Software::GenoScan - Software for pre-miRNA discovery in genomic sequences
use Software::GenoScan qw(runGenoScan);
This module is the implementation of GenoScan, a computational method for prediction of pre-miRNAs in genomic DNA. MicroRNAs (miRNAs) are small non-protein-coding RNAs that act as post-transciprional regulators of gene expression. They are initially transcribed from intergenic or intronic DNA as primary miRNAs with a characteristic hairpin secondary structure, which are processed into shorter precursor miRNAs (pre-miRNAs). The precursors are exported to the cytoplasm and cleaved to liberate the mature miRNA, which is incorporated into the RNA-induced silencing complex (RISC). The RISC molecule then binds to specific target mRNAs by complementary base-pairing between the mature miRNA and the 3' UTR of the mRNA. This either prevents mRNA translation to protein or mediates mRNA degradation.
GenoScan takes genomic DNA as input, and optional sequence annotation used for filtering. Sequences that pass the filters are segmentized into 200 nt windows with 20 nt overlaps and folded with RNAfold. Hairpins are then extracted from the folded sequences based on a a number of sequence and structure criteria. Finally, the extracted hairpins are classified as miRNAs or non-miRNAs using a logistic regression model trained on human datasets. See the publication for a detailed description of the algorithm and its evaluation. The publication reference can be found in "SEE ALSO".
Following installation, GenoScan can be run from the terminal using the genoscan.pl script supplied in the t/ folder. The command should be in this form
perl genoscan.pl [Options]
GenoScan can be run in four different modes. Depending on the mode, different command line arguments must be given. The next section gives general options that apply to several or all modes, while the following sections give options specific to each mode.
-m value
Specifies the GenoScan running mode. Possible values are genome, classify, benchmark and retrain. The default mode is genome, which is used to scan a whole genome or part thereof for pre-miRNAs. GenoScan can also be used to classify a set of hairpins as miRNAs or non-miRNAs, without first extracting hairpins from a genome. This is done in classify mode. To facilitate comparison with other methods, the benchmark mode enables leave-one-out cross valdiation on datasets containing positive, negative and genomic hairpins. This generates a table with performance measures similar to those in the publication. Finally, the retrain mode is used to retrain the regression model on custom datasets.
genome
classify
benchmark
retrain
-t value
The probability threshold used for hairpin classification. When the regression model evaluates the hairpins, it calculates a probability that the hairpin is a miRNA. By setting a threshold for this probablity, it is possible to adjust the stringency of classification. The value should be between 0 and 1 (default: 0.5). Only hairpins with a probability greater or equal to the threshold will qualify as miRNA candidates by GenoScan. According to the evaluation in the publication, performance peaks around probability treshold 0.5-0.6. Higher stringency will give higher specificity, but lower sensitivity.
-r value
A file containing regression model coefficients to use instead of the default ones. The value should be the file name including path. The format of the file is as follows:
(Intercept) 10.5280777518279 stem_len 0.0500355477383941 loop_size 0.320050551436161 mfe_nor -6.61410292089138 prop_gap -17.1624469004961 prop_bulge 4.80460097209541 prop_wobble -3.18891064122477 gc_content -8.89487058546057 PPP -4.61085343984701 UUU -9.76065756139253 UUP -26.0725578749436 PPU -27.0092130076774 PUP -11.7589439172359 UPP 23.1858571677642
The coefficient name appears at the beginning of each row, followed by a tab and the value.
-o value
The output directory to which the results of GenoScan will be written. This option is mandatory.
-s value
Species abbreviation, which determines the parameters used for extraction and classification. The initial version of GenoScan has only been trained on human, so the only possible value is hsa. Training on multiple species is planned as a future development.
hsa
-l value
In addition to filter predictions by probability, GenoScan also supports filtering by low sequence complexity. A sequence with low complexity has long streches of the same nucleotide, or repeats of a nucleotide segment. For this to work, low-complexity regions need to be masked by lower case letters. Two possibilities are to downloaded masked genomic sequences from NCBI RefSeq database, or mask custom sequences with RepeatMasker. To filter GenoScan predictions by low complexity, a value between 0 and 1 should be specified to the -l option, indicating the percentage of nucleotides that are allowed to be masked. Only sequences with a percentage less than or equal to this value will pass (1 mean 100 %). The default is 1, which allows the entire hairpin sequence to be masked.
-l
-v
This option specifies if GenoScan should be verbose, i.e. continuously report on its progress during a run.
-h
This option causes argument documentation to be printed to the terminal.
-d value
Directory where genomic sequences in FASTA format are located, in which to search for miRNAs. The value should be the path to the folder, including the folder name.
-e value
Directory where genomic annotation files in GBS format are located. This argument is optional and will tell GenoScan to ignore annotated regions (except miRNAs) when searching for miRNAs.
-i value
Name of a file containing regions to include when searching for miRNAs. Regions outside of those specified will not be searched. The format of the file should be as follows:
chr1:1000000-2000000 chr3:1500000-1700000 chr9:500000-1500000
-f value
A file containing hairpin extraction parameters to use instead of the default ones. The value should be the file name including path. The format of the file is as follows:
SEQ_LEN_MIN = 40 CG_COMP_MIN = 0.2 CG_COMP_MAX = 0.8 TR_SIG_MAX = 5 STEM_LEN_MIN = 20 LOOP_LEN_MAX = 20 STEM_BP_PR_MIN = 0.5 UP_STR_NOR_MAX = 0.35 UP_LEN_NOR_MAX = 0.35
-j value
Indicates that GenoScan should start processing at a given step. The value should be between 2 and 5 and allows the run to start working on existing data, rather than starting from the beginning. The steps of genoscan include (1) read annotation filters, (2) seqmentize and fold input sequences, (3) extract hairpins, (4) classify harirpins and (5) write miRNA candidates to output file.
-c value
File in FASTA format containing hairpins to classify as miRNAs or non-miRNAs. The classify mode allows hairpins to be classified directly, without first extracting them from genomic sequences.
-p value
File containing the positive dataset, which consists of known miRNAs in FASTA format.
-n value
File containing the negative dataset, which consists of pseudo-hairpins in FASTA format.
-g value
File containing the genomic dataset, which consists of extracted genomic hairpins in FASTA format. This dataset is only used during benchmarking.
-a value
Custum path to R script used for benchmarking or retraining. The scripts are part of the GenoScan distribution and are located in the scripts/ directory. If GenoScan is run from the t/ directory, the -a option need not be set.
-a
The output generated by GenoScan depends on the mode specified. For genome and classify, GenoScan generates two final output files (in step 5): a list of pre-miRNA candidates in FASTA format and a log file with a summary of the run. The header of each miRNA candiadate starts with the hairpin ID, followed by species abbreviation and the loci in the genome where the hairpin is located. The hairpins appearing in this file are the final list of candidates that pass the probability threshold and low-complexity filtering.
In benchmark mode, the list of candidates is replaced by a table indicating the performance of the regression model for different values of the probability threshold. The performance measures included are sensitivity (SN, ratio of positive hairpins classified as miRNAs), specificity (SP, ratio of negative hairpins classified as non-miRNAs), Matthews correlation coefficient (MCC), false discovery rate (FDR, ratio of miRNA predictions that are negative hairpins) and genome prediction rate (GPR, fraction of hairpins in a genome classified as miRNAs). Output from this mode is written to the benchmark folder instead of step_5.
For retrain mode, a new regression model file is created, containing the coefficients of the model fit to the training datasets. This model can be used to classify hairpins by supplying the file to the -r option. Output from this mode is written to the retrain folder instead of step_5.
-r
To test GenoScan following installation, navigate to the t/ folder and type
perl genoscan.pl -v -d fake_genome -o output
This will run GenoScan in genome mode on a fake genome sequence, composed of random sequences concatenated with sequences from the positive and negative datasets. The result files are written to the output folder and organized into five sub-folders. The first folder (step_1) contains inclusion and exclusion filters, based on the sequence annotation supplied (if any). The second folder contains sequence windows derived from the genomic sequence. These are located in the chunk folder and are organized into chunks, with 50 000 windows per file (maximum). The windows are folded with RNAfold and the folds are placed in the folded directory. In step 3, hairpins are extracted from the folded windows and written to step_3 in the output folder. Extracted hairpins are then refolded, visualized in ASCII format and annotated based on their secondary structure. The annotated hairpins are used by the logistic regression model to classify the hairpins, and the result from the model is written to the log_reg_model_report file in the step_4 folder. Finally, hairpins that pass the probability and low-complexity thresholds qualify as miRNA candidates and are written to the miRNA_candidates.fasta file in step_5. In addition, a summary of the Genoscan run is written to the genoscan_log file.
Ulfenborg B, Klinga-Levan K, Olsson B (2014) GenoScan: Genomic Scanner for Putative miRNA Precursors. Bioinformatics Research and Applications, LNBI 8492, pp. 266-277.
The publication can be found by this link: http://link.springer.com/chapter/10.1007%2F978-3-319-08171-7_24
None by default. The runGenoScan subroutine is exported on request.
Benjamin Ulfenborg, <wolftower85@gmail.com>
Copyright (C) 2014 by Benjamin Ulfenborg
This module is free to use, modify and redistribute at will.
To install Software::GenoScan, copy and paste the appropriate command in to your terminal.
cpanm
cpanm Software::GenoScan
CPAN shell
perl -MCPAN -e shell install Software::GenoScan
For more information on module installation, please visit the detailed CPAN module installation guide.