How Does rRNA-Depleted RNA-Seq Differ from Conventional RNA-Seq?
RNA-Seq libraries can be prepared using rRNA depletion in addition to poly(A) selection.
Poly(A) selection mainly captures mature mRNAs with poly(A) tails. In contrast, rRNA depletion removes rRNA, which accounts for a large proportion of total RNA, while retaining a broader range of other RNA molecules.
As a result, rRNA-depleted RNA-Seq data may contain not only mRNA, but also lncRNA, antisense RNA, non-poly(A) RNA, pre-mRNA, and intron-derived RNA.
A major feature of rRNA-depleted RNA-Seq data is that they may retain information about a broader range of RNA species than conventional mRNA expression analysis.
Older Analyses May Not Have Fully Used the Information in the Data
In the mid-2010s, a common RNA-Seq analysis workflow involved mapping reads to a reference genome using a tool such as STAR, counting genes registered in a GTF file, and performing differential expression analysis with a method such as DESeq2. This was a standard approach at the time and is still widely used for analyzing known genes.
However, the GTF files available at that time did not contain as many lncRNAs, antisense RNAs, or transcript isoforms as current annotations. Therefore, even when reads corresponding to such RNAs were present in the FASTQ files, RNAs not registered in the GTF file did not appear in the gene count table.
Mapping reads to the genome and counting them as gene-level expression values are not the same thing. The annotations and analysis environments available at the time may not have allowed the information obtained through rRNA depletion to be used fully.
Some RNAs Cannot Be Found Using GEO-Provided Gene Counts Alone
For some RNA-Seq datasets, GEO provides gene counts generated through a standardized processing workflow.
These counts are useful for differential expression analysis of known genes, but they are generally created using a predefined gene annotation. Therefore, unannotated antisense transcripts, novel lncRNAs, and new isoforms may also be absent from GEO-provided gene counts. The value of reanalyzing such data does not lie only in repeating statistical analysis using the existing gene counts.
The greater value lies in returning to the FASTQ files and using current annotations and transcript assembly to recover RNAs that were not previously counted.
rRNA-Depleted RNA-Seq Data Can Be Particularly Valuable for Reanalysis
rRNA depletion retains a broader range of RNA species than poly(A) selection. Therefore, even for data generated many years ago, publicly available FASTQ files may contain information that can now be recovered using current annotations and analysis methods.
Possible reanalyses include:
- Requantifying lncRNAs and antisense RNAs included in current annotations
- Searching for new transcript isoforms
- Detecting transcripts in intergenic regions or within introns
- Examining the positional relationships between novel transcripts and known genes
Novel Transcripts Can Now Be Included in the Reanalysis
Current GTF files include more lncRNAs and antisense RNAs than older annotations. However, rRNA-depleted RNA-Seq data may still contain novel ncRNAs and transcripts that are not registered even in current GTF files.
Therefore, simply recounting known transcripts using the latest GTF may still fail to make full use of the information contained in rRNA-depleted RNA-Seq data.
To search for novel transcripts, reads are aligned to the genome using STAR or a similar tool, and transcript structures are reconstructed using StringTie or a similar tool. A new GTF is then created by integrating known transcripts with high-confidence novel transcripts, and transcriptome sequences are prepared based on that GTF.
The integrated transcriptome is then used as the reference to requantify all Samples using Salmon or a similar tool. This makes it possible to compare known RNAs and newly identified transcripts within the same expression table.
FASTQ
↓
QC
↓
STAR
↓
StringTie
↓
Classification of known and novel transcripts
↓
Removal of low-confidence candidates
↓
Annotation of genomic position, nearby genes, coding potential, and other features
↓
Creation of a new GTF/transcriptome integrating
known transcripts and high-confidence novel transcripts
↓
Requantification of all Samples with Salmon
using the integrated transcriptome
↓
Differential expression analysis and visualization
In this workflow, genome alignment with STAR or a similar tool provides the basis for reconstructing novel transcripts, whereas Salmon or a similar tool is used to quantify the transcriptome after it has been constructed.
Confirm Strandedness Before Analyzing Antisense RNA
To distinguish antisense RNA from sense RNA, the original RNA-Seq library must retain strand information.
Even when a paper does not explicitly state that a strand-specific library was used, files such as STAR's ReadsPerGene.out.tab may show whether counts are strongly biased toward the forward or reverse orientation.
However, strandedness should be confirmed again from the FASTQ files when performing the reanalysis.
If the library is unstranded, sense and antisense RNAs overlapping the same genomic region cannot be accurately quantified separately.
Before searching for antisense transcripts, the following points should be confirmed:
- Whether the library is stranded or unstranded
- Whether it is forward-stranded or reverse-stranded
- Whether the strand settings in the analysis tools match the library type
Turning Novel Transcripts into Interpretable Information
When novel transcripts are assembled using StringTie or a similar tool, they are assigned identifiers such as MSTRG.1234.1, which provide little biological information by themselves.
The original identifiers should therefore be retained while adding annotations such as:
- Genomic position
- Strand
- Exon count and transcript length
- Overlap with known genes
- Sense or antisense relationship
- Classification as intergenic, intronic, novel isoform, or another category
- Nearest known gene
- Distance to the nearby gene
- Protein-coding potential
- Expression level and differential expression across Samples
Organizing the results in this way makes it possible to interpret previously uninformative transcript identifiers based on their positional relationships with known genes and their expression patterns.
Considerations When Analyzing the Data in Subio Platform
A standard “Platform” used by Subio Platform is based on a set of identifiers from public databases and the corresponding gene annotations.
In contrast, MSTRG identifiers generated through transcript assembly are defined for that particular analysis and are not registered in public databases. Therefore, data containing such novel transcripts should not be added to a standard “Platform.” Instead, they need to be incorporated into a dedicated Platform created for that Series.
However, this dedicated “Platform” is based on a transcriptome constructed from the group of Samples used when the Platform was created. It is therefore generally specific to that Series. Because the structures and identifiers of novel transcripts may change depending on the Samples included in the analysis and their read coverage, a new Sample cannot simply be added later to the existing “Platform.”
When reanalysis needs to include additional Samples, all existing and new Samples must be processed together again starting from the FASTQ files, and a separate new “Platform” must be created.
Building the Pipeline with the Help of AI
In the past, performing this type of analysis required researchers to examine the manuals, commands, and input and output formats of each tool individually. Today, AI can be used to help design a pipeline suited to the purpose of the analysis.
For example, AI can help answer questions such as:
- How STAR and Salmon should be used for different purposes
- How to confirm strandedness
- Which reference genome and GTF file should be used
- How to assemble and classify novel transcripts
- How to remove low-confidence candidates
- How to add nearby-gene information and other annotations to novel IDs
- How to create a GTF/transcriptome that integrates newly identified transcripts with known transcripts
- How to requantify the constructed transcriptome using Salmon
It is no longer necessary to memorize every command in advance. The analysis can be built step by step by telling AI the objective, input files, and expected output.
However, procedures proposed by AI should not be executed without verification. The following points still need to be checked:
- Library preparation method
- Whether the reads are paired-end or single-end
- Strandedness
- Compatibility between the reference genome and GTF file
- Whether the objective is to quantify known transcripts or discover novel transcripts
- Reliability of the novel transcript candidates
- What the output of each tool actually represents
AI does not automatically make an analysis correct.
However, it has made it much easier than before to select the necessary processing steps and translate them into an executable analysis pipeline.
Conclusion
RNA-Seq data generated using rRNA depletion may contain information about not only mRNA, but also lncRNA, antisense RNA, non-poly(A) RNA, pre-mRNA, and other RNA species.
However, older analyses and standardized gene counts may not fully reflect this information because they rely on the GTF files and known-gene-centered counting methods available at the time. For such data, it may be valuable not only to reuse the existing gene counts, but also to search for novel transcripts from the FASTQ files, create a new integrated transcriptome, and requantify all Samples using that reference. This may lead to new findings.
Older rRNA-depleted RNA-Seq datasets are not necessarily less valuable simply because they were generated many years ago. They may be especially valuable for reanalysis precisely because they can still contain information that could not be identified at the time.
