Structure of this document:
First parameters and their meaning is described.
Second, prediction methods are listed. Each prediction method is described (what and how the structures are predicted). For each method, there is section called "parameters", which serves as an example for parameter definition in json format. Parameters are specified for each method independently on other prediction methods.
The pipeline offer a way to interact with most parts of tools used for secondary structure prediction.
These tools are used for specific tasks and the parameters controlling such behaviour are always set and are not exposed for user control. (io flags, mode flags, etc.)
is string with commandline arguments for RNAfold
(see RNAFold documentation).
It must be specified with double quotes.
Default: No parameters specified.
is string with commandline arguments for RNAalifold
(see RNAalifold documentation).
It must be specified with double quotes.
Default: No parameters specified.
is string with commandline arguments for cmscan
(see cmscan documentation).
It must be specified with double quotes.
Default: No parameters specified.
is string with commandline arguments for cmalign
(see cmalign documentation).
It must be specified with double quotes.
Default: No parameters specified.
specifies parameters for hybrid-ss-min program where P, W, and X are integers
see UNAFold documenation.
It must be specified without double quotes.
The hybrid-ss-min is additionally used with --suffix=DAT --NA=RNA --noisolate parameters.
is string of commandline arguments for clustalo
(see clustal omega documentation)
Default: No parameters specified.
is string of commandline argument for clustalo. Allows specification
of different clustalo parameters for the profile alignment stage.
(see clustal omega documentation)
Default: clustalo called with --profile1 only profile align compatible option can be specified.
is string of commandline argument for centroid_homfold
(see centroid rna package documentation).
By default -g -1 is used. If -g or -t is specified by user then the -g -1 is not used.
is a string argument for muscle aligner. For more details, see muscle
documentation.
By default -seqtype rna -quiet -clwstrict are used and cannot be changed.
Selection of estimated full-length sequences to provide reference for secondary structure prediction (reference sequences)
Parameters used for selection of sequences similar to the query sequence. The selection parameters influence how much each sequence can differ from covariance model, how much the sequences can be similar to each other and what is maximum accepted length difference between the estimated full-length sequence and the query sequence.
Selection of estimated full-length sequences: All estimated full-length sequences are filtered based on cm bit-score, then the remaining sequences are filtered according to their length to be within specified length difference from query sequence. After that, the remaining sequences are filtered based on sequence similarity with each other (from pairwise sequence identity), so only those sequences with similarity less then defined similarity threshold are retained.
Defines inclusion threshold in percent of bit-score value obtained by
aligning query sequence to covariance model.
It serves as filter for not sufficiently related sequences.
The inclusion threshold is computed by CMSCORE_PERCENT * QUERY_BITSCORE / 100,
only estimated full-length sequences with CM alignment bit-score higher then inclusion
threshold are considered for further computation.
CMSCORE_PERCENT ranges from 0 to 100, allowed values are integers.
The higher the threshold, the more conservative setting for trusted sequences (i.e. more similarity to cm is required).
Defines a maximum length difference between the query sequence and the estimated full-length sequence. This serves complementary to cmscore_percent to allow setting low cmscore_percent, while preventing very short or very long estimated full-length sequences (for any reason) to be part of selected sequences set.
MAX_LEN_DIFF_to_QUERY ranges from 0 to 1, allowed values are floating point (with decimal dot).
The higher the threshold, the more difference is allowed.
Defines exclusion threshold in percent of sequence similarity.
This serves as a protection from populating set of selected sequences with
too many too similar sequences (as it may skew alignment and other prediction methods).
PRED_SIM_THRESHOLD ranges from 0 to 100, allowed values are integers.
The query is included as first sequence and all sequences which are similar to it down to defined threshold are excluded. Then next sequence from the remaining sequences is analyzed in same manner.
The higher the similarity the more similar sequences are accepted.
(i.e. PRED_SIM_THRESHOLD = 100 means that only exact duplicate sequences are removed)
Serves for prediction methods where the conserved unpaired positions in alignment are taken as constraints.
Defines how much MSA column must be conserved at to denote the position
as single-strand constraint for RNAfold.
This parameter is always used together with conseq_conserved.
Serves for prediction method where the conserved unpaired positions in alignment are taken as constraints.
Defines how many bases in a row must be conserved to denote the position
as single-strand constraint for RNAfold.
This parameter is always used together with repred_unpaired_tr.
Used with TurboFold to set required (and maximum) number of sequences in prediction. Allowed values are integers >= 2. Beware of setting this too high. Then the prediction is very memory and time expensive.
Predict secondary structure for each estimated full-length sequence with RNAFold.
parameters:
{"rnafold" : {
"RNAfold": "RNAFOLD PARAMETERS"
}
}
Obtain the related covariance model (either find highest scoring one in Rfam by cmscan
or use the one provided with --cm_file), align estimated full-length sequences to it (cmalign),
then take conserved bp as constrains for for folding with RNAfold -C.
parameters:
{"rfam-Rc": {
"RNAfold": "RNAFOLD PARAMETERS",
"cmscan": "CMSCAN PARAMETERS",
"cmalign": "CMALIGN PARAMETERS"
}
}
Non optional parameters:
RNAFold:-C(constraints)cmscan:-g(global aligment of models to query sequence)cmalign:--notrunc(disable truncated hits detection - speeds up computation).
Obtain the related CM (either find highest scoring one in Rfam by cmscan
or use the one provided with --cm_file), extract reference secondary structure,
run hybrid-ss-min (UNAFold) for suboptimal structures and select the most similar
one to the one from CM.
parameters:
{"rfam-sub" : {
"mfold": [P, W, X]
}
}
Predict structure of query with RNAFold and take it as reference structure,
then predict suboptimal structures with hybrid-ss-min (UNAFold) and select
structure most similar to the predicted structure of query (by RNAdistance score).
parameters:
{"fq-sub" : {
"RNAfold": "RNAFOLD PARAMETERS",
"mfold": [P, W, X]
}
}
clustalo - RNAalifold - hybrid-ss-min
Select reference estimated full-length sequences, align them with clustalo (Clustal Omega),
predict consensus structure with RNAalifold.
For each estimated full-length sequence compute suboptimal structures with hybrid-ss-min
and select structure from predicted suboptimal structures the one most
similar to the consensus structure (by RNAdistance score).
parameters:
{"C-A-sub" : {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"clustalo": "CLUSTALO PARAMETERS",
"alifold": "ALIFOLD PARAMETERS",
"mfold": [P, W, X]
}
}
clustalo - RNAalifold - hybrid-ss-min
Select reference estimated full-length sequences, align them with muscle,
predict consensus structure with RNAalifold.
For each estimated full-length sequence compute suboptimal structures with hybrid-ss-min
and select structure from predicted suboptimal structures the one most
similar to the consensus structure (by RNAdistance score).
parameters:
{"M-A-sub" : {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"muscle": "MUSCLE PARAMETERS",
"alifold": "ALIFOLD PARAMETERS",
"mfold": [P, W, X]
}
}
clustalo - RNAalifold - refold.pl - rnafold-C
Select reference estimated full-length sequences, compute alignment with clustalo (MSA1)
and then compute consensus secondary structure with RNAalifold from the MSA1.
Then compute profile alignment of MSA1 with all estimated full-length sequences
and add the consensus structure to all estimated full-length sequences from profile alignment,
finally run refold.pl and RNAFold for the result.
parameters:
{"C-A-r-Rc": {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"clustalo": "CLUSTALO PARAMETERS",
"alifold": "ALIFOLD PARAMETERS",
"RNAfold": "RNAFOLD PARAMETERS"
}
}
clustalo - RNAalifold - unpaired conserved - refold.pl - rnafold-C
Select reference estimated full-length sequences, compute alignment with clustalo (MSA1)
and then compute consensus secondary structure with RNAalifold from the MSA1.
Then compute profile alignment of MSA1 with all estimated full-length sequences,
and add the consensus structure to all estimated full-length sequences from profile alignment.
Then select conserved parts of alignment where the consensus secondary
structure annotation is single-strand (i.e. no base-pairs) and use them as constrains for RNAFold -c.
parameters:
{"C-A-U-r-Rc": {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"repred_unpaired_tr": REPRED_UNPAIRED_TR,
"conseq_conserved": CONSEQ_CONSERVED,
"clustalo": "CLUSTALO PARAMETERS",
"alifold": "ALIFOLD PARAMETERS",
"RNAfold": "RNAFOLD PARAMETERS"
}
}
muscle - RNAalifold - refold.pl - rnafold -C
Select reference estimated full-length sequences, compute alignment with muscle (MSA1)
and then compute consensus secondary structure with RNAalifold from the MSA1.
Then compute profile alignment of MSA1 with all estimated full-length sequences
and add the consensus structure to all estimated full-length sequences from profile alignment,
finally run refold.pl and RNAFold for the result.
parameters:
{"M-A-r-Rc": {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"muscle": "MUSCLE PARAMETERS",
"alifold": "ALIFOLD PARAMETERS",
"RNAfold": "RNAFOLD PARAMETERS"
}
}
muscle - RNAalifold - unpaired conserved - refold.pl - rnafold-C
Select reference estimated full-length sequences, compute alignment with muscle (MSA1)
and then compute consensus secondary structure with RNAalifold from the MSA1.
Then compute profile alignment of MSA1 with all estimated full-length sequences,
and add the consensus structure to all estimated full-length sequences from profile alignment.
Then select conserved parts of alignment where the consensus secondary
structure annotation is single-strand (i.e. no base-pairs) and use them as constrains for RNAFold -c.
parameters:
{"M-A-U-r-Rc": {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"repred_unpaired_tr": REPRED_UNPAIRED_TR,
"conseq_conserved": CONSEQ_CONSERVED,
"muscle": "MUSCLE PARAMETERS",
"alifold": "ALIFOLD PARAMETERS",
"RNAfold": "RNAFOLD PARAMETERS"
}
}
Select reference estimated full-length sequences and pass them to centroid_homfold as
homologous sequences, predict secondary structure for all estimated full-length sequences.
The centroid_homfold is by default called with -g -1 parameter.
With this setting the centroid_homfold predicts multiple structures
(see cetroid_homfold documentation),
the structure with best score is chosen.
parameters:
{"centroid": {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"centroid_homfold": "CENTROID_HOMFOLD PARAMETERS"
}
}
With centroid-fast the reference estimated full-length sequences are selected as follows:
we take non-redundant estimated full-length sequences without ambiguous
bases with cm_score > 0 within allowed length-to-query-length difference.
Then we take up to N reference sequences as homologous sequences for centroid_homfold.
The sequences are added in order of the original BLAST HSPs, starting with query.
For this method there is also parameter preset avalible which sets N to 1
meaning that only query will be used as homologous sequence in prediction.
parameters:
{"centroid-fast": {
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"centroid_homfold": "CENTROID_HOMFOLD PARAMETERS",
"max_seqs_in_prediction": MAX_SEQS_IN_PREDICTION,
}
}
Use covariance model (CM) to generate set of reference sequences for centroid_homfold.
By default the highest scoring CM in Rfam is found by cmscan
(or provided one is used --cm_file option).
Then requested number of sequences (N_SEQS) is generated from the covariance model with cmemit.
The sequences are used as homologous sequences for centroid_homfold.
The generated sequences can differ between runs. If repeatable behaviour
is desired the cmemit can be seeded (see it's options).
parameters:
{"rfam-centroid": {
"n_seqs": N_SEQS,
"cmscan": "CMSCAN PARAMETERS",
"cmemit": "CMEMIT PARAMETERS",
}
}
With Turbo-fast the reference estimated full-length sequences are selected as follows:
we take non-redundant estimated full-length sequences without ambiguous
bases with cm_score > 0 within allowed length-to-query-length difference.
For each estimated full-length sequence, we make non-redundant group of sequences
consisting of the estimated full-length sequence for which we want to predict secondary structure
and up to N-1 reference sequences.
The sequences are added in order of the original BLAST HSPs, starting with query.
That means that if N is 2 and query does not contain ambiguous bases,
then each estimated full-length sequence secondary structure is computed with query sequence as a reference.
This setting is also available as commandline argument with --turbo_fast_preset flag and
will override the "max_seqs_in_prediction" value from prediction_parameters file.
The N can be defined in parameters as "max_seqs_in_prediction": MAX_SEQS_IN_PREDICTION.
parameters:
{"Turbo-fast": {
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"max_seqs_in_prediction": MAX_SEQS_IN_PREDICTION,
"TurboFold": "TurboFold PARAMETERS"
}
}
Select reference estimated full-length sequences without ambiguous bases. Then continue as with Turbo-fast.
parameters:
{"TurboFold": {
"cmscore_percent" : CMSCORE_PERCENT,
"pred_sim_threshold": PRED_SIM_THRESHOLD,
"query_max_len_diff": MAX_LEN_DIFF_to_QUERY,
"max_seqs_in_prediction": MAX_SEQS_IN_PREDICTION,
"TurboFold": "TurboFold PARAMETERS"
}
}