Enhancing mRNA translation efficiency with discriminative and generative artificial intelligence by optimizing 5' UTR sequences

The mRNA-based therapeutics, notably mRNA vaccines, represent a new era of powerful tools to combat various diseases. However, the relatively low translation efficiency of exogenous mRNA often limits its wide application. Here, we propose a computational framework called UTailoR (UTR tailor), which significantly improves the challenge by optimizing 5' UTR sequences based on a two-step artificial intelligence strategy. We first develop a deep-learning-based discriminative model for predicting mRNA translation efficiency with 5' UTR sequences and then present a generative model to generate optimized 5' UTR sequences, which are designed to be highly close to the original sequences but predicted to result in high translation efficiency. The experimental results show that the UTailoR-optimized sequences outstrip the corresponding original sequences by ∼200%. This work provides an efficient and convenient method for mRNA 5' UTR optimization, which can be easily accessed online.