RDDpred

RDDpred predicts condition-specific RNA-editing events from RNA-seq data using a Random Forest classifier to distinguish true edits from sequencing and mapping artefacts.

Key Features:

• RNA-seq input: Operates on RNA-seq–derived candidate sites to detect RNA-editing events.
• Random Forest classifier: Uses a Random Forest–based classifier to separate true RNA-editing events from false positives and artefacts.
• Automatic training example compilation: Compiles condition-specific training examples automatically without requiring experimental validation.
• Validation on public datasets: Reproduced reported RNA-editing sites with accuracies of 90% and 95% on two publicly available RNA-editing datasets.
• Negative predictive value (NPV): Achieved NPVs of 75% and 84% on the two validation datasets.
• Biochemical specificity: Accounts for canonical RNA-editing biochemistry including adenosine (A) to inosine (I) edits mediated by ADAR and cytidine (C) to uridine (U) edits mediated by APOBEC.
• False discovery reduction: Aims to reduce false discoveries inherent to RNA-seq–based detection of RNA-editing.
• Reference database statistic: References the public RDD database DARNED reporting over 300,000 editing sites in the human genome (hg19).
• Condition specificity statistic: Incorporates the observation that 97.62% of registered editing sites are detected in single tissues or specific conditions.
• Biological impact context: Enables investigation of RNA-editing impacts on protein activity modulation, alternative splicing of mRNA, and miRNA target site substitutions.

Scientific Applications:

• Condition-specific RNA-editing detection: Identify RNA-editing events that are specific to tissues or experimental conditions from RNA-seq data.
• Transcriptome-level studies: Support investigations of RNA-editing contributions to protein function, alternative splicing, and miRNA targeting.
• False-positive filtering: Reduce artefactual calls in RNA-editing discovery workflows derived from RNA-seq.
• Validation and reproducibility assessment: Reproduce and assess reported RNA-editing sites from public RNA-editing datasets.

Methodology:

Uses a Random Forest classifier trained on automatically compiled condition-specific examples from RNA-seq candidate sites to classify true RNA-editing events versus artefacts, with validation reported on two public datasets (accuracies 90% and 95%; NPVs 75% and 84%).