RNASTOP

This repository contains codes, data and trained models for RNASTOP. RNASTOP is a a novel framework that accurately predicts mRNA degradation at both single-nucleotide and full-length levels, while also enhancing mRNA stability through codon optimization. RNASTOP encompasses an mRNA degradation prediction model and an mRNA codon optimization module. The mRNA degradation prediction model primarily comprises three modules: the mRNA feature embedding module, the dual-branch feature decoupling-and-aggregating network and the prediction layer. Firstly, the model integrates nucleic acid LLMs and the structure embedding module to obtain the multi-source feature embeddings of mRNA sequences. These embeddings are then fed into the dual-branch feature decoupling-and-aggregating network for feature fusion. Finally, the fused feature embeddings are fed into the prediction layer for the prediction of mRNA degradation. mRNA codon optimization module further integrates the mRNA degradation prediction model with beam search for codon optimization to improve the stability of mRNA. You can find more details about RNASTOP in our paper, "RNASTOP: Prediction and optimization of mRNA stability by integrating deep learning and thermodynamic property-guided heuristic search" (Lin et al., 2024).