The method combines RNA sequences with secondary-structure information during pre-training. The scientific paper describes a dataset of about 21 million sequence–structure pairs and evaluations across several prediction tasks.
The researchers report improvements in analysing RNA structure and function. These findings concern their experimental evaluations; they do not establish that a medical treatment or clinical tool is ready for use.
Opening the resources allows others to inspect and reproduce the work. The reviewed sources do not yet establish broad independent validation of the reported performance.