Mapping Circular RNA Could Be the Next Frontier in AI-Driven Biology

Researchers at South Korea’s Pusan National University have developed an artificial intelligence tool to predict where microRNAs bind to circular RNA molecules, a development that could help scientists investigate gene regulation and identify targets for further biological experiments.

The tool, called circMAC, predicts potential binding sites at the level of individual nucleotides across full-length circular RNA molecules, accounting for the distinctive structure of the molecules and the specific microRNA being studied.

The research, led by Professor Giltae Song and colleagues at the university’s School of Computer Science and Engineering, was published in the journal Briefings in Bioinformatics on September 9, 2026.

The findings come as AI is increasingly being applied to biological research, including the analysis of genetic sequences, the prediction of protein structures and the identification of potential drug targets. Such tools can help researchers narrow large search spaces, although computational predictions generally require laboratory validation before they can establish biological function or clinical relevance.

Circular RNAs, or circRNAs, are RNA molecules whose ends are joined to form a closed loop. They have attracted scientific interest because they can influence gene expression, including through interactions with microRNAs, small RNA molecules that help regulate how genetic information is translated into proteins.

Pinpointing these interactions can be difficult because circular RNAs may occur in different isoforms, or sequence variants, and their circular structure creates a junction where the ends of the sequence meet. Computational models that treat the molecules as linear sequences can miss patterns spanning that junction.

“The key advance is moving beyond asking whether a circRNA and miRNA interact to asking where that interaction is likely to occur,” Prof. Song told Impact Newswire.

CircMAC combines several machine-learning techniques to analyse RNA sequences. Attention mechanisms capture relationships across the sequence, a Mamba-based model learns sequential dependencies, and a one-dimensional convolutional neural network identifies local patterns associated with potential binding sites.

The model uses circular padding in its convolutional component to preserve sequence continuity across the junction. It also applies cross-attention to incorporate information about the specific microRNA being analysed and uses structure-aware pretraining based on RNA base-pairing predictions.

The researchers tested the system against binding-site annotations derived from AGO-CLIP experiments, which use cross-linking and immunoprecipitation to identify RNA regions associated with Argonaute proteins involved in microRNA-mediated gene regulation.

The main evaluation dataset comprised 19,674 training pairs and 7,838 test pairs, with 5.7% of nucleotide positions labelled as binding sites. At a 1,000-nucleotide setting, the model using base-pairing prediction pretraining achieved an F1-macro score of 0.762, an area under the receiver operating characteristic curve of 0.902 and an area under the precision-recall curve of 0.545.

The researchers reported that circMAC outperformed the evaluated pretrained RNA language models and rule-based screening methods under their benchmark conditions. It also recorded the strongest overall performance among the evaluated full-length models in tests designed to separate RNA isoforms and back-splice junctions between training and test data.

“By retaining full-length sequence context and conditioning predictions on the specific miRNA, the framework can provide more focused candidate regions for follow-up experiments,” said Prof. Song.

In one illustrative example, circMAC concentrated its predictions near an annotated junction region in circCDYL2, a circular RNA molecule. The researchers said the results could help scientists prioritise regions for laboratory investigation rather than testing every possible interaction.

The approach addresses a broader challenge in computational biology: identifying meaningful molecular interactions from large volumes of sequence data. RNA molecules can have different structures and regulatory roles, making accurate predictions dependent on both sequence context and the biological partners involved.

However, predicted binding sites do not by themselves demonstrate that an interaction occurs in living cells or produces a measurable biological effect. Further validation using independent datasets and laboratory experiments will be needed to establish the tool’s reliability across different RNA molecules and biological conditions.

The researchers said circMAC could support more targeted studies of RNA-mediated gene regulation, while its potential applications in disease research and clinical development remain to be established.

Stay ahead of the Stories shaping our world. Subscribe to Impact Newswire and join our 
WhatsApp Channel for updates on global tech, business, and innovation—all in one place.

Dive deeper into the future with the Cause Effect 4.0 Podcast, where we explore the ideas, trends, and technologies driving the global AI conversation.

Got a story to share? Contact Us to reach a global audience with Impact Newswire.