https://orcid.org/0009-0004-7522-9878
Figures
Abstract
MicroRNAs (miRNAs) are small non-protein-coding RNAs that regulate gene expression in many eukaryotes. Next-generation sequencing of small RNAs (small RNA-seq) is central to the discovery and annotation of miRNAs. Newly annotated miRNAs and their longer precursors encoded by MIRNA loci are typically submitted to databases such as the miRBase microRNA registry following the publication of a peer-reviewed study. However, genome-wide scans using small RNA-seq data often yield high rates of false-positive MIRNA annotations, highlighting the need for more robust validation methods. miRScore was developed as an independent and efficient tool for evaluating new MIRNA annotations using sRNA-seq data. miRScore combines structural and expression-based analyses to provide rapid and reliable validation of new MIRNA annotations. By providing users with detailed metrics and visualization, miRScore enhances the ability to assess confidence in MIRNA annotations. miRScore has the potential to advance the overall quality of MIRNA annotations by improving accuracy of new submissions to miRNA databases and serving as a resource for re-evaluating existing annotations.
Author summary
MicroRNAs (miRNAs) play a major role in gene regulation in most eukaryotic organisms. Genome-wide analysis of miRNAs and miRNA-encoding precursors (here, MIRNAs), can lead to numerous false positive annotations. Criteria for MIRNA annotation often use a combination of short RNA sequencing and predicted RNA secondary structural properties of precursors. However, implementation of these criteria varies. Here, we introduce a tool, miRScore, developed to standardize MIRNA validation using accepted annotation criteria. miRScore is intended to improve the quality of MIRNA annotation. miRScore takes as input one or more putative miRNA sequences, one or more corresponding precursor RNAs, and short RNA sequencing data. miRScore quickly and accurately evaluates each candidate in the context of the provided data and determines whether each annotation meets all criteria. These results can be used to determine high-confidence miRNAs for cataloging and downstream analysis.
Citation: Vanek A, Griffiths-Jones S, Meyers BC, Shahid S, Axtell MJ (2025) miRScore: A rapid and precise microRNA validation tool. PLoS Comput Biol 21(11): e1013663. https://doi.org/10.1371/journal.pcbi.1013663
Editor: Mark Ziemann, Burnet Institute, AUSTRALIA
Received: December 16, 2024; Accepted: October 25, 2025; Published: November 3, 2025
Copyright: © 2025 Vanek et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The miRScore software is an open-source python project available under the permissive MIT license. Documentation, source code, and test examples are freely available on the github page at https://github.com/Aez35/miRScore. Newly generated small RNA-seq data are available from NCBI GEO under accession GSE282265.
Funding: This work was supported by the National Science Foundation (2130884 to MJA and BCM; 2450802 to BCM), the Biotechnology and Biological Sciences Research Council (BB/W018438/1 to SGJ), and a seed grant from The Huck Institutes of the Life Sciences at Penn State (un-numbered award to MJA). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
MicroRNAs (miRNAs) are a class of small, non-coding RNAs that regulate gene expression within eukaryotes. This regulation typically occurs when a miRNA, which is loaded into an RNA-induced silencing complex (RISC), imperfectly base pairs to a target messenger RNA (mRNA). The RISC then frequently acts as an endonuclease to cleave the mRNA or to otherwise inhibit its translation [1–4]. miRNA-directed regulation of mRNAs is crucial in various biological processes such as developmental timing [5–7], metabolism [8,9], and defensive pathways [10–13] in both plants and animals. Although miRNA biogenesis varies somewhat between animals and plants, the fundamental aspects of miRNA structure and function are conserved [14]. In both plants and animals, the precursors of miRNAs are generally transcribed by RNA polymerase II from an endogenous MIRNA gene. While many MIRNA primary transcripts are transcribed as independent genes from intergenic regions, some are processed from the introns of protein-coding mRNAs [4]. Transcription results in a long single-stranded RNA containing a hairpin, called the primary miRNA (pri-miRNA). The hairpin embedded within the primary transcript is then processed by sequential endonuclease activity (Drosha and Dicer in animals, or by a single Dicer-Like protein in plants) to release a miRNA duplex. The miRNA duplex is a double-stranded RNA, typically with a few mismatched and/or bulged nucleotides, which consists of the mature functional strand (miRNA) and passenger strand (miRNA*). The miRNA duplex is unwound, and a single-stranded mature miRNA is bound to an Argonaute protein to form the RISC. Most frequently a single strand from this duplex is incorporated into RISC and regulates mRNAs; in some cases both strands from the miRNA duplex become separately bound to different RISCs and have two distinct constellations of mRNA targets. For details of microRNA biogenesis, see [4,15,16].
Alignment of deep small RNA-sequencing (sRNA-seq) data to a reference genome is a common method for MIRNA annotation and quantification. Several tools such as ShortStack [17,18], miRador [19], miRDeep [20], and miRDeep-P2 [21], have been developed to annotate miRNAs and other small RNAs using sRNA-seq data. These tools typically work by aligning sRNA-seq data to a reference genome, followed by evaluation of potential miRNA-encoding loci (MIRNA). One way candidate MIRNAs are identified is by the distinctive alignment pattern of the miRNA/miRNA* duplex reads to the hairpin precursor. miRNA and miRNA* reads from sRNA-seq align to a single genomic strand, as their precursors are single-stranded transcripts. These reads align a short distance from each other, forming two distinct “stacks” of read coverage [17,22]. MIRNA primary transcripts are typically short-lived and hard to detect using sRNA-seq or regular mRNA-seq. Most sRNA-seq centered MIRNA identification tools thus annotate “hairpin” sequences that encompass the stem-loop region and some adjacent sequence of pre-determined length. The start and stop positions of these annotations do not necessarily correspond to the ends of the actual primary transcripts. The secondary structure of this putative hairpin precursor is then predicted. For true MIRNAs, the predicted secondary structure of the putative precursor RNA is an imperfect stem-loop. Furthermore, two stacks of aligned sRNA-seq reads from the miRNA and the miRNA* are found on opposite arms of the predicted stem-loop with a diagnostic two nucleotide 3’-overhang. Generally, the sequence with the most abundant set of reads is termed the ‘mature’ miRNA, while the sequence with less abundant reads is the ‘star’ sequence. Detection of reads from both arms of the miRNA duplex is required to confirm the predicted duplex [23–25]. Identification of candidate MIRNAs using sRNA-seq is therefore dependent on empirical evaluation of read alignment patterns in the context of the presumed precursor’s predicted RNA secondary structure.
The identification of MIRNAs through deep sequencing data poses some challenges. One is the handling of multimapping reads, in which there are multiple best-scoring alignments for a single read. This occurs frequently with sRNA-seq data due to shorter read lengths and the fact that identical miRNAs can be encoded by paralogous loci [18]. Another challenge is distinguishing true MIRNAs from other sRNA classes such as short-interfering RNAs (siRNAs), which have their own unique alignment patterns and criteria [23,26]. Each MIRNA discovery tool employs distinct methods for handling these challenges, with varying degrees of performance for identification of novel MIRNAs in plants and animals [17,19–21]. The lack of uniform implementation of well-defined MIRNA criteria, coupled with the challenging nature of informatically distinguishing miRNAs from noise or other sRNA species, has led to diminishing confidence in the overall quality of existing MIRNA annotations [24,27–30].
There have been considerable efforts to define MIRNA criteria to improve the quality of annotations [23–25,31]. Some miRNA databases contain a significant number of false positive annotations [28,30,32]. miRBase for example relies on researchers and peer reviewers to assess the validity of miRNAs before submission, and has adopted methods of determining confidence in these community-based annotations [27,32]. The current release of miRBase (V. 22.1) contains over 48,000 mature miRNA sequences from 271 diverse species including animals, plants, and some protists [32]. MirGeneDB has taken a different approach, manually curating MIRNA annotations of metazoan species through structural, expression, and conservation analysis [25,31,33]. Whether by database curators or the research community, the assessment of novel miRNAs relies on a degree of manual inspection and evaluation. However, manual inspection of incoming annotations takes significant effort and currently lacks standardized implementation.
While there are many de novo miRNA annotation tools and miRNA databases available, a secondary method to quickly analyze novel and annotated MIRNAs following genome-wide sRNA annotation is not available. Such a tool, to rapidly check new annotations, could be useful for database curators by removing the need for time-consuming manual inspection of new submissions. A standardized and quick method of automatically validating new MIRNA annotations would improve the quality of annotations published and subsequently submitted to online repositories. Retrospective application of such a method could also be used to flag and remove problematic entries in repositories such as miRBase. To address this need, we developed miRScore – a rapid and precise miRNA validation tool. miRScore can rapidly evaluate the annotation of both existing and novel miRNAs against specific sRNA-seq datasets using widely accepted MIRNA criteria in plants and animals. It offers a comprehensive evaluation of MIRNA loci, analyzing each criterion and producing visualizations of hairpin secondary structure and expression patterns. In this study, miRScore is described and tested using both annotated and novel MIRNAs from plants and animals.
Design and implementation
miRScore is implemented as a Python script that requires several commonly used bioinformatic tools including samtools [34], ViennRNA [35], and bowtie [36]. miRscore is an open-source software available under a permissive MIT license from GitHub at https://github.com/Aez35/miRScore, and is easily installed using Bioconda [37].
Workflow
miRScore validates MIRNA loci by analyzing the hairpin precursor sequence, miRNA duplex, and sRNA-seq data. The validation process is based on a set of previously described criteria which can be categorized as either structural (based on the predicted RNA secondary structure of the precursor) or expression (based on observations of miRNA and miRNA* abundance) (Table 1) [23–25,27,38,39]. miRScore utilizes a “pass/fail” system of reporting: each input MIRNA locus will ultimately either pass or fail. In some cases, one or more warnings may be raised for a “passed” entry if certain atypical features are present. A full list of flags and their explanations can be found in Table 2 and in the miRScore README. If one or more flags with a “fail” result are present, the locus will fail. Loci with no flags will pass, as will loci that have one or more flags associated with a “warning” result but no flags with a “fail” result (Table 2).
The primary use of miRScore is to rapidly assess novel MIRNA annotations prior to publication and submission to miRNA databases (Fig 1A). Users input properly formatted FASTQ or FASTA files containing sRNA sequencing reads, as well as miRNA duplex sequences and hairpin sequences in FASTA format (Fig 1B). The precursor sequences should be extended past the endonuclease cut sites, and the miRNA/miRNA* should not start or end the precursor sequence. This is to allow proper evaluation of the miRNA duplex structure. The identifier of each hairpin should be nearly identical to the corresponding mature miRNA, with the exception being that the mature identifier may contain ‘3p’, ‘5p’, or ‘mature’ and still be discerned. Users may also include miRNA* sequences in the mature FASTA file, provided they be distinguishable from the mature sequence by the either a “-5p”,”-3p”, “.star”, or “*” at the end of the name (i.e. miR399-3p, miR399.star, miR399*) (Fig 1C). The workflow of miRScore is to evaluate structural and expression criteria of all loci, assign a pass or fail result, reanalyze each failed locus for potential rescue (see below), and generate visualizations.
(A) miRScore is designed to follow MIRNA annotation in the miRNA analysis workflow. (B) Flow chart describing the inputs and steps of miRNA analysis by miRScore (C) Example of suitable names for sequences for input FASTA files. MIRNA hairpin identifier (ath-MIR399a) must match the mature miRNA sequence identifier (ath-miR399a); however, the miRNA* (ath-miR399a*) must have an identifier that distinguishes it from the mature miRNA sequence within the file. Created in BioRender [40].
Structural evaluation
miRScore predicts the secondary structure of single-stranded hairpin precursors using RNAfold from ViennaRNA [35]. The location of the miRNA and miRNA* sequences are indexed on the hairpin. If the user does not provide a miRNA* sequence, miRScore predicts it by determining the sequence that forms a miRNA/miRNA* duplex with a two-nucleotide 3’ overhang. miRScore then evaluates the miRNA duplex and hairpin against structural criteria (Table 1). This predicted secondary structure is used to determine characteristics such as the number of mismatches and large bulges within the duplex, or whether there is a two-nucleotide 3’ overhang.
Expression evaluation
The next step of the process is to evaluate expression-based criteria (Table 1). In this phase, miRScore quantifies miRNA abundance and calculates precision for each MIRNA locus. Reads from each library are mapped to the hairpin using bowtie version 1.3.1 [36]. All perfect alignments are retained if they align to the forward strand of the putative hairpin sequence. When counting miRNA and miRNA* reads, miRScore allows for one-nucleotide positional variance, which is included to account for biological variation in endonuclease processing during miRNA biogenesis [23]. miRScore requires at least ten reads within a single library to align to the miRNA/miRNA* duplex, allowing for one-nucleotide variants. Raw read counts are used as opposed to normalized values because we are primarily concerned with reproducibility (i.e., observing the miRNA/miRNA* multiple times in a sample) rather than comparative quantification between samples. miRScore then calculates precision for each locus in a library. Precision is defined as the number of reads that map to the miRNA/miRNA* duplex (including one-nucleotide positional variants) divided by the total number of reads which map to the hairpin precursor. The precision threshold is >= 75% (S1 Fig). Only libraries which meet these requirements will have their read count and precision values reported in the results file, but metrics for all libraries can be found within the ‘reads.csv’ file.
Identifying potential alternative mature miRNAs in failed loci
Optionally, MIRNA loci that fail the initial miRScore analysis are reanalyzed to determine if a different potential mature miRNA exists on the hairpin. This optional procedure is triggered if the user specifies the “-rescue” option in the run command. This feature may be helpful in cases where the initially annotated location of the mature miRNA within the hairpin does not agree with the observed sRNA-seq data. Reanalysis begins by determining the most abundant 20–24 nucleotide sequence that maps to the failed hairpin. miRScore then evaluates this sequence as an ‘alternative miRNA’ using structural and expression criteria (Table 1). If the locus now passes (Table 2), miRScore includes this potential ‘alternative miRNA’ in a separate alternative results CSV file. Read counts for all alternative miRNAs are reported in an additional alternative reads CSV file. Any potential “rescued” loci that emerge from this optional pipeline should be scrutinized manually before final annotation and submission to a miRNA registry.
Output and visualization
After assessing structural and expression-based criteria, miRScore generates a CSV file containing details about each locus along with the relevant flags (Table 2) and a pass/fail result. Lastly, miRScore generates figures for each submitted MIRNA locus for visualization of secondary structure and read depth.
Results and discussion
Performance analysis for annotated MIRNAs
miRScore is primarily designed as a quick secondary filter to analyze new MIRNA annotations prior to submission or acceptance into a permanent repository. Because it assesses the validity of an annotation with respect to specific sRNA-seq datasets, it is not appropriate to conclude that a miRNA whose annotation is not supported by specific datasets is not a bona fide miRNA. However, existing repositories contain multitudes of diverse annotations that have already been peer-reviewed and curated and as such are a good source of input data to evaluate the use and performance of miRScore. To this end, we obtained MIRNA annotations from miRBase version 22.1 and MirGeneDB version 3.0 from two animal species (Homo sapiens and Mus musculus) and three plant species from miRBase version 22.1 (Arabidopsis thaliana, Oryza sativa, and Zea mays) (Table 3). For each plant, five sRNA-seq libraries (S1 File) were acquired from the most frequently cited publication on miRBase that included suitable sRNA-sequencing data [41,42]. sRNA-seq data for animal species were acquired from MirGeneDB website. SRA accession numbers of sRNA-seq data from each species can be found in S1 File. miRScore version 0.3.2 was run using default settings for each dataset.
The primary output of miRScore is a pass/fail result for each locus, accompanied by flags which indicate specific criteria that a MIRNA locus did not meet with respect to the provided sRNA-seq datasets (Table 2). A single MIRNA locus may receive multiple flags if it fails to meet multiple criteria, and some flags are warnings instead of failures (Table 2). We evaluated the distribution of failed MIRNAs across structural and expression-based categories for all tested species (Fig 2A and 2B).
(A) miRScore results for animal MIRNA datasets from two databases. (Tan) Number of MIRNAs which failed miRScore due to expression criteria. (Blue) Number of MIRNAs which failed miRScore due to structural criteria. (Mauve) Number of MIRNAs which failed miRScore due to both expression and structural criteria. (Green) Number of MIRNAs which met all criteria and passed. (B) miRScore results for plant MIRNA datasets sourced from miRBase.
In miRBase animal datasets, many of the failed MIRNA loci failed due to both structural and expression criteria (Fig 2A). For example, 885 out of 1615 submitted H. sapiens MIRNAs failed due to structural reasons. 526 of these were due to having no 2 nt 3’ overhang within the duplex, often off by a single nucleotide (S2 Fig and S2 File). This was observed in several of the failed MIRNAs in the MirGeneDB dataset as well (S2 Fig and S2 File). The 2-nt overhang can be more challenging to interpret for miRNA/miRNA* duplexes which contain asymmetric bulges or large sets of mismatches near the ends. For example, hsa-MIR-9-P1 in MirGeneDB has an asymmetric bulge within the annotated 3’ overhang (S3 Fig and S2 File). These overhangs are interpreted by miRScore as a 3-nt overhang based on the pairing of the first nucleotide of the duplex on the 5’ arm and therefore fails. Most MIRNAs from MirGeneDB, which is a curated database, met all criteria within both tested datasets (Figs 2A and S2 and S2 File). For plant MIRNAs within miRBase, many failed to meet expression criteria in the given sRNA-seq libraries (Figs 2B and S4 and S3 File). For example, of the 256 failed A. thaliana MIRNAs, 173 had no mature or star reads in the analyzed sRNA-seq data, and 41 had a precision of less than 75% (S4 Fig and S3 File). Some of the failures could be attributed to possible tissue-specific or conditional accumulation of the mature miRNA such that the miRNA and/or the miRNA* were absent in the sRNA-seq data used for analysis [24]. These MIRNAs meet all structural criteria and would potentially pass given a set of libraries which support expression. Structure failures and some lowly expressed miRNAs could reflect the subset of miRBase annotations that are not true MIRNAs [27] or miRNAs made by non-canonical pathways, such as isomiRs and miRtrons.
miRScore visualizations of hairpin secondary structure and read depth for each input locus (Fig 3A–D) allow easy inspection of results with respect to the MIRNA criteria (Fig 3E). For example, inspection of the visualizations of ath-MIR399a (Fig 3A and 3B), an endogenous A. thaliana MIRNA, visually confirms that this locus meets all criteria (Fig 3E). Conversely, ath-MIR405a (Fig 3C and 3D) failed miRScore analysis due to unmet expression criteria (Fig 3E).
(A) ath-MIR399a RNAplot depicting secondary structure. Mature miRNA (orange) and miRNA* (blue) in RNAplot indicate where the user-provided sequence can be found within the hairpin precursor secondary structure. (B) ath-MIR399a Strucvis plot depicting read depth of all submitted libraries. (C) ath-MIR405a RNAplot depicting secondary structure. (D) ath-MIR405a Strucvis plot depicting read depth of all submitted libraries. (E) miRScore criteria and whether each locus met or failed those criteria.
Manual validation of annotated MIRNAs
To evaluate miRScore’s classification performance, each MIRNA locus across all five species was manually inspected to determine its actual condition with respect to the input sRNA-seq datasets (pass or fail). Manual inspection used a combination of data including plots of RNA secondary structure overlaid with annotation and alignment data (Fig 3A–D) and genome browser visualizations of aligned small RNA-seq data. Manual inspection using the criteria defined in Table 1 yielded no observations of false positives or false negatives in any of the analyzed results (Table 4). The difference in the number of true positive and true negative MIRNAs in each dataset is striking. There are several factors that affect the number of failed miRNAs. Between 26–58% of miRNAs in the various datasets failed due to expression criteria. In some cases, this is likely due to the small subset of sRNA-seq libraries used to evaluate performance, as miRNA expression can be tissue and condition specific [24,43]. Therefore, when using a more comprehensive set of libraries, these loci may well pass miRScore evaluation. For this reason, it is recommended that miRNAs be validated through miRScore using the same libraries used to annotate them whenever possible. As a corollary, it must be emphasized that these types of “failures” due to lack of accumulation in selected sRNA-seq libraries do not necessarily reflect incorrect annotations in the databases. Painstaking manual curation activities, including literature analysis, will still be required to confirm the validity of many existing annotations. Secondly, the number of miRNAs that fail due to structural criteria in miRBase is likely due to slightly offset annotations of the miRNA/miRNA* duplex positions. One example is hsa-MIR-202, whose miRBase annotation failed, but has an annotation in MirGeneDB which passes. Another reason for the number of structural failures in the animal datasets is the number of loci with mismatches or bulges near the terminal regions of the duplex. This can be challenging and interpretation of the 2-nt 3’ overhang may vary in these contexts. For example, hsa-let-7a, which contains an asymmetric bulge at the 5’ end of the miRNA, has a miRNA* that is annotated to have what miRScore interprets as a 1-nt 3’ overhang (S3 Fig).
Performance analysis for de novo MIRNAs
miRScore was primarily designed to evaluate new MIRNA annotations as a quick screen before or upon submission to databases. One common source of new annotations is those produced by tools that perform genome-wide de novo annotation such as ShortStack [17], miRador [19], and miRDeep-P2 [21]. We generated de novo MIRNA annotations in four plant species: Oryza sativa, Zea mays, Arabidopsis thaliana, and the parasitic plant Striga hermonthica. Striga hermonthica was included as it is currently an unannotated species with no MIRNAs cataloged in miRBase or any other database, allowing us to test novel MIRNA validation. Each annotation tool was run using the same five libraries used for the plant species in the annotated MIRNA dataset (Table 3). For S. hermonthica, novel small RNA-seq libraries were generated from leaf and haustorial tissue. miRScore was then run using annotated miRNAs from these results and the sRNA-seq data used for annotation.
miRDeep-P2 annotated the largest number of MIRNAs in each species, with over 2,000 MIRNAs from O. sativa (Fig 4A). Many loci annotated by miRDeep-P2 failed miRScore evaluation (Fig 4B and S4 File). Interestingly, many failed miRNAs had no miRNA* reads, which is a stated requirement for miRDeep-P2 [21]. Nearly all MIRNAs annotated by ShortStack passed miRScore inspection (Fig 4B and S4 File). miRador annotations had a pass rate between 80 and 95% with failing loci flagged for various criteria, most of which were expression-based (S4 File). For novel Striga hermonthica annotations, both ShortStack and miRador reported 68 passing MIRNAs, while miRDeep-P2 reported 127 passing loci.
(A) The number of MIRNAs predicted by three de novo annotation software by species. (B) Proportion of de novo MIRNAs from each software which passed or failed miRScore evaluation in each species including Arabidopsis thaliana (Ath), Oryza sativa (Osa), Striga hermonthica (She), and Zea mays (Zma).
Unlike annotation software which uses a merged sRNA-seq alignment file to evaluate expression of miRNAs, miRScore evaluates sRNA-seq libraries on an individual basis. This helps account for tissue specificity where a merged library may dilute precision to the point of failure. It also provides a readout of which libraries specifically a miRNA passes in and evidence of replication, which can be useful for downstream analysis. However, this may explain why some miRNAs pass during annotation but fail to meet miRScore criteria. It is worth noting that ShortStack, miRador, and miRScore all use RNAfold [35] to predict secondary structure. In addition, the congruence between Shortstack and miRScore at least partially reflects the shared authorship of the two tools. A workflow for each software is included in the supplementary material (S5 File). Overall, miRScore effectively evaluated the outputs of several MIRNA annotation tools in plants, confirming its utility in diverse annotation workflows.
Availability and future directions
miRScore is an open-source python code, and instructions are available on GitHub at https://github.com/Aez35/miRScore. A Bioconda recipe for miRScore is available, allowing easy installation using the conda package manager [37]. In this study, we demonstrate that miRScore effectively validates MIRNAs in both annotated and novel MIRNA datasets across plant and animal species. miRScore enables rapid and robust analysis of MIRNA loci, without requiring a reference genome, and offers detailed metrics and visualizations for each locus to support comprehensive analysis of MIRNA datasets. We believe that miRScore will contribute to improving MIRNA annotation quality and provide a tool for researchers to quickly verify annotations prior to downstream analysis. In addition, miRScore will provide an automated validation tool which will reduce the time it takes to validate novel MIRNAs prior to publication and database submission. miRScore’s ability to identify high confidence MIRNAs using widely accepted criteria quickly and accurately provides a valuable tool, contributing to the enhancement of MIRNA annotation quality in future studies.
Methods
Annotated datasets
Annotated MIRNAs from Oryza sativa, Arabidopsis thaliana, Zea mays, Mus musculus, and Homo sapiens were tested using miRScore version 0.3.2. Mature miRNA and hairpins sequences for each species were downloaded from either miRBase version 22.1 or MirGeneDB version 3.0. The miRNA file submitted to miRScore for each species contained sequences for both the annotated miRNA and miRNA* sequence where applicable. Several miRNAs from miRBase in some species began at position one of their respective hairpins and did not allow for evaluation of the miRNA duplex, particularly the two-nucleotide 3’ overhang structure. In these instances, miRScore will report the offending sequences and quit the run. These miRNAs were therefore removed from the dataset prior to running miRScore. This was not an issue with miRNAs sourced from MirGeneDB, as the database included an option to download extended precursors. Both miRBase and MirGeneDB data require some processing to harmonize naming conventions prior to miRScore analysis: Details are provided in S6 File. sRNA-seq data accession numbers and sources used for evaluation are listed in Table 3 and S1 File.
Processing and alignment of small RNA-seq data
Small RNA-seq data for each dataset were trimmed to remove 3’ adapters using the ‘-autotrim’ feature of miRScore. Trim keys for each dataset were set to the miRScore default of ath-miR166a (UCGGACCAGGCUUCAUUCCCC) for plants and hsa-let-7a (UGAGGUAGUAGGUUGUAUAGUU) for animals. The trimmed sRNA-seq data were aligned to hairpin sequences using bowtie [36] version 1.3.1 with options ‘-v0 -a –no-unal –norc’. The BAM alignment files were merged and read groups were used to count reads that mapped to each hairpin using samtools [34] version 1.20.
Striga hermonthica growth and sRNA-seq library preparation
Striga hermonthica Kibos ecotype was grown on host Oryza sativa ssp. japonica variety Kitaake under 16-hour light conditions in a quarantine facility at 30°C for 45 days. Haustorium and leaf tissue were collected from S. hermonthica and total RNA was extracted using Zymo Quick-RNA Plant Miniprep Kit. sRNA-seq libraries were prepared essentially as described in Maguire et al. [44]. Sequencing of the prepared libraries was performed on an Illumina NextSeq2000. New small RNA-seq libraries from S. hermonthica have been deposited at NCBI GEO under accession GSE282265.
de novo MIRNA datasets
MIRNAs were annotated in four plant species using ShortStack version 4.0.4 [17], miRador commit c68c153 [19], and miRDeep-P2 version 1.1.4 [21] (See S5 File). All annotation software were run on default settings for de novo MIRNA discovery. Genome assembly versions were: Arabidopsis thaliana (TAIR 10), Oryza sativa (IRGSP-1.0), Striga hermonthica assembly SHERM (GCA_902706635.1), and Zea mays (Zm-B73-REFERENCE-NAM-5.0). sRNA-seq data for Striga hermonthica was generated as described above. sRNA-seq data for the other plant species were acquired from accession numbers in S1 File. Mature miRNA and hairpin sequences from resulting annotations were parsed and saved to two separate FASTA files. These FASTA files were used to test validation of de novo MIRNAs using miRScore version 0.2.0. The same sRNA-seq data used for MIRNA annotation were used to run miRScore for each species.
Supporting information
S1 File. SRA accession numbers of sRNA-seq libraries used for testing miRScore on each dataset.
https://doi.org/10.1371/journal.pcbi.1013663.s001
(XLSX)
S2 File. miRScore results file for miRNA datasets from MirGeneDB and miRBase for two animal species (Homo sapiens and Mus musculus).
https://doi.org/10.1371/journal.pcbi.1013663.s002
(XLSX)
S3 File. miRScore results file for miRNA datasets from MirGeneDB and miRBase for three plant species (Arabidopsis thaliana, Oryza sativa, and Zea mays).
https://doi.org/10.1371/journal.pcbi.1013663.s003
(XLSX)
S4 File. miRScore results files from de novo MIRNAs from three annotation software of four plant species.
https://doi.org/10.1371/journal.pcbi.1013663.s004
(XLSX)
S5 File. Markdown pdf file describing de novo annotation pipeline for plant miRNAs used to test miRScore handling of de novo annotations.
https://doi.org/10.1371/journal.pcbi.1013663.s005
(PDF)
S6 File. Markdown file describing how to prepare miRBase and MirGeneDB data for running miRScore.
https://doi.org/10.1371/journal.pcbi.1013663.s006
(PDF)
S7 File. R script file for generating Figs 2 and 4.
Format: Plain text/ R code (.R).
https://doi.org/10.1371/journal.pcbi.1013663.s007
(R)
S1 Fig. Explanation of read alignment, precision, and variance.
(A) When counting miRNA duplex reads, a variance window of -/ + 1 nt from the indexed start/stop position of the miRNA and miRNA*. Reads which start and stop within this window are counted towards the total miRNA duplex count and used to determine precision. (B) Example of reads which are included in total count of hsa-mir-212 miRNA (red), miRNA* (blue), and those that are not included in count (black). Read length (len) and number of reads aligned at that position (al) can be found on the right side. (C) Example of reads included in ath-MIR167a miRNA (red), miRNA* (blue), and those that are not included (black).
https://doi.org/10.1371/journal.pcbi.1013663.s008
(DOCX)
S2 Fig. Upset plot of result and flags for MIRNAs sourced from miRBase and MirGeneDB for animal species.
See S2 File for source data. (A) Mus musculus (mmu) mirbase MIRNA results and flags. (B) Homo sapiens (hsa) mirbase MIRNA results and flags. (C) mmu MirGeneDB MIRNA results and flags. (D) hsa MirGeneDB MIRNA results and flags.
https://doi.org/10.1371/journal.pcbi.1013663.s009
(DOCX)
S3 Fig. RNAplots of MIRNA secondary structures.
(A) Plot of Hsa-Mir-9-P1 from MirGeneDB with annotated miRNA (orange) and miRNA* (blue). (B) Plot of hsa-let-7a-1 from miRBase with annotated miRNA (orange) and miRNA* (blue).
https://doi.org/10.1371/journal.pcbi.1013663.s010
(DOCX)
S4 Fig. Upset plot of results and flags for MIRNAs sourced from miRBase for plant species.
See S3 File (A) Arabidopsis thaliana (ath) MIRNA results and flags. (B) Oryza sativa (osa) MIRNA results and flags. (C) Zea mays (zma) MIRNA results and flags.
https://doi.org/10.1371/journal.pcbi.1013663.s011
(DOCX)
Acknowledgments
We thank the Penn State Genomics Core Facility (RRID: SCR_023645) for sRNA-seq services. We thank Steven Runo and Claude dePamphilis for the gift of Striga hermonthica seed.
References
- 1. Reinhart BJ, Weinstein EG, Rhoades MW, Bartel B, Bartel DP. MicroRNAs in plants. Genes Dev. 2002;16(13):1616–26. pmid:12101121
- 2. Tang G, Reinhart BJ, Bartel DP, Zamore PD. A biochemical framework for RNA silencing in plants. Genes Dev. 2003;17(1):49–63. pmid:12514099
- 3. Ambros V. The functions of animal microRNAs. Nature. 2004;431(7006):350–5. pmid:15372042
- 4. Bajczyk M, Jarmolowski A, Jozwiak M, Pacak A, Pietrykowska H, Sierocka I, et al. Recent Insights into Plant miRNA Biogenesis: Multiple Layers of miRNA Level Regulation. Plants (Basel). 2023;12(2):342. pmid:36679055
- 5. Olsen PH, Ambros V. The lin-4 regulatory RNA controls developmental timing in Caenorhabditis elegans by blocking LIN-14 protein synthesis after the initiation of translation. Dev Biol. 1999;216(2):671–80. pmid:10642801
- 6. Palatnik JF, Allen E, Wu X, Schommer C, Schwab R, Carrington JC, et al. Control of leaf morphogenesis by microRNAs. Nature. 2003;425(6955):257–63. pmid:12931144
- 7. Wu G, Park MY, Conway SR, Wang J-W, Weigel D, Poethig RS. The sequential action of miR156 and miR172 regulates developmental timing in Arabidopsis. Cell. 2009;138(4):750–9. pmid:19703400
- 8. Liang G, He H, Yu D. Identification of nitrogen starvation-responsive microRNAs in Arabidopsis thaliana. PLoS One. 2012;7(11):e48951. pmid:23155433
- 9. Li M, Chen T, Wang R, Luo J-Y, He J-J, Ye R-S, et al. Plant MIR156 regulates intestinal growth in mammals by targeting the Wnt/β-catenin pathway. Am J Physiol Cell Physiol. 2019;317(3):C434–48. pmid:31166713
- 10. Gao N, Qiang XM, Zhai BN, Min J, Shi WM. Transgenic tomato overexpressing ath-miR399d improves growth under abiotic stress conditions. Russ J Plant Physiol. 2015;62(3):360–6.
- 11. Isik M, Blackwell TK, Berezikov E. MicroRNA mir-34 provides robustness to environmental stress response via the DAF-16 network in C. elegans. Sci Rep. 2016;6:36766. pmid:27905558
- 12. Barciszewska-Pacak M, Milanowska K, Knop K, Bielewicz D, Nuc P, Plewka P, et al. Arabidopsis microRNA expression regulation in a wide range of abiotic stress responses. Front Plant Sci. 2015;6:410. pmid:26089831
- 13. Ma X, Zhao F, Zhou B. The characters of non-coding RNAs and their biological roles in plant development and abiotic stress response. Int J Mol Sci. 2022;23(8):4124.
- 14. Millar AA, Waterhouse PM. Plant and animal microRNAs: similarities and differences. Funct Integr Genomics. 2005;5(3):129–35. pmid:15875226
- 15. O’Brien J, Hayder H, Zayed Y, Peng C. Overview of microRNA biogenesis, mechanisms of actions, and circulation. Front Endocrinol. 2018;9:402.
- 16. Medley JC, Panzade G, Zinovyeva AY. microRNA strand selection: Unwinding the rules. Wiley Interdiscip Rev RNA. 2021;12(3):e1627. pmid:32954644
- 17. Axtell MJ. ShortStack: comprehensive annotation and quantification of small RNA genes. RNA. 2013;19(6):740–51. pmid:23610128
- 18. Johnson NR, Yeoh JM, Coruh C, Axtell MJ. Improved Placement of Multi-mapping Small RNAs. G3 (Bethesda). 2016;6(7):2103–11. pmid:27175019
- 19. Hammond RK, Gupta P, Patel P, Meyers BC. miRador: a fast and precise tool for the prediction of plant miRNAs. Plant Physiol. 2023;191(2):894–903. pmid:36437740
- 20. Friedländer MR, Mackowiak SD, Li N, Chen W, Rajewsky N. miRDeep2 accurately identifies known and hundreds of novel microRNA genes in seven animal clades. Nucleic Acids Res. 2012;40(1):37–52. pmid:21911355
- 21. Kuang Z, Wang Y, Li L, Yang X. miRDeep-P2: accurate and fast analysis of the microRNA transcriptome in plants. Bioinformatics. 2019;35(14):2521–2. pmid:30521000
- 22. Brown M, Suryawanshi H, Hafner M, Farazi TA, Tuschl T. Mammalian miRNA curation through next-generation sequencing. Front Genet. 2013;4:145. pmid:23935604
- 23. Axtell MJ, Meyers BC. Revisiting Criteria for Plant MicroRNA Annotation in the Era of Big Data. Plant Cell. 2018;30(2):272–84. pmid:29343505
- 24. Taylor RS, Tarver JE, Hiscock SJ, Donoghue PCJ. Evolutionary history of plant microRNAs. Trends Plant Sci. 2014;19(3):175–82. pmid:24405820
- 25. Fromm B, Billipp T, Peck LE, Johansen M, Tarver JE, King BL, et al. A Uniform System for the Annotation of Vertebrate microRNA Genes and the Evolution of the Human microRNAome. Annu Rev Genet. 2015;49:213–42. pmid:26473382
- 26. Zhao Y, Kuang Z, Wang Y, Li L, Yang X. MicroRNA annotation in plants: current status and challenges. Brief Bioinform. 2021;22(5):bbab075. pmid:33754625
- 27. Kozomara A, Griffiths-Jones S. miRBase: annotating high confidence microRNAs using deep sequencing data. Nucleic Acids Res. 2014;42(Database issue):D68-73. pmid:24275495
- 28. Fromm B, Zhong X, Tarbier M, Friedländer MR, Hackenberg M. The limits of human microRNA annotation have been met. RNA. 2022;28(6):781–5. pmid:35236776
- 29. Ludwig N, Becker M, Schumann T, Speer T, Fehlmann T, Keller A, et al. Bias in recent miRBase annotations potentially associated with RNA quality issues. Sci Rep. 2017;7(1):5162. pmid:28701729
- 30. Meng Y, Shao C, Wang H, Chen M. Are all the miRBase-registered microRNAs true? A structure- and expression-based re-examination in plants. RNA Biol. 2012;9(3):249–53. pmid:22336711
- 31. Clarke AW, Høye E, Hembrom AA, Paynter VM, Vinther J, Wyrożemski Ł. MirGeneDB 3.0: improved taxonomic sampling, uniform nomenclature of novel conserved microRNA families and updated covariance models. Nucleic Acids Research. 2025;53(D1):D116–28.
- 32. Kozomara A, Birgaoanu M, Griffiths-Jones S. miRBase: from microRNA sequences to function. Nucleic Acids Research. 2019;47(D1):D155-62.
- 33. Umu SU, Paynter VM, Trondsen H, Buschmann T, Rounge TB, Peterson KJ, et al. Accurate microRNA annotation of animal genomes using trained covariance models of curated microRNA complements in MirMachine. Cell Genom. 2023;3(8):100348. pmid:37601971
- 34. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics. 2009;25(16):2078–9.
- 35. Lorenz R, Bernhart SH, Höner Zu Siederdissen C, Tafer H, Flamm C, Stadler PF, et al. ViennaRNA Package 2.0. Algorithms in Molecular Biology. 2011;6(1):26.
- 36. Langmead B, Trapnell C, Pop M, Salzberg SL. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 2009;10(3):R25. pmid:19261174
- 37. Grüning B, Dale R, Sjödin A, Chapman BA, Rowe J, Tomkins-Tinch CH, et al. Bioconda: sustainable and comprehensive software distribution for the life sciences. Nat Methods. 2018;15(7):475–6. pmid:29967506
- 38. Rolle K, Piwecka M, Belter A, Wawrzyniak D, Jeleniewicz J, Barciszewska MZ, et al. The Sequence and Structure Determine the Function of Mature Human miRNAs. PLoS One. 2016;11(3):e0151246. pmid:27031951
- 39. Ambros V, Bartel B, Bartel DP, Burge CB, Carrington JC, Chen X, et al. A uniform system for microRNA annotation. RNA. 2003;9(3):277–9. pmid:12592000
- 40.
Zvarick A. Figure 1. Workflow and input of miRScore. [Internet]. Created in Biorender; Available from: https://BioRender.com/p32t100
- 41. Zhang L, Chia J-M, Kumari S, Stein JC, Liu Z, Narechania A, et al. A genome-wide characterization of microRNA genes in maize. PLoS Genet. 2009;5(11):e1000716. pmid:19936050
- 42. Breakfield NW, Corcoran DL, Petricka JJ, Shen J, Sae-Seaw J, Rubio-Somoza I, et al. High-resolution experimental and computational profiling of tissue-specific known and novel miRNAs in Arabidopsis. Genome Res. 2012;22(1):163–76. pmid:21940835
- 43. Londin E, Loher P, Telonis AG, Quann K, Clark P, Jing Y, et al. Analysis of 13 cell types reveals evidence for the expression of numerous novel primate- and tissue-specific microRNAs. Proc Natl Acad Sci U S A. 2015;112(10):E1106-15. pmid:25713380
- 44. Maguire S, Lohman GJS, Guan S. A low-bias and sensitive small RNA library preparation method using randomized splint ligation. Nucleic Acids Res. 2020;48(14):e80. pmid:32496547