AI Tool Maps microRNA Binding Sites in Circular RNAs

Circular RNAs (circRNAs) are loop-shaped molecules that can influence gene regulation through interactions with microRNAs (miRNAs). These interactions are complex, as circRNAs can exist in different full-length isoforms. Additionally, the back-splicing junction (BSJ) connects the two ends of the RNA, and treating a circRNA as a linear transcript can ignore sequence information near this junction ².

Tool Functionality

The developed tool, called circMAC, uses artificial intelligence to predict miRNA binding sites in circRNAs. It takes as input a complete circRNA isoform and a mature miRNA sequence, and predicts a binding probability for each nucleotide of the circRNA. To do this, circMAC combines three sequence modeling approaches: attention, which captures information from the entire circRNA sequence; Mamba, which models sequential dependencies; and a one-dimensional convolutional neural network, which identifies short local patterns associated with binding sites ^1,2.

Evaluation and Results

The researchers evaluated circMAC using complete circRNA isoforms with AGO-CLIP binding site annotations. The main dataset contained 19,674 training pairs and 7,838 test pairs, with 5.7% of nucleotide positions labeled as binding sites. The results indicated that circMAC improves nucleotide-level binding site localization compared to other models ¹.