AbstractFor many RNA molecules, the secondary structure is essential for the correction function of the RNA. Predicting RNA secondary structure from nucleotide sequences is a long-standing problem in genomics, but the prediction performance has reached a plateau over time. Traditional RNA secondary structure prediction algorithms are primarily based on thermodynamic models through free energy minimization. Here we propose a deep learning-based method, called UFold, for RNA secondary structure prediction, trained directly on annotated data without any thermodynamic assumptions. UFold improves substantially upon previous models, with approximately 31% improvement over traditional thermodynamic models and 24.5% improvement over other learning-based methods. It achieves an F1 score of 0.96 on base pair prediction accuracy. An online web server running UFold is publicly available at http://ufold.ics.uci.edu.