SLRanger: an integrated approach for spliced leader detection and operon prediction using long RNA reads

Jinpeng Chen, Yanwen Shao, Zhihao Guo, Jinpeng Chen 0003, Runsheng LiBriefings in Bioinformatics 2025 · DOI

TL;DR

Researchers have developed SLRanger, a novel computational tool that leverages long-read RNA sequencing to accurately identify spliced leader (SL) sequences and predict operon structures in eukaryotic transcriptomes. This advancement overcomes limitations of previous methods, enabling a more comprehensive understanding of gene regulation and operon organization, particularly in organisms like nematodes.

What Problem Does This Paper Address?

Many eukaryotes, including nematodes, utilize a process called spliced leader (SL) trans-splicing, where a short RNA fragment is added to the beginning of messenger RNAs (mRNAs). This process is crucial for RNA stability, translation efficiency, and, importantly, for processing polycistronic transcripts (operons) into individual, functional mRNAs. However, accurately identifying SL trans-splicing events has been a significant challenge. Traditional short-read sequencing methods struggle to capture the full SL sequence and resolve operon structures due to fragmented reads and potential loss of the 5' end. While long-read sequencing offers a promising solution, its inherent error rates and the short, variable nature of SL sequences make detection difficult. This paper addresses the need for a robust and integrated tool that can reliably detect SL sequences and predict operon structures from long RNA reads, overcoming these technical hurdles.

Methods

The core of SLRanger lies in its two main modules: SL detection and operon prediction.

  • SL Detection: SLRanger refines the traditional Smith-Waterman (SW) alignment algorithm with an optimized scoring scheme specifically tailored for SL detection in long RNA reads. This involves:
  • Candidate Read Preprocessing: Long RNA reads are mapped to a reference genome, and only reads with unaligned soft-clipped sequences (at least 5 nucleotides) at the 5' end are considered candidate reads.
  • Optimized Scoring: Candidate sequences are aligned against known SL sequences (e.g., SL1 and SL2 variants in C. elegans) and random sequences (as controls). SLRanger's scoring system incorporates several modifications to the raw SW score, including penalizing gaps and rewarding continuous matches, proximity to the 5' end of the aligned region, and proximity to the 3' end of the reference. This results in an "SLRanger score."
    • Dynamic Cutoff: A dynamic cutoff is applied to identify "high-confidence SL reads." This cutoff is determined by comparing the number of reads aligning to SL references versus random sequences, allowing for adjustable stringency. This approach aims to balance sensitivity and specificity, effectively distinguishing genuine SL reads from background noise.
  • SL Type Classification: SLRanger can distinguish between different SL types (e.g., SL1 and SL2 variants), which is crucial for operon prediction. Reads that cannot be definitively classified are labeled as "unknown."
  • Operon Prediction: This module leverages the SL type information from high-confidence SL reads and their genomic mapping positions.
  • SL Type Quantification: The number of SL1 and SL2 reads mapped to each gene is quantified.
  • Upstream/Downstream Gene Identification: Genes with a higher proportion of SL1 reads are predicted as upstream operon genes, while those with a higher proportion of SL2 reads are considered downstream genes.
  • Operon Structure Inference: Genes are grouped based on their genomic proximity. Candidate operons are identified by looking for adjacent genes exhibiting an upstream-downstream SL pattern (SL1 followed by SL2). A minimum number of supporting reads is required to confirm an operon. The tool also accounts for "fusion reads" that span multiple genes.

Datasets

The study utilized several publicly available long-read RNA sequencing datasets for validation:

  • Direct RNA Sequencing (ONT):
    • Caenorhabditis elegans embryo dataset (Li et al.)
    • Caenorhabditis elegans datasets from various developmental stages (Roach et al.)
    • Brugia malayi dataset (a species with SL sequences)
    • Human direct RNA dataset (used as a negative control to assess false positive rates)
  • cDNA Sequencing (ONT):
    • Caenorhabditis elegans cDNA dataset (Bernard et al.)

The authors also used the C. elegans genome annotation (WBcel235 from WormBase) and its known operon annotations as a ground truth for evaluating operon prediction accuracy.

Key Insights and Ideas

  • Long reads are game-changers for SL detection: The ability to sequence full-length transcripts with long reads is critical for capturing the entire SL sequence and its context, which is nearly impossible with short reads.
  • A specialized scoring system is essential: Raw SW alignment scores are insufficient for reliably distinguishing SLs from background noise in high-error long-read data. SLRanger's custom scoring scheme significantly improves this discrimination.
  • SL2 is a strong indicator of operons: The strong association between SL2-type trans-splicing and operon structures is a key principle exploited by SLRanger for operon prediction.
  • SL trans-splicing is widespread and complex: The study highlights the prevalence of SL trans-splicing across diverse eukaryotic taxa and suggests potential stage-specific or condition-specific SL usage preferences (e.g., SL10 abundance).
  • Operon structures are more prevalent than previously thought: SLRanger's ability to predict operons, even those not yet annotated, suggests that operon organization might be more common in eukaryotes than currently recognized.

Key Results

  • Improved SL Detection Accuracy: SLRanger achieved a 1.8-2 fold higher Area Under the Curve (AUC) for distinguishing SL reads from random sequences compared to raw SW scores in C. elegans datasets.
  • Higher Confidence SL Read Recovery: SLRanger identified 7-15% more high-confidence SL reads compared to methods relying solely on SW scores.
  • Accurate SL Type Classification: SLRanger effectively distinguished between SL1 and SL2 types in C. elegans, with only 0.05% of reads being unclassifiable between SL1 and SL2.
  • Robust Operon Prediction:
    • Gene-level: SLRanger achieved 81% precision and 73% sensitivity (for the Li et al. dataset) in predicting operon genes.
  • Structure-level: 52% of predicted operon structures were fully concordant with the ground truth, demonstrating the tool's capability in reconstructing operon organization.
  • Low False Positive Rate: When applied to human datasets lacking SLs, SLRanger did not identify any SL reads, demonstrating its ability to control false positives.
  • Cross-species Applicability: SLRanger successfully identified SL sequences in Brugia malayi using the C. elegans SL1 reference.

Limitations and Future Work

  • Data Quality Dependence: The performance of SLRanger, particularly in operon prediction sensitivity, can be influenced by the quality (accuracy and length) of the input long-read data. Datasets with higher error rates or more 5' truncation may yield lower sensitivities.
  • Unannotated Features: The paper acknowledges that some unaligned 5' regions beyond SL detection might represent biologically meaningful events (e.g., alternative promoters, unannotated 5' exons) that are not fully characterized by the current tool.
  • Fine-tuning Cutoffs: Users may need to fine-tune the dynamic cutoff parameter to balance recall and precision based on their specific dataset and research goals.
  • Future Directions: The authors suggest that SLRanger's pipeline is adaptable to other long-read sequencing platforms (e.g., PacBio Iso-Seq) and could be further applied to investigate the complexity of gene regulation and operon organization in various eukaryotic genomes. The tool's ability to identify novel operons and SL-containing transcripts opens avenues for further functional and mechanistic studies.

Summary updated

BibTeX
@article{chen2025slranger,
  author = {Jinpeng Chen and Yanwen Shao and Zhihao Guo and Jinpeng Chen and Runsheng Li},
  title = {{SLRanger: an integrated approach for spliced leader detection and operon prediction using long RNA reads}},
  journal = {Briefings in Bioinformatics},
  year = {2025},
  doi = {10.1093/bib/bbaf437},
}
Figure