InteRRact

Analyse single dataset

This module enables interactive exploration of a single RNA–RNA interaction dataset derived from proximity ligation or duplex probing experiments. It is designed for hypothesis-driven inspection of one experiment or condition at a time. Trans interactions are visualised as a gene-level interaction network, while locus-level evidence can be inspected in a linear track view. You can dynamically filter interactions to focus on robust duplex groups and explore community structure within the network.

Key Controls & data filters

• Dataset selection
  Choose experiment / cell line / condition.

• Minimum reads
  Require a minimum number of supporting reads per duplex group.

• Minimum replicates
  Require detection in multiple replicates.

• Minimum component size
  Remove small disconnected subgraphs to focus on hubs and structured communities.

• RRIs per gene pair
  Require more than N duplex groups between two genes.

• Hybridisation energy (ΔG)
  Filter interactions by predicted duplex stability.

• P-value threshold
  Retain statistically supported interactions for the binominal random ligation test.

• P-value threshold for the odds ratio test
  Retain statistically supported interactions for the odds-ratio ‘interaction orrured by chance’ test.

• Gene search / coordinate search
  Focus on a gene of interest or a genomic locus.

• Network physics & colouring
  Adjust layout behaviour and colour nodes by community.

These controls expose the confidence metrics directly derived from the experimental library and downstream statistical processing.

Outputs

• Interactive trans RNA–RNA interaction network
• Community detection results
• Selected node panel with gene metadata and external links
• Selected edge panel with underlying duplex groups
• Embedded linear view for locus-level inspection
• GO enrichment per network community
• Tabular view pane to browse and export displayed RRI data

Show-case tutorial

This tutorial explains how to use InteRRact in the single-dataset mode using a biologically motivated example based on recently published work. Bergeron et al. analysed PARIS, LIGR-seq and SPLASH datasets and identified pletora of snoRNA-host transcript interactions. The central validated interaction in their analysis is snoRNA-mRNA interaction: SNORD2 <-> EIF4A2. SNORD2 resides on EIF4A2 and interacts with its host intron near exon 4 and regulates EIF4A2 splicing by sequestering the nucleotides at the branch point.

We will use InteRRact to locate the SNORD2-EIF4A2 interaction in the trans RNA-RNA interaction network, inspect its supporting duplex groups, and examine the corresponding locus in the linear genomic view. We then zoom out to broader outlook in order to demonstrate that the same workflow can be extended to many other snoRNAs in the dataset.

1. locating the SNORD2-EIF4A2 interaction in the network

We will use the LIGR-seq dataset, as this LIGR-seq protocol produces the largest amount of small RNA interactions, including snoRNAs. For this example, we keep the default confidence filters unchanged. To make the graph easier to interpret and pinpoint the RNA biotypes, the colouring rule can be set to gene type.

To focus on the interaction of interest, we enter SNORD2 in the gene search field and request display of only n = 1 interacting neighbour in the graph. This reduces the network to the immediate local neighbourhood of SNORD2 and makes it easier to inspect the relevant interaction.

The resulting graph shows SNORD2 together with its interacting partner EIF4A2. Selecting the SNORD2 node fills the node information panel with the basic properties of the gene, including its identifier, type and number of interacting partners. Selecting the SNORD2-EIF4A2 edge fills the edge information panel and exposes the interaction as a set of underlying duplex groups.

2. Examining the interaction details and the genomic locus

Once the SNORD2-EIF4A2 edge is selected, the detailed edge view shows the number of duplex groups supporting the interaction together with their confidence-related measures. The full set of confidence features can also be downloaded from this panel.

Selecting an entry in the duplex groups table shows the schematic representation of the hybrid on the two duplex-group arms. The same selection can then be used to switch directly to the linear genomic view.

To load the linear representation, go to the corresponding panel and click load linear tracks. The IGV browser then loads the gene annotation and interaction tracks. Because a specific duplex group has already been selected in the edge view, the option focus both panels on selected duplex group can be used to centre the top and bottom browser panels on the two arms of that duplex group.

In this example, SNORD2 interacts with its host gene, so both browser panels point to the same EIF4A2 locus. Bergeron et al. proposed that SNORD2 may affect host-gene splicing before its own maturation by forming an intramolecular RNA structure involving the branch point. In the linear view, duplex group ID:11862 is visibly extended beyond the annotated genomic boundaries of SNORD2. This indiates that underlying chimeric reads map outside the snoRNA gene coordinates, supporting the interpretation that the detected interaction may act in cis, in agreement with the model discussed in the paper.

This step shows how InteRRact can be used to move from a network edge to the underlying duplex evidence and then to the locus-level representation of the same interaction.

3. Extending the same workflow to other snoRNAs

The SNORD2-EIF4A2 interaction is an example of functional snoRNA-mRNA interaction which has been validated independently from the proximity-ligation data. However, many other intreactons of this type, that may have non-canonical biological function in trans could be observed in the LIGR-seq.

The same workflow can be applied more broadly to other snoRNAs present in the dataset. We remain in the LIGR-seq dataset and use snoRNA-centred search to browse for other snoRNA- interactions. Confidence filters can be tightened at this stage to retain only more strongly supported interactions.

With the graph coloured by gene type, snoRNA nodes can be located easily in the network represetation. For any interaction of interest, you can move from the network to the selected edge, inspect the associated duplex groups, and then immidiatly switch to the complementary linear genomic view .

In this way, InteRRact can be used not only to folow up on the investigated examples such as SNORD2-EIF4A2, but also to identify many additional interactions that may serve as candidates for further exploration.

Compare two datasets

This module allows the qualitative comparison of two RNA–RNA interaction datasets. It visualises the union of the two RNA-RNA interaction networks and encodes whether nodes and edges are shared or dataset-specific. The comparison mode addresses the lack of standardised approaches for comparing the outcome of two RNA–RNA interaction experiments.

Key Controls & data filters

Input filters for the Datasets A & B

Confidence filters are applied to each dataset prior to gathering the super-set of the RRI. They are similar to the filters used in single-dataset module.

• minimum reads
• replicate support
• ΔG threshold
• p-value thresholds
• coverage score

Input filters for the combined RRI network

After two datasets are compared, the combined RRI set is created. It can be filtered in following ways:

• Show dataset-specific communities You can either select the RRI specific to the dataset A or B, or both. Additionally, if the community consists of RRI from different datasets i.e same gene has different RRI, such community is labelled as mixed. Mixed communities also could be isolated and expored.

• Minimum component size
  Removes small disconnected subgraphs to focus on hubs and structured communities.

• Minimum RRIs per gene pair
  Require more than N duplex groups between two genes.

Aestatics for the node and edge in the RRI network

• Show/hide edges with unmatched RRI

Whether to show the edges where distinct (unmatched) RRIs present between the gene in each dataset.

• Minimum matched interaction threshold

• Minimum % of matched RRI to label edge as matched
  Define how many/which fraction of all edge RRI are required for an edge to be annotated as matched and drawn normally in green. I.e the edge between the two genes that is formed by four RRI, of which the half is present in both datsaets, and another half are distinct can be drawn as the conserved or mixed one. These options allow you to customise the rule for this.

• Gene / coordinate search
  Focus on specific genes within the union graph.

Output :

• Union network of both datasets
• Encoding node by color and shape: shared vs dataset-specific genes
• Encoding edge by color:
  • Dataset A only
  • Dataset B only
  • Shared (matched)
  • Shared (unmatched)
• Clustered duplex groups for selected edges
• Community structure on pooled interaction graph
• Linear genomic browser for RRI in each dataset
• Tabular view pane to browse and export displayed data or the combined trans RRI where matched and not matched RRIs are annotated.

Module usage example

Exploring H1-specific communities

To showcase the pairwise comparison module, we will use H1 and hNPC cell lines from RIC-seq data. Our aim will be to the combined network to identify interactions that are specific to the pluripotent or neuronal progenitor state. We configure the data filtering parameters to moderately strict, requiesting only RRI which are observed in 3/4 replicates and use the odds-ratios test p-values.

We first select communities present only in H1. This allows us to focus on RNA interaction programs that are absent from, or rewired in, hNPC cells.

Several signal appear immediately: the graph contains well-known pluripotency-associated markers, including LIN28A. LIN28A is a well-known regulator of stem-cell identity. Another gene HMGA1 is a non-histone chromatin protein implicated in embryonic development and maintenance of stem-like cellular states.

In addition to these genes, MALAT1 and several ENSG-annotated loci, many of which correspond to lncRNAs or less well-characterized transcripts suggest that non-coding RNAs are also part of the H1-specific interaction landscape. This highlights an important feature of the pairwise comparison view: it captures not only known regulators, but also candidate non-coding components of the same cell-type-specific network.

Importantly, not all biologically relevant changes form large communities. Some of the most specific differences between the cell lines are represented by two-nodes, single-edge components, which we filtered out by restricting the analysis to components larger than n = 3.

Exploring hNPC-specific communities

We observe the same pattern in the opposite direction when selecting communities present only in hNPCs. In this case, the graph no longer highlights pluripotency-associated factors but instead suggests interactions related to neural progenitor identity and neuronal differentiation, consistent with the cell type. Several cliques containing genes associated with the hNPC phenotype are present. For example, QKI RBPs are known regulators of glial differentiation, and APP has established roles in neurodevelopment.

Together with the H1-specific graph, this comparison shows that the pairwise trans interaction view can recover biologically meaningful differences between cell states: H1-specific communities highlight pluripotency-associated programs, whereas hNPC-specific communities reveal neural progenitor-associated rewiring.

Exploring mixed communities

Finally, we swith to communities, which contain genes interating with different partners in NPCs and H1. Such mixed communities highlight a different type of relationship between the two cell types. Rather than being fully specific to either H1 or hNPC, these components contain a shared interactor* together with **cell-type-specific edges

In such communities, the same gene can participate in different interactions in the two datasets. This makes the pairwise trans interaction graph useful not only for identifying fully H1- or hNPC-specific programs, but also for detecting cases where a gene is present in both cell types while its interaction partners change between conditions.

Thus, mixed communities provide a complementary view to the exclusive ones: instead of showing fully gained or lost modules, they reveal shared RNA interaction neighborhoods that are rewired between H1 and hNPC.

Compare multiple datasets

This module allows you to see the results of multiple comparisons across more than two conditions or experiments. It summarises overlaps between matched interactions across selected datasets and presents them as an interactive UpSet plot. Additionally, you can include one RRI dataset passed from the visualise your data module. Interactions contributing to each overlap category can be browsed, filtered by confidence measures and exported in bedpe format.

Input and options

Because datasets may originate from different experimental protocols and may differ in the range and availability of confidence features, the initial superset is assembled without applying confidence thresholds.

The required inputs for the UpSet plot are:

• the list of datasets to include in the comparison
• the number of overlap categories to show
• selection from one of the pre-sets for controlling the display of labels and their size in the plot.

Please note that rendering of the plot is handled by the client browser. If more than six datasets are included in the comparison, rendering of the plot may take up to 1-2 minutes.

Output

• Interactive UpSet plot summarising overlaps between selected datasets
• Tabular browser for matched interactions in the selected overlap category
• Export pane for downloading the selected RRIs

Once the comparison plot is ready, you can click on the overlap category, which will display the tabular view and activate the data filters. Tabular data include all RRIs in each dataset and can be ordered by a special variable Comp.Group that labels the overlapping RRIs.

User-provided set of RRIs

You can upload your custom set of RRIs to the visualise your data module, which, once the import is successful and the module has been launched, can also be matched against datasets already hosted on InteRRact. Upon successful saving of the data, you will see a short message in the bottom-right corner (while in visualise your data). In the multiple comparison module, the filename of your upload should appear in the list of available data.

Computing the overlaps between your RRIs and datasets could take up to several minutes, depending on the size of your input. The current maximum allowed number of user-imported RRIs is 5000.

Note that user input is treated as a special case, as it is not included while the superset is being computed. Instead, user-provided RRIs are directly matched against the superset. The unique user RRIs are added to the superset as mismatching, while overlapping ones are recorded as matched. Importantly, only a single hit of a user-provided RRI per corresponding superset RRI is added to the comparison statistics. In other words, if your RRIs are redundant, i.e. you have many duplex groups where both arms overlap, only one overlap per such redundant set adds count to the comparison. You still retain all uploaded RRIs in the tabular browser.

Underlying comparison procedure

Each selected dataset contributes its processed RNA-RNA interactions to a common superset of matched interactions. Overlaps are then computed across all selected datasets and shown in the UpSet plot. The following figure details the procedure.

Infromation on source data & processing

RNA proximity ligation experiments, as well as the methods reliant on capturing disjoint RNA fragments bound by proteins yeild information on the RNA-RNA contacts. Contrary to the SHAPE- and DMS map RNA probing, these methods aim at identifying the interaction loci of single or multiple RNA rather than accessing the flexibility of RNA molecule though devising the reactivity scores. Very simplisticly, It coud be viewed as the RNA analog DNA-DNA contacts (Hi-C), although RNA duplex probing methods have considerably lower coverage and differ in most aspects from experiment to data handling and interepretation. For the more detailed information we refer the user to the technology review and here for the details on data processing method we use for the data hosted on InteRRact. The acession number and statistics on the filtered data could be found in the dataset statistics page.

Overview of the main steps for RNA-RNA interaction data processing.

Raw sequencing data are processed using a standardised bioinformatic workflow to detect, cluster, annotate and filter RNA–RNA duplex groups prior to visualisation.

1. Fetching the raw read data
2. Quality filtering with fastp
3. Read mapping and chimeric alignment with STAR
4. Duplex group detection with DuplexDiscovereR
5. Aggregating replicates, addition of the complementary RNA-seq data
6. Annotation and metadata integration: labelling of the repetitive elements and blacklisted region.
7. Computing confidence metrics and filtering
8. Representing data for the InteRRact views: trans RRI as a network, all RRI in linear genomic browser.

Data retrieval and pre-processing

Hosted datasets in InteRRact were collected from published RNA duplex probing experiments and pre-processed using a common workflow before loading into the server.

Raw sequencing reads were first cleaned with fastp to remove low-quality reads and adapters. For PARIS datasets, additional protocol-specific preprocessing was applied using the icSHAPE pipeline scripts, following the procedures used for the original protocol.

Reads were then aligned to the human reference genome (GRCh38) with STAR in chimeric alignment mode. Following alignment settings were chosen to recover reags originating from RNA-RNA ligation.

--chimOutType Junctions
--chimOutJunctionFormat 1
--alignIntronMin 1
--alignIntronMax 10
--outSJfilterReads All
--chimSegmentMin 15
--chimMultimapNmax 10
--chimScoreDropMax 30
--chimScoreJunctionNonGTAG 0

The resulting Chimeric.out.Junction files were used as the input for downstream interaction calling.

Duplex group detection

Candidate interactions are clustered into duplex groups using DuplexDiscovereR. A duplex group represents a set of overlapping or closely spaced chimeric reads supporting the same RNA-RNA interaction.

This grouping reduces redundancy at the read level and provides a compact interaction unit for filtering, visualisation and comparison. For each duplex group, InteRRact stores a set of summary measures that can be used during downstream exploration, including:

• number of supporting reads
• replicate support
• locus coverage
• predicted hybridisation energy (ΔG)
• statistical significance measures

For datasets with replicates, duplex groups detected in individual replicates of the same condition are further aggregated into condition-level duplex groups. This allows the server to display a more compact representation while preserving replicate support and the main confidence-related features.

Statistical asessment

InteRRact employs two statistical models for asessment the RRI confidence.

1. A binomial random-ligation model, where pair probability is derived from per-gene/transcript counts and tested using the right-tail binomial test.

2. An odds-ratio / Fisher exact test based on a 2x2 contingency table using chimeric counts and singleton counts.

Annotation and metadata integration

Detected duplex groups are annotated with:

• gene identifiers (gene-level mapping)
• transcript coordinates
• repeat annotations (RepeatMasker)
• ENCODE blacklist regions
• gene biotype

Duplex groups mapping to simple repeats, low-complexity regions or high-mapping-signal regions were excluded from the main set hosted datasets. This filtering step reduces interactions that are more likely to reflect mapping artefacts or transcriptional noise than bona fide RNA-RNA duplexes. If you are interested in the RNA-RNA interactions that may occur with RNA transcribed from those problematic regions, you can disable pre-filtereing in the Advanced options sidebar pane (Visualise single dataset and Pairwise comparison modules).

Representation of data in InterRRact

Network construction

Filtered duplex groups are aggregated at the gene level to construct trans RNA–RNA interaction networks.

• Nodes represent genes.
• Edges represent one or more duplex groups between gene pairs.

Community detection is performed using a greedy modularity optimisation algorithm (Louvain). Gene Ontology (GO) over-representation analysis is performed per detected community.

Network construction for the pairwise comparison module

For dataset comparison:

• Filtered duplex groups from two datasets are pooled.
• Matching criteria are applied based on overlapping or clustered duplex groups.
• Edges are classified as:
  • dataset A specific
  • dataset B specific
  • shared (matched)
  • shared (unmatched)

Community detection is performed on the pooled interaction graph.

Confidence metrics and filtering

RNA–RNA interaction datasets are inherently sparse and prone to experimental artefacts. Therefore, multiple orthogonal confidence metrics are retained and exposed in the interface:

• Read count (number of supporting chimeric reads)
• Replicate support
• Hybridisation energy (ΔG)
• Two statistical tests output (p-values)
• Locus coverage score
• Repeat / blacklist annotations

These parameters allow users to apply principled filtering based on interaction robustness.

InteRRact features

• Interactive visualisation of RNA-RNA interactions in multiple modes: Combination of network representation of trans RRIs and linear browser views provides flexibility and convenience over traditional RNA database views that store tabular data.
• Interactive filtering by confidence, gene and coordinate search: Quickly focus on genes or genomic loci of interest, and examine the underlying duplex groups and base-pairing evidence.
• Multiple comparison modes: Novel graph-based framework for pairwise comparison and interactive UpSet plot for intuitive browsing conserved and unique RRIs across multiple datasets.
• Data export: Export interaction subsets for your own analysis.
• Standards compatibility: Interactions are represented in standard BEDPE format and are compatible with the bedtools and commonly used ecosystems for working with genomic tracks (Bioconductor, UCSC).

Comprehensive and explicit measures of the RRI detection confidence.

RNA–RNA interaction datasets are inherently sparse and prone to experimental and computational artefacts. Only a small fraction of sequencing reads are chimeric, crosslinking and ligation efficiencies vary, and bioinformatic processing can introduce false positives.

InteRRact exposes multiple confidence-related parameters to allow data filtering:

• number of supporting reads
• replicate support
• hybridisation energy (ΔG)
• statistical tests output (two types of p-values)
• loci coverage score derived from the continious fragments in the library

These metrics provide orthogonal evidence for duplex reliability and help assess robustness of detected interactions.

Technical limitations of the web-server.

• Rendering networks with more than 1000 nodes may slow down your browser performance. We address this by disabling the network physics and additionally, by providing the convenient interface option to reset the page, while keeping the previous parameters. This may speed-up the responsiveness after displaying large graph.
• We provide the browser safe mode that prevents the display of particulary large graphs larger than 2000 nodes. Very alrge graphs are generally not intuitive for browsing, but potentially can crash the browser. You need to be disable the browser safe options to allow such displays to be rendered it. We suggest to render the graph with disabled physics and enable it only once it is displayed.
• A current limit for the user-uploaded data is set to 5000 RRI.
• GO enrichment annotation depends on detected communities and takes substantial ammount of time to compute. InteRRact empoys internal caching, however, computation can take 2-10 min per single view. As an alternative, the enrichment analysiscould be conducted in the specialised online tool. To do this, InteRRact supports quick export of the the gene set defined by the trans RRI community.
• Currently InteRRact is focused to human cell lines and utilises the hg38 assembly.
• The interaction network focusses on genes rather than individual transcripts. This is a deliberate choice, as resolving the transcript identity for the RRI is not always possible and simplifying decisions such as only considering longest or the most expressed isoform may not constitute the correct assumption to your analysis. This is why we provide gene coordinates which constitute a solid foundation for any subsequent transcript-specific analysis.

Citation is not avaliable yet

News

• 14.04.2026 InteRRact is put online. The manuscript is submitted.
• 02.06.2026 Data is updated to the unfiltered version. Filtered datasets are kept to be the default. New options for IGV and minor UI changes in landing page.