User Tutorial

CHARM

Comprehensive Hub of Alternative Regulatory Mapping

A guide to exploring RNA-binding protein expression, splicing, and eCLIP binding data, all in one webapp.

Based on: Kaizeler, A. & Barbosa-Morais, N.L. “CHARM: Comprehensive Hub of Alternative Regulatory Mapping of RNA-Binding Proteins.”  |  App: v0.9.3  |  Looking for the local pipeline instead? See eCLIPSE Local Pipeline.

1. Introduction

RNA-binding proteins (RBPs) regulate almost every stage of RNA metabolism, and among their most consequential roles is the control of alternative splicing. Studying this regulation properly means combining three different kinds of evidence: how much an RBP is expressed, how splicing changes when the RBP is depleted, and where the RBP physically binds its RNA targets. Assembling this evidence normally requires custom, code-heavy pipelines that align CLIP peaks to genomic coordinates, filter genuine alternative-splicing events from constitutive noise, and apply the right statistical test at each step, a barrier that keeps ENCODE's rich public RBP resources largely inaccessible to wet-lab researchers without computational support.

CHARM (Comprehensive Hub of Alternative Regulatory Mapping) is a no-code, browser-based R/Shiny application built to remove that barrier. It organises RBP-centric data across three modules: Expression, Splicing, and Binding; built on ENCODE shRNA-knockdown RNA-seq, ENCODE eCLIP binding data, and VastDB splicing annotations. The first three modules can be used in either of two modes: Explore mode lets you query this precomputed reference resource directly, while Discovery mode lets you upload your own differential expression or splicing results and compare them against it.

CHARM is deliberately a hypothesis-generating tool rather than a source of validated mechanistic claims. Every relationship it surfaces , be it an expression-independent functional change, or a candidate indirect regulatory interaction between two RBPs, is a candidate for downstream experimental confirmation, not a substitute for it. Two published analyses illustrate what this can reveal. First, because CHARM assesses RBP activity through binding rather than through expression alone, it can detect functional alterations invisible to expression-only analysis. Sequestration of TIA1 into stress granules produced splicing changes resembling those of TIA1 knockdown, even though TIA1 expression itself was unchanged. Second, CHARM can flag indirect regulatory relationships: knockdown of RBM39 altered the binding of U2AF1 and U2AF2, suggesting RBM39 functions within, or in cooperation with, this splicing complex.

Good to know: This tutorial walks through all four tabs of the app (Home, Expression, Splicing, Binding), then covers how to read each plot type CHARM produces (Section 4).

2. Getting Started

Accessing CHARM

CHARM runs entirely in your web browser. Open the app here:
https://diseasetranscriptomicslab.github.io/CHARM/

Please be aware that the interface was not designed for mobile screens, so a desktop or laptop display is best for reading the plots and tables comfortably.

Good to know: CHARM's initialisation, (as well as the first search after switching cell lines) can take a few seconds while the app decompresses the relevant reference dataset into memory (at maximum 1 minute). This is expected and is sadly a limitation of reading the large amounts of data CHARM hosts.

The Home Tab

Purpose: Orient yourself and jump straight to the module you need.

CHARM Home tab, showing the logo, the three module cards (Binding, Splicing, Expression) and the Explore and Discovery modes
The CHARM Home tab: three module cards (Binding, Splicing, Expression), followed by the two usage modes, Explore and Discovery.

The Home tab greets you with the CHARM logo, a short introduction to what CHARM does, and three large, clickable module cards, one for each layer of evidence:

  • Binding — takes you to the Binding tab, to see where an RBP binds around splicing events and what happens to that binding when an RBP is knocked down (eCLIP-based RNA splicing maps).
  • Splicing — takes you to the Splicing tab, to see which splicing events (exon skipping and intron retention) change after an RBP knockdown.
  • Expression — takes you to the Expression tab, to see which genes and Hallmark gene sets change after an RBP knockdown.

Below the module cards, Explore and Discovery are shown as the two ways of using every module (they are sub-modes of each tab, not separate tabs):

  • Explore — investigate how a known RBP affects expression, splicing, and binding, using ENCODE's gene-silencing series and eCLIP data. This is the default mode of every tab. Clicking its card opens the Expression tab in Explore mode.
  • Discovery — upload your own expression, splicing, or binding data to discover which RBPs are most likely altered in your biological system. Clicking its card opens the Expression tab in Discovery mode.

You can also reach any tab directly at any time using the tab bar at the top of the page (Home, Expression, Splicing, Binding), and switch between Explore (default) and Discovery with “Select mode” in the sidebar of each tab.

Glossary of Key Terms

CHARM's interface uses a small set of recurring terms drawn from RNA biology and statistics. This glossary is meant as a quick reference list.

TermMeaning
Explore vs. Discovery modeExplore mode queries CHARM's precomputed reference resource (ENCODE knockdowns). Discovery mode compares a file you upload against that same reference resource.
ENCODE knockdownsShort-hairpin RNA or CRISPR system used by ENCODE to knockdown a specific RBP, so its effect on expression, splicing, and binding of other factors can be measured against a control.
eCLIPEnhanced Crosslinking and Immunoprecipitation. The assay ENCODE used to map genome-wide, where an RBP physically binds RNA.
eCLIPSECHARM's in-house pipeline (eCLIP for Splicing Evaluation) that aligns eCLIP binding peaks to standardised positions around splicing events, producing the RNA binding maps shown in the Binding tab.
PSI / dPSIProportion Spliced-In the proportion of transcripts that include a given alternative exon or intron. dPSI is the change in PSI between two conditions (e.g. knockdown vs. control); positive values mean more inclusion.
VastDB event IDA standardised identifier for a specific alternative splicing event, from the VastDB database.
betASThe R package/web app CHARM used to facilitate differential splicing analysis; Discovery-mode splicing and binding uploads are more easily formatted as betAS output.
Cohen's dA standardised effect size describing how far apart two groups are (e.g. knockdown vs. control), independent of sample size. Larger absolute values mean a bigger, more consistent difference.
Increased / Decreased / Maintained eventsThe three categories CHARM's Binding tab splits a knockdown's splicing events into, using a dPSI threshold: Increased and Decreased events changed inclusion by more than the threshold, in each direction; Maintained events are the unaffected background.
Chi-squared differential-binding statisticThe statistic CHARM's Binding tab plots at each position of a splicing map: whether an RBP's eCLIP peaks are more or less frequent at Increased events than at Maintained events, and separately at Decreased events than at Maintained events. This gives two independent signed tracks; a positive peak means enrichment relative to Maintained, negative means depletion (see Section 5).
GSEA / NESGene Set Enrichment Analysis and its Normalised Enrichment Score: whether a MSigDB Hallmark gene set is coordinately shifted, up or down, among genes ranked by differential expression.
Similar RBPs / Similar ProfilesA CHARM feature (in the Expression, Splicing, and Binding tabs) that ranks other RBP knockdowns, or RBP-knockdown/target-RBP binding profiles, by how similar their signature is to a chosen RBP or query profile. Expression and Splicing rank by Spearman correlation of t-statistics or NES; Binding ranks by Pearson correlation of the positional binding-density profiles.

3. Interface Walkthrough

This section describes every tab. All tabs share the same overall aesthetics, with a left-hand sidebar for mode and search controls, and a right-hand results area where the plots appear.

3.1 Expression Tab

Purpose: Investigate how a chosen RBP's knockdown affects gene expression and pathway enrichment, or find which RBP knockdowns most affect a gene or pathway of interest.

Why use this: Expression is the most direct readout of a gene-silencing experiment.

Explore Mode

  1. In the sidebar, leave “Select mode” on Explore (the default).
  2. Choose a dataset scope from “Select option”: Both Cells, K562, HEPG2, or Similar RBPs. The first three options relate to which cells you will use for the knockdown vs. control comparison. The latter, is a separate sub-mode.
  3. Search by RBP: type an RBP symbol, click Search. This shows shRNA knockdown-efficacy violin plots, a differential gene expression (DGE) volcano plot with its results table, and a Hallmark GSEA barplot with its results table, all for that RBP's knockdown vs. control.
  4. Search by Gene (reverse search): type a gene symbol to see, across every profiled RBP knockdown, which knockdown most changes that gene's expression.
  5. Search by Hallmark Gene Set (reverse search): type a Hallmark gene set name to see which RBP knockdown most shifts that gene set's enrichment (NES).
  6. Similar RBPs: set “Select option” to Similar RBPs, choose whether to correlate by gene expression or by GSEA enrichment. You can chose to obtain a scatter of your selected RBP with another RBP of interest. Alternatively, you can obtain a bar plot of selected RBPs, or restrict to the top RBPs whose knockdowns lead to most similar or different effect, or top overall (module). All of these correlations compare t-statistics of differential expression, or alternatively, if the GSEA option is selected, NES.
Good to know: Gene-set window (step 3). Click a bar of the GSEA barplot (or select a row of the GSEA table) to open a window for that gene set. The top panel shows the NES of that gene set across all RBP knockdowns, with your RBP marked and its percentile; the bottom panel shows the NES obtained for random gene sets of the same size and the resulting false-positive rate (FPR). Both backgrounds are precomputed and are the same for every RBP.
Good to know: Significance of similarities (step 6). Every bar is labelled with its correlation, its percentile among all RBP pairs (e.g. “top 0.8%”) and a permutation p-value (the false-positive rate, FPR). Below the bars, a density plot shows the observed distribution of all RBP-pair correlations, the permutation background, and one vertical line per displayed RBP. See “Percentile and permutation background” in Section 4 for how to read them.
CHARM Expression tab in Explore mode, showing the sidebar search controls and the resulting knockdown-efficacy violin plots, differential expression volcano plot, and results table
The Expression tab in Explore mode: search by RBP, gene, or Hallmark gene set to see knockdown-efficacy, differential expression, and GSEA results.
CHARM Expression tab, Similar RBPs sub-mode, showing bar plots ranking RBP knockdowns by correlation of their expression signature with a chosen RBP
The Similar RBPs sub-mode: rank other RBP knockdowns by how closely their expression signature correlates with a chosen RBP's.

Discovery Mode

  1. Switch “Select mode” to Discovery.
  2. Upload a .txt, .csv, or .tsv file with a header row: the first column must contain HGNC gene symbols, and the second column the corresponding differential-expression t-statistics (e.g. from a limma–voom analysis of your own data). Alternatively, any "differential expression metric" should suffice, but bear in mind these will always be compared with t-statistics. Once the file is accepted, CHARM runs a live GSEA on your ranked metric and shows how your data's expression signature and pathway enrichment correlate with each reference RBP knockdown, letting you ask “which RBP knockdown most resembles my experiment?”
  3. Load example data: not ready to upload a file? Click “Load example data” under the upload box to load a stress-response (U2-OS) differential-expression table. It is processed exactly like an uploaded file.
  4. Users can then choose to compare their uploaded data with expression (t-statistics) or GSEA (NES). Like in the Similar RBPs sub-mode, users can chose to obtain a Scatter, or a bar plot, and the top N RBPs that (anti)-correlate with their data. Percentile and permutation p-values (FPR) are shown here too, relative to the same backgrounds.
CHARM Expression tab in Discovery mode, showing an uploaded differential expression file, the detected gene/t-statistic table, and the correlation-type options
The Expression tab in Discovery mode: upload your own differential expression file and compare it against every reference RBP knockdown.

3.2 Splicing Tab

Purpose: Investigate how a chosen RBP's knockdown affects alternative splicing, or find which RBP knockdown most affects a specific splicing event.

Explore Mode

  1. Leave “Select mode” on Explore, and choose a dataset scope (Both Cells, K562, HEPG2, or Similar RBPs).
  2. Search by RBP: type an RBP symbol and click Search to see dPSI violin plots (exon skipping and intron retention, separately) and a differential splicing analysis (DSA) volcano plot with its results table. The table has an rMATS ID column and hovering over a point in the volcano plot also shows the rMATS ID(s) of the event (see the note below).
  3. Search by Event ID: enter a VastDB event identifier to see, across every profiled RBP knockdown, which knockdown most changes that specific event's inclusion. You can also type an rMATS event ID (for example SE:37285 for exon skipping or RI:1234 for intron retention): CHARM opens the VAST-TOOLS event it is anchored on and indicates this in the plot subtitle.
  4. Similar RBPs: correlate a chosen RBP's splicing signature against all others to find RBPs with similar splicing footprints. As in Expression, each correlation is shown with its percentile and permutation p-value (FPR), together with the density plot of the backgrounds.
About rMATS IDs: CHARM quantifies splicing with VAST-TOOLS, so every event is identified by a VastDB ID. The rMATS IDs shown (exon skipping and intron retention only) are not measured: they were assigned by matching splice-site coordinates (hg38, ±1 nt) between VastDB and the rMATS annotation. The match is based on the alternative exon (or intron); rMATS may pair it with different neighbouring exons. One VAST-TOOLS event can match several rMATS events and some have none (“no rMATS match”). dPSI values and p-values always come from VAST-TOOLS.
CHARM Splicing tab in Explore mode, showing dPSI violin plots for exon skipping and intron retention, a differential splicing volcano plot, and the knockdown-efficacy plot
The Splicing tab in Explore mode: search by RBP or event ID to see dPSI violin plots, the differential splicing volcano plot, and knockdown efficacy.

Discovery Mode

  1. Switch “Select mode” to Discovery.
  2. For ease of use, we recommend uploading a differential-splicing results table exported directly from the betAS app. Alternatively, please upload a table of VastDB events with their respective dPSI values.
  3. CHARM then shows an initial plot of your data and lets you compare it against the reference resource's RBP knockdowns to identify which most closely resembles your splicing signature; or how a certain RBP's knockdown correlates with your splicing data.
  4. Load example data: click “Load example data” under the upload box to try the module with an SF3B1-knockdown splicing table. It behaves exactly like an uploaded file.
CHARM Splicing tab in Discovery mode, showing an uploaded betAS/VastDB splicing table, its initial dPSI violin plot, and a correlation scatter against a chosen RBP knockdown
The Splicing tab in Discovery mode: upload your own betAS or VastDB splicing table and correlate it against the reference RBP knockdowns.

3.3 Binding Tab

Purpose: Visualise where a chosen RBP binds relative to alternative splicing events, and whether one RBP's knockdown alters another RBP's binding. CHARM's most distinctive capability.

Why use this: Because CHARM also measures activity through binding rather than inferring it from expression, it can detect functional changes that expression analysis alone would miss entirely.
How the categories work: Every binding map starts from one knockdown's splicing results, split into three categories using a dPSI threshold: Increased and Decreased events are those whose inclusion moved by more than that threshold, in each direction, upon knockdown; Maintained events are everything below threshold and serve as the unaffected background. At every position along the metagenomic exon–intron–exon window, CHARM runs a chi-squared test asking whether the target RBP's eCLIP peaks are more or less frequent at Increased events than at Maintained events, and, separately, at Decreased events than at Maintained events. That produces two independent signed tracks (red for Increased, blue for Decreased, plotted under the raw binding densities): a positive peak means the target is enriched at that position among events that changed in that direction; a negative value means it is depleted there, relative to the unaffected background.

Explore Mode

  1. Choose an Event Type: Exon Skipping or Intron Retention.
  2. Leave “Select mode” on Explore, and choose a dataset scope (Both Cells, K562, or HEPG2).
  3. Search by RBP: enter the knockdown RBP and click Search.
  4. eCLIPSE Binding Map: choose one or more target RBPs (whose binding you want to inspect under that knockdown). Selecting one target renders a full RNA binding map; selecting several renders a heatmap overview across all of them.
  5. dPSI: choose the threshold used to define the Increased/Decreased/Maintained categories described above. Besides the fixed 0.05 and 0.1 options, CHARM also offers thresholds chosen automatically for the searched knockdown: the one that maximises the chi-squared statistic, or the one that maximises the number of FDR-significant positions.
  6. Metric: choose what the bottom track plots. FDR shows the signed, Benjamini–Hochberg-adjusted statistical significance of enrichment or depletion at each position; Effect Size shows the underlying signed chi-squared statistic directly.
CHARM Binding tab in Explore mode, showing the exon-intron-exon schematic, the normalised binding density profile with Increased/Decreased/Maintained event counts, and the chi-squared differential-binding track
The Binding tab in Explore mode: normalised binding density (top) and the signed chi-squared differential-binding track (bottom), split into Increased and Decreased tracks measured against Maintained events.

Similar Profiles

Set “Select option” to Similar Profiles to rank every RBP-knockdown / target-RBP combination in CHARM by how closely its binding profile's shape matches a chosen query pair (written as shRNA RBP__Target RBP), separately for the Increased-event and Decreased-event tracks. Similarity here is the Pearson correlation between the two profiles' positional density curves: close to +1 means the profiles rise and fall together across the map (the same shape), close to −1 means they are close to mirror images of one another, and near 0 means unrelated. You can restrict the comparison to specific profiles, or ask for the top N most similar and/or most dissimilar instead of the full ranking.

Good to know: Where does the profile fall? Under each bar plot, a density plot places your query among reference distributions: (1) all profile pairs, (2) profiles of the same eCLIP target under different knockdowns, and (3) the target’s own profile (Target__Target) against the other profiles of that target. The vertical line is the correlation of your query with that self profile, and its percentile is given in the legend. A third plot per cell line pools the Increased and Decreased profiles together (every pair of profiles counts, including Increased vs. Decreased) and draws one line per direction. No permutation p-value (FPR) is given for binding: neighbouring positions along a profile are not independent, so a shuffling background would be misleading. Use the percentiles as a guide to how unusual the similarity is.
CHARM Binding tab, Similar Profiles sub-mode, showing bar plots ranking RBP-knockdown/target-RBP profile pairs by Pearson correlation with a chosen query profile, for Increased and Decreased events across cell lines
The Similar Profiles sub-mode: rank other RBP-knockdown/target-RBP binding profiles by Pearson correlation with a chosen query profile.

Discovery Mode

  1. Switch “Select mode” to Discovery.
  2. For ease of use we suggest uploading a differential-splicing table obtained directly from betAS, exactly as in the Splicing tab. Alternatively, please upload a table of VastDB events with their respective dPSI values.
  3. Choose a target RBP (or “All” for a heatmap overview), a dPSI threshold, and a metric, exactly as in Explore mode. CHARM builds RNA binding maps from your uploaded data's Increased/Decreased/Maintained categories, letting you identify which RBPs' binding patterns are most altered in your own perturbation.
  4. Load example data: click “Load example data” under the upload box to try the module with the AllwSG example table. It behaves exactly like an uploaded file.
CHARM Binding tab in Discovery mode, showing an uploaded splicing table and heatmaps of binding density for all RBPs across Decreased and Increased events
The Binding tab in Discovery mode: build RNA binding maps directly from your own uploaded splicing categories.

4. Interpreting Results

This section is a reference for reading each plot type CHARM produces, independent of which tab generated it.

Violin plots (Expression, Splicing)

Each violin shows the distribution of a value (log2 CPM for Expression, dPSI for Splicing) across samples in one group. In Expression, comparing the knockdown violin to the control violin lets you confirm the shRNA actually reduced the target's expression before trusting any downstream result. The log2 fold-change and Cohen's d shown above the plot summarise the separation between groups: Cohen's d near 0 indicates little separation, around 0.5 a moderate effect, and 0.8 or above a large, consistent effect.

Volcano plots (Expression DGE; Splicing DSA)

Each point is one gene (Expression) or one splicing event (Splicing). The x-axis is the magnitude of change (log2 fold-change, or dPSI); the y-axis is statistical support (the moderated B-statistic for Expression, or Pdiff for Splicing). Points further from the plot's centre and higher up are both larger in effect and more statistically supported. The accompanying results table lets you sort and export these values directly.

GSEA barplots and NES

Genes are ranked by their differential-expression statistic, then tested for whether a gene set is concentrated toward one end of that ranking. A positive Normalised Enrichment Score (NES) means the gene set is enriched among up-regulated genes; a negative NES means it is enriched among down-regulated genes. The accompanying table reports NES together with a Benjamini–Hochberg-adjusted p-value for each gene set.

RNA binding maps (Binding tab)

Each map has two aligned panels along the same x-axis, representing standardised positions across the splicing-regulatory region flanking an exon or retained intron (vertical lines mark the boundaries between the different regions of interest).

  • Top panel: normalised binding density of the selected RBP, shown separately for splicing events whose inclusion decreased (blue), increased (red), or was unchanged (black) upon the knockdown you searched.
  • Bottom panel: the chi-squared differential-binding statistic at each position, comparing decreased- or increased-inclusion events against unchanged events. Peaks indicate positions where the RBP's binding differs most between event categories.

How to interpret this:

A peak in the bottom track at a position where the RBP is also enriched in the top track (for the matching colour) is the strongest evidence that the RBP directly regulates splicing at that position.

Percentile and permutation background (Similar RBPs and Discovery)

Whenever CHARM correlates two knockdown profiles (or your data with a knockdown), it reports two numbers next to the correlation, and draws them in the density plot below the bars:

  • Percentile (e.g. “top 0.8%” or “bottom 2%”): the share of all RBP-pair correlations in the resource that are at least as extreme, in the same direction. It answers “how unusual is this similarity compared with what other RBP pairs show?”.
  • Permutation p-value (false-positive rate, FPR): the share of correlations obtained from a shuffled background that are at least as extreme, in the same direction. It answers “how often would a correlation this large arise by chance?”. For Expression and Splicing, the background is built by shuffling the values of every gene (or event) across the RBPs in which it was measured, then recomputing every RBP-pair correlation. For GSEA, it is built from random gene sets of the same size as each Hallmark set, scored on every RBP’s real ranking. When no shuffled value reaches your correlation, the p-value is shown as “< x”, the smallest value the background can resolve.
  • Density plot: the grey curve is the permutation background, the purple curve is the observed distribution of all RBP pairs, and each vertical line is one displayed correlation (red positive, blue negative). Each curve is scaled to its own peak, so compare positions, not heights.

How to interpret this:

Correlations between knockdown profiles can be spurious: two profiles may correlate for reasons that have nothing to do with a shared biological mechanism, such as common control samples, batch or cell-line effects, a general stress or toxicity response, or a handful of very strongly changed genes. The two numbers help you judge this in different ways:

  • The permutation p-value (FPR) only tests whether the correlation is larger than what random, unrelated profiles of the same size would give.
  • The percentile compares the correlation with what other RBP pairs in the resource actually show, including whatever technical structure they share. It is the more demanding reference: a correlation in the top 1% is unusual even among real knockdown pairs. Prefer it when ranking candidates.

Gene-set window (Expression, GSEA)

The window opened by clicking a gene set has two panels sharing one x-axis (NES). Top: the density of the NES of that gene set across all RBP knockdowns in the chosen cell line, with small ticks for individual RBPs, the selected RBP marked with a vertical line, and its percentile in the subtitle. Bottom: the density of NES obtained for random gene sets of the same size scored on all RBP rankings, with the selected RBP’s NES marked and the FPR in the title: the share of random gene sets with an NES at least as extreme, in the same direction. The random background is the same for every RBP, so it is computed once and stored.

Binding profile backgrounds

For Binding → Similar Profiles, the density plot compares your query with all profile pairs, with profiles of the same eCLIP target, and with the target’s own profile. Only percentiles are given, not a permutation p-value (see Section 3.3).

5. FAQ / Troubleshooting

My Discovery-mode file upload was rejected or produced a warning.

Check the required format for the specific tab:

  • Expression: a .txt/.csv/.tsv file with a header row; first column HGNC gene symbols, second column differential-expression t-statistics.
  • Splicing and Binding: for ease of use, a table exported directly from betAS, uploaded unmodified. Alternatively, a table with VastDB event identifiers and corresponding dPSI values.

Why does an RBP's own knockdown sometimes not enrich its own binding?

Absence of a clean binding enrichment at the searched RBP's own knockdown does not necessarily mean the map is wrong. Some RBPs regulate splicing indirectly (through a partner or complex member) rather than through direct, local binding at the affected exon. This is precisely the kind of case the Binding tab's cross-RBP search is designed to help you investigate further.

How should I interpret a result CHARM shows me?

As a hypothesis, not a conclusion. Every relationship CHARM surfaces, whether an expression-independent functional change or a candidate indirect regulatory interaction, is derived from correlative binding and splicing data. A result that agrees with your own experiment is encouraging but still requires case-by-case biological interpretation, and the absence of a correlation does not, by itself, rule out a real relationship. Use CHARM to prioritise what to test next, not as a final answer.

What do “top x%” and the p-value next to each bar mean?

“Top x%” (or “bottom x%” for negative correlations) is the percentile of the correlation among all RBP-pair correlations in the resource. The p-value is the permutation-based false-positive rate (FPR): the share of correlations from a shuffled background that are at least as extreme. See “Percentile and permutation background” in Section 4.

Why is there no p-value for Binding → Similar Profiles?

Binding profiles are continuous curves over hundreds of neighbouring positions that are strongly dependent on each other. Shuffling values would give an unrealistically narrow background, so CHARM shows percentiles against the real distributions (all pairs, same target, self profile) instead.

Why does the Binding tab's Similar Profiles use Pearson correlation, when Expression and Splicing use Spearman?

Because the two comparisons are fundamentally different kinds of data. Expression's and Splicing's Similar RBPs compare t-statistics or NES values across a set of genes or gene sets, an unordered collection that can include extreme outliers, where only the relative ranking between RBP knockdowns is meant to matter; Spearman correlation is the more robust choice there. Binding's Similar Profiles instead compares two continuous, smoothed binding-density curves across hundreds of ordered genomic positions, where the biologically meaningful question is whether the two curves rise and fall together, position by position, which is a genuinely linear notion of shape that Pearson correlation captures directly.

Why are the dPSI thresholds that maximise the χ² statistic called "oddsrat"?

The name "oddsrat" is a legacy from an earlier version of CHARM. Originally, CHARM used Fisher's exact test to quantify differential binding, and the resulting effect size was an Odds Ratio. Later in the development of CHARM, we replaced Fisher's test with the χ² test, which we considered more appropriate for quantifying differential binding. The thresholds are now selected by maximising the χ² statistic, rather than an Odds Ratio. However, many downstream functions and variables had already been built around the name "oddsrat", so we kept the original name for compatibility. In other words, despite the name, "oddsrat" now refers to the χ² statistic used to select the dPSI threshold, not to an Odds Ratio.

Why aren't alternative 5′/3′ splice-site (Alt5/Alt3) events available?

CHARM's splicing analyses currently cover exon-skipping and intron-retention events only; alternative splice-site usage is not yet supported. This is a known limitation described in the CHARM paper's Discussion. Extending the eCLIPSE splicing-genome framework to these event types is a natural direction for future development.

Is iCLIP or iCLIP2 binding data planned?

It's an intended future extension, though not yet implemented. CHARM's binding data currently comes exclusively from ENCODE eCLIP. As the CHARM paper's Discussion notes, iCLIP2 offers higher spatial resolution of RBP-binding sites, and integrating a sufficiently large iCLIP2 dataset, should one become available, is described as a natural extension of the eCLIPSE pipeline — CHARM's architecture is deliberately built so that additional binding assays can be incorporated without changing the interface, but no timeline exists yet.

I'd rather not use VAST-TOOLS. Can I use rMATS or MAJIQ instead?

  • rMATS: yes. rMATS reports events in a fixed format that turned out to be structurally compatible with VastDB's event-based approach, so a corresponding rMATS-compatible splicing genome already exists — see the eCLIPSE Local Pipeline tutorial. This route runs through the local pipeline (for computational users), not through the Shiny app's Discovery-mode upload, which still expects a betAS- or VastDB-formatted dPSI table.
  • Seeing rMATS IDs in the app: the Splicing tab shows rMATS IDs for exon-skipping and intron-retention events next to the VAST-TOOLS IDs, and you can search an event by its rMATS ID. These IDs are assigned by matching splice-site coordinates (±1 nt), not measured, and dPSI and statistics are always from VAST-TOOLS.
  • MAJIQ: not currently. MAJIQ's event definitions and coordinate system differ enough from VastDB's that they aren't structurally compatible the way rMATS's are, so no MAJIQ-CHARM-compatible splicing genome exists yet.

Who do I contact if something looks broken, or if I have a request?

Please contact the lead author at alexandre.afonso@gimm.pt. Alternatively, please open an issue on CHARM's GitHub repository describing it.