Self-Playing RNA Inverse Folding

Stephen Obonyo, Nicolas Jouandeau, Dickson Owuor

Abstract: Ribonucleic acid (RNA) sequence controls many biological functions in the body of living organisms including protein synthesis and gene regulations. The sequence folds according to the low Minimum Free Energy (MFE) which enables it to achieve various roles. Given a target RNA structure, an RNA sequence can be designed to fold accordingly in a process called RNA Inverse Folding. In this paper, we present an RNA Inverse Folding model that performs a single-step look-ahead to select optimal RNA base and base pairs to design RNA sequences. Our proposed model (SPRNA) learns to accurately design RNA sequences based on self-improving policy function. The model recorded state-of-the-art results on two test datasets and very competitive results on others.