Skip to content

jamesguoxin/nwRNAFoundation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 

Repository files navigation

Nüwa.RNA Foundation Model

English | 中文


Model Description

Nüwa.RNA is a large-scale generalist foundation model developed by the Cheng Yuan and Guo Xin team at Shanghai Academy of Artificial Intelligence for Science (SAIS). It establishes a unified representation of RNA sequence, structure, and function.

The model is available in multiple sizes, scaling up to 30 billion parameters, and is trained on a massive corpus of diverse RNA types using a novel multi-level masked self-supervised learning framework. This approach synergizes synchronized single-token masking with span-based masking strategies and explicitly incorporates secondary structure information within an optimized architecture. This multi-modal training regime enables the emergence of advanced cognitive capabilities regarding RNA structural features and chemical modifications.

Nüwa.RNA achieves state-of-the-art performance across 43 evaluation metrics spanning sequence generation, structure prediction, and functional inference, ranking first in comprehensive benchmarks. Notably, Nüwa.RNA attains optimal results in 42 of these 43 metrics, surpassing leading models such as RNA-FM, RNAGenesis, and AIDO.RNA.

Beyond in silico benchmarking, Nüwa.RNA's practical utility has been validated through a "Lab-in-the-loop" system for nucleic acid drug design. In experimental validation targeting five distinct targets across aptamer and siRNA modalities, Nüwa.RNA reduced wet-laboratory validation costs by over 90%.

Model Architecture

Nüwa.RNA is built upon a modernized encoder-only Transformer architecture with:

  • Rotary Positional Embeddings (RoPE) for better relative positioning between nucleotides
  • GeGLU activation functions for improved training stability
  • Hybrid attention combining sliding window and global attention mechanisms
  • Specialized heads for masked language modeling and structural constraint prediction

Nüwa.RNA-1.6B Configuration:

Hyperparameter Value
num-layers 32
hidden-size 2,048
ffn-hidden-size 5,440
num-attn-heads 32
vocab-size 16

Model Access

🔗 NovaInspire Platform: https://aistudio.ai4s.com.cn/galaxy-model/model/167


License

Please refer to the model portal for licensing information.

Citation

If you use Nüwa.RNA in your research, please cite our technical report.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors