Stanford tech advances RNA structure prediction

A deep neural network approach developed by Stanford University researchers could improve the prediction of RNA structures, laying the foundation for discovery of RNA-targeted drugs. In recent rounds of the RNA-Puzzles blind structure prediction challenge, the network, dubbed ARES, outperformed at least nine other methods in predicting structures of four RNAs, all of which were larger and more complex than the 18 RNA molecules used to train the network. ARES predicted structures of complex RNAs at a root-mean-square deviation, a measure of the similarity between the known and modeled structures, of about 12 angstroms, a four-angstrom improvement over prior methods, according to RNA researcher Kevin Weeks in a Perspective on the Science study.

Until recently, the 3-D structures of RNAs were not thought to be important for their functional roles. A 2018 Cell study led by Weeks made the case that complex, functionally relevant RNA structures are the rule, not the exception...