MORAD

Reinforcement Learning with Multi-Objective Rewards for RNA Inverse Design

MORAD post-trains a single backbone-conditioned RNA diffusion policy online with a six-objective reward covering tertiary geometry, base pairing, ensemble behavior and sequence composition. The trained policy designs sequences for a new target backbone directly, with no per-target optimization.

• 🧬 MORAD-RIDE: the released model, post-trained from pretrained RIDE
• 📦 MORAD-Targets: 527 training and 153 test RNA backbone targets
• 💻 Code: training, evaluation and reproduction scripts
• 🌐 Project page: overview, method and results