#South Korea #Pusan National University #Busan #microRNA #circRNA

Introduction

In a significant advancement in bioinformatics, researchers from Pusan National University have developed circMAC, an innovative AI tool that focuses on mapping microRNA (miRNA) binding sites on circular RNAs (circRNAs) at nucleotide resolution. This breakthrough addresses the complex nature of circRNA–miRNA interactions, which can often be challenging to delineate precisely, thus opening up new avenues for gene regulation research.

Understanding Circular RNAs and Their Importance

Circular RNAs are a unique class of RNA molecules that take on a looped formation, and play a crucial role in regulating gene expression through their interactions with miRNAs. Unlike linear RNAs, circRNAs exist in multiple isoforms, each possessing distinct functions and potential binding sites for miRNAs. These binding interactions are critical as they can influence gene activity and cellular processes, making accurate mapping essential for understanding gene regulation mechanisms. However, treating these circular molecules as if they were linear can lead to inefficient binding site predictions, primarily due to neglecting important sequence context near the back-splice junction (BSJ).

Introducing circMAC

CircMAC aims to enhance the accuracy of mapping miRNA interactions with circRNAs. Developed under the guidance of Professor Giltae Song and his team from the School of Computer Science and Engineering, the tool takes full-length circRNA isoforms and mature miRNAs as inputs, providing a comprehensive analysis of binding probabilities for each nucleotide within the circRNA. This refined focus is significant, as it moves from merely identifying whether interaction occurs to pinpointing exactly where it is likely to transpire.

Technology Behind circMAC

CircMAC employs a sophisticated, three-pronged approach to preserve the unique circular structure of RNA. The tool's core technology incorporates:
1. Attention Mechanisms: This allows the model to consider information from all parts of the circRNA simultaneously, enhancing the prediction’s reliability.
2. Sequential Dependency Modeling: Utilizing Mamba models, circMAC effectively captures patterns of nucleotide sequences, maintaining the integrity of the circular form.
3. Convolutional Neural Networks: These identify short, local patterns linked to binding sites, with circular padding that ensures continuity across critical junctions.

Additionally, circMAC employs cross-attention techniques that adapt predictions based on the specific miRNA being analyzed. This factor is crucial as different miRNAs may interact with circRNAs in distinct ways, thus influencing the predicted binding probabilities.

Performance Evaluation of circMAC

The effectiveness of circMAC was rigorously tested using a dataset comprising over 27,000 training and testing pairs relating specifically to AGO-CLIP binding-site annotations. These evaluations demonstrated that circMAC significantly outperformed existing RNA language models and other approaches in precise binding site predictions. For instance, at a 1000-nucleotide benchmark, circMAC achieved an impressive F1-macro score of 0.762 and an area under the receiver operating characteristic curve (AUROC) of 0.902.

Implications for Future Research

CircMAC provides a powerful new resource for computational biology researchers, especially those focused on RNA-mediated regulation. By allowing researchers to prioritize and pinpoint potential miRNA binding regions, it can lead to more targeted experimental validations. An illustrative case analyzed by the research team showed predictive signals concentrated around the BSJ region of the circRNA isoform circCDYL2, demonstrating the tool's applicability in real-world investigations.

Despite these advances, it is essential to note that circMAC's predictions are intended to guide researchers toward candidate binding regions and do not establish functional interactions. Continued validation through independent datasets and experiments will be necessary before circMAC can be widely applied to clinical or broader biological contexts.

Conclusion

The introduction of circMAC not only represents a leap forward in our ability to understand and map circRNA–miRNA interactions but also highlights the emerging role of AI in genomics. By leveraging machine-learning techniques to model RNA sequences accurately, circMAC can substantially contribute to future studies on gene regulation, potentially leading to discoveries that enhance our understanding of various biological processes.

Reference

The detailed findings of this research were published in Volume 27, Issue 5 of Briefings in Bioinformatics on September 9, 2026. For further information, interested readers can consult the original publication titled "circMAC microRNA-conditioned binding-site localization on circular RNA isoforms."