, or through a windowed approach (see next paragraph).

Structure modelling

The structure modelling approach is inspired by the original method described in Siegfried et al., 2014 (PMID: 25028896). Since version 2.8.0, the underlying logic of the windowed approach has been slightly changed, by performing the detection of pseudoknots as the last step. The procedure is outlined below:



In step I, a window is slid along the RNA (if the -w option is enabled, otherwise the entire RNA is used), and partition function is calculated. If provided, soft-constraints from structure probing are applied. Predicted base-pair probabilities are averaged across all windows in which they have appeared, and base-pairs with >99% probability are retained, and hard-constrained to be paired in step III.
In step II, a window is slid along the RNA, and MFE folding is performed, including (where present) soft-constraints from probing data, and base-pairs from step I. Predicted base-pairs are retained if they appear in >50% of analyzed windows.
In step III (optional), a window is slid along the RNA, and putative pseudoknots are detected using the same approach employed by the ShapeKnots algorithm (Hajdin et al., 2013 (PMID: 23503844)). Our implementation of the ShapeKnots algorithm relies on the ViennaRNA package (instead of RNAstructure as in the original implementation), thus it is much faster:



Nonetheless, both algorithms work in single thread. Alternatively, the multi-thread implementation ShapeKnots-smp shipped with the latest RNAstructure version can be used.
If constraints from structure probing experiments are provided, these are incorporated in the form of soft-constraints. Predicted pseudoknotted base-pairs are retained if they apper in >50% of analyzed windows and if they do not clash with the nested base-pairs indentified in step II. In case structure probing constraints are provided, pseudoknots are retained only if the average reactivity of bases on both sides of the pseudoknotted helix is below a certain reactivity cutoff.

Along with the predicted structure, the windowed method also produces a WIGGLE track file containing per-base Shannon entropies.
Regions with higher Shannon entropies are likely to form alternative structures, while those with low Shannon entropies correspond to regions with well-defined RNA structures, or persistent single-strandedness (Siegfried et al., 2014).
Shannon entropy is calculated as:

$$H_i=-\sum _{j=1}^J{p}_{\mathrm{i,j }}{\log }_{\mathrm{10 }}{p}_{\mathrm{i,j}}$$

where p_i,j is the probability of base i of being base-paired to base j, over all its potential J pairing partners.
Since version 2.9.1, RF Fold can use R to generate PDF graphical reports for each structure, reporting reactivity, MEA structure, Shannon entropy, and base-pairing probabilities:

Additionally, since version 2.9.4, RF Fold can use the RNAplot tool of the ViennaRNA package (v2.7.0 or greater) to generate SVG secondary structure plots with overlaid reactivities:

Combining multiple experiments

Since version 2.9.0, RF Fold can combine multiple experiments into a single prediction. These can either be replicates prepared using the same chemical probe, or experiments performed by using different chemical probes.

The combination is achieved by iterating the folding process (as described in the "Structure modelling" paragraph above) over all experiments. Windows from all experiments are combined using a majority voting approach, so that only the pairs passing the above-detailed thresholds (i.e., pairs with >99% probability and pairs appearing in >50% windows) are retained. Therefore, if, for example, a certain base-pair appears in 100% of the windows for one replicate and in just 25% of the windows for another replicate, it will be included in the final model as, in total, it appears in 62.5% of the windows.

In the example below:

$ rf-fold -sl 2.4 -in -0.2 experiment_1/ experiment_2/

the structure of the transcripts common to both experiments will be modelled y combining both sets of reactivities, while transcripts unique to either experiments will be modelled using a single set of reactivities.
Single XML files and entire XML folders can also be combined:

$ rf-fold -sl 2.4 -in -0.2 experiment_1/ experiment_2/ transcript_1.xml

In the above example, transcript_1 will be modelled using data coming from both experiment #1 and #2 (if the file transcript_1.xml is present in the respective folders), as well as data from transcript_1.xml, while the all other transcripts potentially shared across experiments #1 and #2 will only be modelled using those two datasets.

Importantly, it is possible combining even experiments generated using different chemical probes. In the following example:

$ rf-fold -sl 2.4 -in -0.2 DMS_data/ CMCT_data/

two experiments, one generated using DMS and the other using CMCT, are combined into a single prediction that incorporates both sets of reactivities.

While in this case, the same slope and intercept values will be used for both experiments, this might not be ideal, as different reagents might require very different slope/intercept pairs. Therefore, RF Fold can also accept multiple slope/intercept pairs, as a comma-separated list, which must follow the same order as the provided experiments, for example:

$ rf-fold -sl 2.4,4.5 -in -0.2 DMS_data/ CMCT_data/

At this point, a slope of 2.4 will be used for the DMS dataset and a slope of 4.5 will be used for the CMCT datset, while the intercept will be -0.2 for both experiments.

Usage

$ rf-fold [options] XML_dir_1/ XML_dir_2/ .. XML_dir_N/
$ rf-fold [options] rna1.xml rna2.xml .. rnaN.xml

To list all available parameters, simply type:

$ rf-fold -h

Parameter	Type	Description
-o or --output-dir	string	Output directory for writing inferred structures (Default: rf_fold/)
-ow or --overwrite		Overwrites the output directory if already exists
-ct or --connectivity-table		Writes predicted structures in CT format (Default: Dot-bracket notation)
-m or --folding-method	int	Folding method (1-2, Default: 1): 1. ViennaRNA 2. RNAstructure
-p or --processors	int	Number of processors (threads) to use (Default: 1)
-oc or --only-common	int	In case of multiple experiments, only transcripts covered across at least this number of experiments will be folded
-g or --img		Enables the generation of graphical reports (requires R and, optionally, RNAplot v2.7.0 or greater)
-vrp or --vienna-rnaplot	string	Path to ViennaRNA RNAplot v2.7.0 (or greater) executable (Default: assumes RNAplot is in PATH)
-R or --R-path	string	Path to R executable (Default: assumes R is in PATH) Note: also check $RF_RPATH under Environment variables
-t or --temperature	float	Temperature in Celsius degrees (Default: 37.0)
-sl or --slope	float	Sets the slope used with structure probing data restraints (Default: 1.8 [kcal/mol])
-in or --intercept	float	Sets the intercept used with structure probing data restraints (Default: -0.6 [kcal/mol])
-md or --maximum-distance	int	Maximum pairing distance (in nt) between transcript's residues (Default: 0 [no limit])
-nlp or --no-lonelypairs		Disallows lonely base-pairs (1 bp helices) inside predicted structures
-i or --ignore-reactivity		Ignores XML reactivity data when performing folding (MFE unconstrained prediction)
-is or --ignore-sequence		In case of multiple experiments, nucleotide differences (e.g. SNVs) between XML files are ignored
-hc or --hard-constraint		Besides performing soft-constraint folding, allows specifying a reactivity cutoff (specified by -f) for hard-constraining a base to be single-stranded
-c or --constraints	string	Path to a directory containing constraint files (in dot-bracket notation), that will be used to enforce specific base-pairs in the structure models
-f or --cutoff	float	Reactivity cutoff for constraining a position as unpaired (>0, Default: 0.7)
-w or --windowed		Enables windowed folding
-pt or --partition	string	Path to RNAstructure partition executable (Default: assumes partition is in PATH) Note: by default, partition-smp will be used (if available)
-pp or --probabilityplot	string	Path to RNAstructure ProbabilityPlot executable (Default: assumes ProbabilityPlot is in PATH)
-fw or --fold-window	int	Window size (in nt) for performing MFE folding (>=50, Default: 600)
-fo or --fold-offset	int	Offset (in nt) for MFE folding window sliding (Default: 200)
-pw or --partition-window	int	Window size (in nt) for performing partition function (>=50, Default: 600)
-po or --partition-offset	int	Offset (in nt) for partition function window sliding (Default: 200)
-wt or --window-trim	int	Number of bases to trim from both ends of the partition windows to avoid end biases (Default: 100)
-dp or --dotplot		Enables generation of dot-plots of base-pairing probabilities
-sh or --shannon-entropy		Enables generation of a WIGGLE track file with per-base Shannon entropies
-pk or --pseudoknots		Enables detection of pseudoknots (computationally intensive)
-ksl or --pseudoknot-slope	float	Sets slope used for pseudoknots prediction (Default: same as -sl <slope>)
-kin or --pseudoknot-intercept	float	Sets intercept used for pseudoknots prediction (Default: same as -in <intercept>)
-kp1 or --pseudoknot-penality1	float	Pseudoknot penality P1 (Default: 0.35)
-kp2 or --pseudoknot-penality2	float	Pseudoknot penality P2 (Default: 0.65)
-kt or --pseudoknot-tollerance	float	Maximum tollerated deviation of suboptimal structures energy from MFE (>0-1, Default: 0.5 [50%])
-kh or --pseudoknot-helices	int	Number of candidate pseudoknotted helices to evaluate (>0, Default: 100)
-kw or --pseudoknot-window	int	Window size (in nt) for performing pseudoknots detection (>=50, Default: 600)
-ko or --pseudoknot-offset	int	Offset (in nt) for pseudoknots detection window sliding (Default: 200)
-kc or --pseudoknot-cutoff	float	Reactivity cutoff for retaining a pseudoknotted helix (0-1, Default: 0.5)
-km or --pseudoknot-method	int	Algorithm for pseudoknots prediction (1-2, Default: 1): 1. RNA Framework 2. ShapeKnots Note: the chosen folding method (specified by -m) affects the algorithm used by RNA Framework (pseudoknot detection method #1) to define the initial MFE structure
		RNA Framework pseudoknots detection algorithm options
-vrs or --vienna-rnasubopt	string	Path to ViennaRNA RNAsubopt executable (Default: assumes RNAsubopt is in PATH)
-ks or --pseudoknot-suboptimal	int	Number of suboptimal structures to evaluate for pseudoknots prediction (>0, Default: 1000)
-nz or --no-zuker		Disables the inclusion of Zuker suboptimal structures (reduces the sampled folding space)
-zs or --zuker-suboptimal		Number of Zuker suboptimal structures to include (>0, Default: 1000)
		ShapeKnots pseudoknots detection algorithm options
-sk or --shapeknots	string	Path to ShapeKnots executable (Default: assumes ShapeKnots is in PATH) Note: by default, ShapeKnots-smp will be used (if available)
		Folding method #1 options (ViennaRNA)
-vrf or --vienna-rnafold	string	Path to ViennaRNA RNAfold executable (Default: assumes RNAfold is in PATH)
-ngu or --no-closing-gu		Disallows G:U wobbles at the end of helices
		Folding method #2 options (RNAstructure)
-rs or --rnastructure	string	Path to RNAstructure Fold executable (Default: assumes Fold is in PATH) Note: by default, Fold-smp will be used (if available)
-d or --data-path	string	Path to RNAstructure data tables (Default: assumes DATAPATH environment variable is already set)

Constraint files

Constraint files allow forcing base-pairing of certain positions in the RNA. These files are standard dot-bracket files and they must be named after the transcript ID used in the corresponding XML files (for instance, if the XML file is named XYZ.xml, the module will look for a XYZ.db file in the constraint folder):

>XYZ
UUUCGUACGUAGCGAGCGAGUAGCUGAUGCUGAUAGCGGCGAUGCUAGCUGAUCGUAGCGCGCGAUCGAUCGAUGC
..(((.............................................................))).......

In the above example, the constraint file instructs the module to force the base-pairing between positions 3-69, 4-68 and 5-67 of the XYZ transcript.

Output dot-plot files

When option -dp is provided, RF Fold produces a dot-plot file for each transcript being analyzed, with the following structure:

1549                                   # RNA's length
i       j       -log10(Probability)    # Header 
8       254     0.459355416499312
9       253     0.446335563943221
10      252     0.456738523239413
11      251     0.454733421725068
12      250     0.46965667808714
13      249     0.47837140333524
21      35      0.268192200569539
22      34      0.0183400615262171
23      33      0.0166665677814708
24      32      0.0128927546134575
25      31      0.0148601207296645
26      30      0.0252017532628297

-- cut --

1497    1510    0.0147874890078331
1498    1509    0.0102803152157546
1499    1508    0.0137510190884233
1500    1507    0.0402352346970943

where i and j are the positions (1-based) of the bases involved in a given base-pair, followed by the -log₁₀ of their base-pairing probability.
These files can be easily viewed using the Integrative Genomics Viewer (IGV) (for additional details, please refer to the official Broad Institute's IGV page).