Design of function-regulating RNA via deep learning and AlphaFold 3

RNAs are programmable macromolecules that play diverse regulatory roles in living organisms. However, the intricate structure-function relationships underlying their regulatory activities pose significant challenges for RNA design. Here, we introduce a computational framework that integrates deep learning and energy-based methods to enhance the sequence diversity of sgRNAs designs. Our approach demonstrates high editing efficiencies of up to 75% for gene knockouts, 100% for large fragment deletions, and 62.5% for multiplex gene editing using the designed sgRNAs. Molecular dynamic simulations suggested the stability of DNA-RNA-protein complex is essential to the functionality of designed RNAs. Moreover, we reveal that the confidence metrics of AlphaFold 3 can effectively distinguish functional sequences, enabling one-shot design of crRNAs. This work presents an efficient strategy for designing regulatory RNAs with complex interactions and establishes the potential of AlphaFold 3 in advancing RNA design.