Evoflow-RNA: Generating And Representing Non-Coding RNA With A Language Model
Patel, S.; Peng, F. Z.; Fraser, K.; Friedman, A. D.; Chatterjee, P.; Yao, S.
Show abstract
RNA plays a critical role across numerous biological functions. Recent advances in language modeling show promise with representing RNA, but the possibility of large-scale RNA design and optimization has not been fully explored. We propose EvoFlow-RNA, a bidirectional non-coding RNA language model leveraging a masked discrete diffusion model (MDM) formulation for both generative modeling and representation learning. EvoFlow-RNA bridges the gap between RNA sequence representation and design. It outperforms leading RNA models on three BEACON tasks critical to understanding RNA function, spanning from structure prediction to gene editing. For unconditional generation, it synthesizes diverse RNA sequences with native-like structural and binding properties. Additionally, EvoFlow-RNA can globally redesign aptamer sequences around preserved binding recognition sites with enhanced functionality. Our results demonstrate the effectiveness of EvoFlow-RNA in RNA modeling, highlighting the capability and potential of masked discrete diffusion for both recapitulating and enhancing existing RNAs.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- DUETT quantitatively identifies known and novel events in nascent RNA structural dynamics from chemical probing data 97%
- Graph neural representational learning of RNA secondary structures for predicting RNA-protein interactions 96%
- Deep learning models for RNA secondary structure prediction (probably) do not generalise across families 96%
Similar papers in this journal
Similar papers in this journal
- Predicting RNA 3D structure and conformers using a pre-trained secondary structure model and structure-aware attention 95%
- Interpreting Neural Networks for Biological Sequences by Learning Stochastic Masks 94%
- Accelerating protein engineering with fitness landscape modeling and reinforcement learning 93%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.