Structure-aware Graph Learning Predicts RNA Editability Across Tissues and Species

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Adenosine-to-Inosine (A-to-I) RNA editing, mediated by ADAR enzymes, is a crucial post-transcriptional modification with therapeutic potential.
• Predicting A-to-I editing sites is difficult due to reliance on double-stranded RNA (dsRNA) geometry and stability, not just sequence.
• Current models struggle to capture the complex structural determinants of ADAR enzyme recognition.

Purpose of the Study:

• To develop a structure-explicit computational framework for predicting A-to-I RNA editing sites.
• To improve the accuracy of predicting RNA editing by incorporating dsRNA structural features.
• To investigate conserved principles of ADAR substrate recognition across different species.

Main Methods:

• Developed A dar E dit , a graph-attention framework representing dsRNA substrates as nucleotide graphs with backbone and base-pair edges.
• Augmented the graph representation with typed interactions and a motif-sensitive sequence branch.
• Trained and evaluated the model on human inverted Alu duplexes using RNAfold-predicted secondary structures and GTEx RNA-seq data.

Main Results:

• A dar E dit consistently outperformed sequence-only CNN, transformer, and RNA language models across multiple tissue contexts.
• Achieved high discrimination performance (AUROC/AUPRC = 0.96; F1 ≈ 0.90) on combined human tissue data.
• Demonstrated successful transferability of the graph representation to evolutionarily distant non-Alu species, revealing conserved ADAR recognition principles.

Conclusions:

• A dar E dit provides a powerful, structure-aware approach for predicting A-to-I RNA editing sites.
• The findings highlight the importance of dsRNA structure in ADAR substrate recognition, with conserved principles across species.
• The model's attention profiles and mutagenesis analysis offer insights into biochemical constraints and long-range structural influences on RNA editing.