A new method for designing RNA sequences to fit specific shapes

The MORAD framework uses multi-objective rewards to improve the accuracy and quality of RNA inverse design.

Researchers introduced MORAD to help with RNA inverse design. This process seeks nucleotide sequences that are compatible with a target three-dimensional backbone.

MORAD is an online reinforcement learning framework. It uses backbone-conditioned group sampling and group-relative reward normalization. It also uses a forward-process diffusion objective.

The method optimizes six different measurements. These include tertiary geometry, base pairing, ensemble behavior, and sequence composition. It maps each measurement to a bounded desirability score and combines them with a weighted geometric mean.

When applied to pretrained RIDE, MORAD improved GDT-TS from 0.33000.3300 to 0.38270.3827. It also increased pairing MCC from 0.61100.6110 to 0.72950.7295. It reduced normalized ensemble defect from 0.39060.3906 to 0.28300.2830.

MORAD outperforms RiboDiffusion, gRNAde, and RDesign. It shows better tertiary structural accuracy, secondary-structure pairing, and thermodynamic quality. This establishes a new state of the art in structure-conditioned RNA inverse design.

Key facts

  • MORAD improves 10 of 11 evaluation metrics that have a preferred direction.
  • The framework trains a single shared policy across diverse target backbones.
  • MORAD improves GDT-TS by 16.0%16.0\% and pairing MCC by 19.4%19.4\%.

All sections

Edition