RF Compare allows comparing RF Fold-inferred secondary structures, with a reference of known secondary structures, reporting for each comparison 4 metrics: the PPV, the sensitivity, the FMI (Fowlkes-Mallows index) and the mFMI (modified FMI). For additional details, check the Metrics section below.

Reference structures can be provided both in Vienna (dot-bracket), or in CT format. Since version 2.8.8, reference structures can be provided both as a single file containing multiple structures, or as a folder of individual structure files.
The sequence ID of the reference structures must match the compared file's name (e.g. "Transcript#1" expects a file named "Transcript#1.ct" or "Transcript#1.db").
RF Compare can be invoked both on a single structure, or on an entire folder of RF Fold-predicted structure files. Structures can be provided either in CT or Vienna (dot-bracket) format.
RF Compare can further generates vector graphical reports (SVG format) for each structure, reporting the reference structure and the compared structure, with base-pairs colored according to their presence in both structures:

RF Compare plot

Metrics

Given a reference and a predicted structure as input, RF Compare calculates 4 metrics. Each metric ranges between 0 (more dissimilar structures) and 1 (more similar structures):

Positive Predictive Value (PPV)

The fraction of base-pairs present in the predicted structure that are also present in the reference structure

Sensitivity

The fraction of base-pairs present in the reference structure that are also present in the predicted structure

Fowlkes-Mallows index (FMI)

The geometric mean of PPV and sensitivity (introduced by Deigan et al., 2009, PMID:19109441):

FMI = Pboth(Pboth+Pref)(Pboth+Ppred)


where Pboth is the number of base-pairs common to both reference and predicted structure, while Pref and Ppred are the numbers of base-pairs respectively unique to reference and predicted structures.

Modified Fowlkes-Mallows index (mFMI)

A variant of the FMI (introduced by Lan et al., 2022, PMID:35236847), which also rewards bases that are unpaired both in the reference and in the predicted structure:

mFMI = u+(1-u)×FMI


where u is the number of unpaired bases common to both reference and predicted structure.

Usage

To list the required parameters, simply type:

$ rf-compare -h
Parameter Type Description
-p or --processors int Number of processors to use (≥ 1; Default: 1)
-r or --reference string Path to a file containing reference structures in Vienna format (dot-bracket)
-g or --img Enables generation of graphical comparison images
-o or --output-dir string Images output directory (Default: rf_compare/, requires -g)
-ow or --overwrite Overwrites output directory (if the specified path already exists)
-x or --relaxed Uses relaxed criteria (described in Deigan et al., 2009) to calculate PPV and sensitivity
-kp or --keep-pseudoknots Keeps pseudoknotted basepairs in reference structure
-kl or --keep-lonelypairs Keeps isolated base-pairs (helices of length 1 bp) in reference structure
-i or --ignore-sequence Ignores sequence differences (e.g. SNVs) between the compared structures

Note

When parameter --relaxed is specified, a basepair i-j is considered to be present in the reference structure if any of the following pairs exist: i/j; i-1/j; i+1/j; i/j-1; i/j+1. For additional details, please refer to Deigan et al., 2009 (PMID: 19109441)