The intricate network of RNA-RNA interactions, crucial for orchestrating essential cellular processes like transcriptional and translational regulation, has been unveiling through high-throughput techniques and computational predictions. With the emergence of deep learning methodologies, the question arises: how do these cutting-edge techniques for base-pairing prediction compare to traditional free-energy-based approaches, particularly when applied to the challenging domain of interaction prediction via chain concatenation? In this study, we employ base pairs derived from three-dimensional RNA complex structures as the gold standard benchmark to assess the performance of 22 different methods, including recently developed deep learning models. Our results demonstrate that the deep-learning-based methods, SPOT-RNA and coevolution-information-powered SPOT-RNA2, can be generalized to previously unseen RNA structures and are capable of making accurate zero-shot predictions of RNA-RNA interactions. The finding underscores the potential of deep learning as a robust tool for advancing our understanding of these complex molecular interactions.

Deep learning models of RNA base-pairing structures generalize to unseen folds and make accurate zero-shot predictions of base-base interactions of RNA complexes