RWKV-IF: Efficient and Controllable RNA Inverse Folding via Attention-Free Language Modeling
Ji, G.; Xu, K.; Zheng, C.
Show abstract
We present RWKV-IF, an efficient and controllable framework for RNA inverse folding based on the attention-free RWKV language model. By treating structure-to-sequence generation as a conditional language modeling task, RWKV-IF captures long-range dependencies with linear complexity. We introduce a decoding strategy that integrates Top-k sampling, temperature control, and G-C content biasing to generate sequences that are both structurally accurate and biophysically meaningful. To overcome limitations of existing datasets, we construct a large-scale synthetic training set from randomly generated sequences and demonstrate strong generalization to real-world RNA structures. Experimental results show that RWKV-IF significantly outperforms traditional search-based baselines, achieving higher accuracy and full match rate while greatly reducing edit distance. Our approach highlights the potential of lightweight generative models in RNA design under structural and biochemical constraints. The code is available at https://github.com/Lyttr/RWKVInverseFolding.git
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Adversarial domain translation networks for fast and accurate integration of large-scale atlas-level single-cell datasets 93%
- ECloudGen: Leveraging Electron Clouds as a Latent Variable to Scale Up Structure-based Molecular Design 92%
- Symphonizing pileup and full-alignment for deep learning-based long-read variant calling 91%
Similar papers in this journal
- Accelerating protein engineering with fitness landscape modeling and reinforcement learning 95%
- 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%
Similar papers in this journal
"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.