From Object Detection to RNA Interaction: Evaluating YOLOv8 and CNNs on RRI Data

The recent convergence of computer vision and bioinformatics has opened new pathways for solving complex biological problems using deep learning. Originally developed for real-time object detection in images and videos, models like YOLOv8 (You Only Look Once, version 8) have demonstrated exceptional speed, accuracy, and generalization capabilities. This chapter investigates the adaptation of YOLOv8 and other Convolutional Neural Network (CNN) architectures to the domain of RNA-RNA Interaction (RRI) prediction, a crucial task for understanding post-transcriptional regulation and identifying disease-associated RNA networks. We present a comprehensive evaluation of how visual and structural encoding of RNA sequences enables CNN-based models to detect meaningful patterns related to interaction potential. Emphasis is placed on model architecture, input encoding (e.g., image-like matrices from sequence/structure data), transfer learning strategies, and evaluation metrics. Our results suggest that CNNs, particularly YOLOv8 with custom-modified layers, can be effectively repurposed to learn latent spatial dependencies in biological sequences-paving the way for interpretable and scalable RRI prediction pipelines. This cross-disciplinary study underscores the power of transferring methodologies from one domain (computer vision) to another (computational biology), demonstrating promising implications for both research and clinical diagnostics.