The trRosettaRNA server for RNA structure prediction

Similar to proteins, many RNAs fold into three-dimensional (3D) structures to perform biological functions. Here we present the trRosettaRNA server, a web-based platform for automated RNA 3D structure prediction using deep learning. The primary input is the nucleotide sequence of a target RNA, with the option to upload custom multiple sequence alignments and secondary structures. The server uses an end-to-end neural network for automated 3D structure prediction, followed by an energy optimization step to resolve structural violations. As an automated server, trRosettaRNA is distinguished by its state-of-the-art modeling accuracy, flexible input options and comprehensive visualization of prediction results. trRosettaRNA has been successfully applied in various contexts, including predicting structures for Rfam families lacking known 3D structures, where representative cases of high-confidence structure predictions were found to align well with subsequent experimental observations. Utilizing up to 5 central processing unit (CPU) cores in parallel on our computer cluster, the server takes a median time of about 1 h to predict structures for RNA sequences with about 200 nucleotides. The standalone package for trRosettaRNA offers distinct advantages such as enhanced data privacy for sensitive sequences, the ability to bypass server queues and integration into high-throughput automated pipelines. Importantly, the open-source nature of the package empowers researchers to directly modify the codebase for specialized research needs or to develop derivative tools by fine-tuning the underlying neural network. The web server and standalone package of trRosettaRNA are available at https://yanglab.qd.sdu.edu.cn/trRosettaRNA/ and https://github.com/YangLab-SDU/trRosettaRNA2 , respectively.