Author: Dana Nachmanson

Sr. Bioinformatics Scientist

Share this post on LinkedIn

RNA-seq analysis has become a staple of modern bioinformatics. It’s reliable, foundational, and applicable across a wide range of biological questions. But if RNA-seq is the bread and butter of transcriptomics, then RNA modification sequencing is sourdough with herbed olive oil–a familiar base, but infused with richer, more complex insights into gene regulation.

What are RNA Modifications?

RNA molecules are not just passive messengers of genetic code. They are dynamically regulated through a diverse array of chemical modifications, like N6-methyladenosine (m⁶A), inosine, pseudouridine (Ψ), and more. Each transcript can carry several of these marks, which influence how RNA is spliced, exported, translated, or degraded¹.

Despite being pervasive, some RNA modifications are dynamically regulated–written, read, and erased in a short window between transcription and translation. Studying them has been challenging, but recent advances have revealed their critical roles in development, cancer, neurodegeneration, immunity, and metabolic disease^2-4.

Why RNA Modification Data Require Specialized Analysis

Although analysis of RNA modification assays borrow elements from RNA-seq and CHIP-seq workflows, they bring unique bioinformatic requirements:

• Tools must support non-uniform, modification-specific coverage influenced by expression, strand, and transcript region
• Metrics integrating modification-specific features to support accurate calls (e.g., DRACH motifs for m⁶A, A-to-I editing events for inosine)
  
• Dedicated algorithms to detect RNA modifications through either peak calling or signal-based models
• Learning models trained to separate signal from noise, particularly in transcripts with low abundance modifications

* Common but not natively performed within example pipeline

Key Bioinformatics Challenges in RNA Modification Analysis

1. Sparse Signal and Low Abundance

Many RNA modifications–such as 5-methylcytidine (m⁵C)–are rare or conditionally expressed. Others, like inosine, are more abundant but often localized to intronic regions that are typically under-covered in typical RNA modification workflows. This makes robust spike-in controls and accurate normalization essential to distinguish low modification sites from background noise. Controls and spike-ins are not optional; they’re foundational.

2. RNA-Specific Peak Calling

Classic peak callers (like MACS2 or HOMER) were built for DNA-centric assays and assume depth uniformity, lack of strand specificity, and constant background noise  ^5-7. These assumptions break down with RNA data. Tools like exomePeak2, HEPeak, m6Aboost and m6ACali offer RNA-focused solutions but often focus solely on m⁶A and depend on training datasets which are very model-specific, limiting their broader use^8-12.

3. Differentially Expressed Genes Modified Regions

Identifying differentially modified regions is not the same as detecting differentially expressed genes. RNA modifications often cluster at 3’ UTRs or introns rather than spanning the entire gene. These features require statistical models with region-level resolution and careful control for baseline gene expression that diverge from the capabilities of DESeq2 or edgeR, which are designed for gene-level count data^13-14.

4. Multi-Modification Detection

Most tools are modification-specific, and most are developed for m⁶A. Yet real biological samples often carry multiple modifications, sometimes on the same transcript. To capture this complexity, you need assays and pipelines that can handle more than one modification at a time. 

Right now, only EpiPlex™ and Oxford Nanopore’s direct RNA sequencing generate signals for multiple RNA modifications¹⁵. But detecting them isn’t straightforward. It takes two things:

• Physical controls with known modifications to match signal patterns
• Smart models—like machine learning tools—to tell similar signals apart

Most current pipelines still focus on one modification at a time, and robust multi-modification detection across platforms is still developing.

5. Reproducibility and Tool Maintenance

Many academic tools become obsolete after publication, making it difficult to reproduce results or apply them to new datasets. Sustained, version-controlled, and documented pipelines–like nf-core and GATK–show how community-driven development and commercial support can ensure long-term usability and broad adoption. 

The need for more sophisticated, reproducible tools has also been recognized at the national level. In 2024, the National Academies released a full-length report calling for stronger informatics infrastructure to support RNA modification research¹⁶. Shared infrastructure is key to supporting growth in the RNA modification space.

Best Practices for Accurate RNA Modification Analysis

Here’s what we recommend at Alida Biosciences:

Normalize for Gene Expression

If a gene isn’t expressed, it can’t be modified. Highly expressed genes can yield false positives due to sheer abundance. Always normalize enrichment signals against baseline expression.

Use Internal Controls

Include synthetic RNA spike-ins carrying known levels of modification. These act as anchors for scaling your data and evaluating technical variation (see our blog “Epitruth or Epifiction? How to know if RNA Modification Data Can Be Trusted“]

Visualize Your Modification Calls

Load your BAM and BedGraph files into IGV or UCSC Genome Browser to confirm modification calls visually. This is especially important before investing in downstream experiments.

Use Modification-Specific Biological Context

Confirm m⁶A peaks with DRACH motif overlap, or identify inosine A-to-I editing using A-to-G mismatches. These patterns strengthen biological plausibility.

From Data to Discovery: Biological Insight from RNA Modifications

Once you’ve mapped RNA modifications, what can you actually learn?

1. Gene and Transcript-Level Insights

RNA modifications like m⁶A don’t just decorate transcripts–they drive essential cell functions and play a critical role in disease biology.

Take Huntington’s disease: Pupak et al. (2024) found that m⁶A marks in intron 1 of mutant HTT RNA trigger the production of a toxic splice variant (HTT1a)–a transcript that escapes normal processing and fuels neurodegeneration¹⁷. When researchers erased those marks, the pathogenic RNA dropped–without touching full-length HTT.

Modification patterns can reveal condition-specific regulation of transcripts, especially when a gene is known to be involved in a phenotype or process.

Pro tip: Modifications often cluster near cryptic splice sites, stop codons, or 3′ UTRs–prime real estate to regulate  gene expression.

Figure 1: Distribution of m6A and inosine across a hypothetical transcript binned into 100 parts. m6A (left) is abundant at the coding domain (CDS)/3’UTR junction. Inosine (right) has a diffuse distribution across introns (included in the CDS), 5’UTR, and 3’UTR.

2. Pathway-Level Insights

Once you’ve identified your modified transcripts, don’t stop at the gene-level, ask what they’re doing. 

Try running GSEA or GO enrichment on your modified transcript list. You might find:

• Immune signaling: Inosine editing protects endogenous dsRNA from being treated as a viral RNA, often dampening innate immunity pathways¹⁸.
• Neuroplasticity: In neurons, m⁶A and A-to-I edits cluster in genes tied to synaptic signaling and calcium transport–fine-tuning circuit dynamics without changing expression¹⁹.
• Cancer control: Modifications like m⁶A and inosine editing hit transcripts in cell cycle, stress response, and apoptosis pathways, helping tumors rewire post-transcriptional control²⁰.

Try it: Compare pathway enrichment from modification-enriched versus differentially expressed transcripts–you may uncover completely different biology.

3. Zoom Out: Epitranscriptome-Wide Patterns

Sometimes the biggest insights come from stepping back.

Global modification patterns–like overall inosine or m⁶A levels–can shift dramatically across conditions, reflecting the activity (or inactivity) of writers and readers. For instance, cancers often show elevated m⁶A, while some tumors dial down ADAR editing to evade immune detection²¹.

Use overall modification patterns as input for data dimensionality reduction tools: PCA (Principal Component Analysis) or UMAP (Uniform Manifold Approximation and Projection), and you may see samples naturally cluster by disease state, cell type, or even molecular subtype–purely from their modification “fingerprints”. 

In a recent study, Santos et al. (2024) showed that global A-to-I editing profiles could distinguish living versus post-mortem brain tissue using PCA–revealing that RNA editing patterns persist as robust markers of tissue viability²².

Figure 2: Principal component analysis (PCA) of inosine of living (left) vs post-mortem (right) brain tissue adapted from Santos et al. 2024 under a Creative Commons Attribution 4.0 International License. PCA reveals distinct RNA editing patterns correlated with tissue viability. Changes were made to the original figure.

4. Integration with Other Omics

RNA modifications can fill in the blanks between transcript levels and protein output.

We all know gene expression doesn’t always predict protein abundance. That’s where modifications come in–they influence mRNA stability, translation, and decay, offering critical insight into why a transcript is (or isn’t) making protein.

You can also treat modification features as input for machine learning models. Adding modification data to expression or splicing features can improve subtype prediction, especially in cancer or treatment response studies.

AlidaBio’s EpiScout™ Analysis Suite

To address many of these challenges, AlidaBio built the EpiScout analysis pipeline, the informatics engine behind the EpiPlex RNA Modification Detection Kits.

What EpiScout Analysis Suite Does:

• Normalizes pileup data to baseline gene expression
• Controls for technical variation using known-modification spike-ins
• Uses a custom Hidden Markov Model-based peak caller
• Detects multiple modifications in parallel (e.g. m⁶A, inosine)
• Identifies statistically significant differentially modified regions (DMRs) using a modification-centric statistical model
• Outputs rich, reproducible reports:
  
  • QC metrics
  • Peak tables and metagene plots
  • DRACH enrichment and A→G mutation analysis
  • Differential peak comparisons

Easy to Access and Use:

The EpiScout pipeline is available on DNAnexus and as a Docker image, making it easy to deploy across a variety of computing environments. It is maintained with version control and technical documentation to support transparent, scalable research.

Conclusion

RNA modifications are a critical but under explored layer of post-transcriptional regulation. As the field matures, so too must the tools that support it. Whether you’re working on neurodevelopment, cancer biology, or building disease classifiers, integrating RNA modification analysis into your workflow unlocks a deeper, more nuanced view of gene regulation.

To explore how EpiPlex™ Assays and EpiScout™ Analysis Suite can empower your research, visit our Resource Hub.

About the Author

Dana Nachmanson

Dana Nachmanson is a Senior Bioinformatics Scientist at Alida Biosciences, where she supports customers in analyzing and interpreting RNA modification data. She earned her Ph.D. in Bioinformatics and Systems Biology from UC San Diego, where she focused on sequencing method optimization for early cancer detection. With a background in error-corrected sequencing and translational genomics, particularly within biotech startup environments, Dana is passionate about connecting cutting-edge technologies with researchers to drive advancements in science.