Repository files navigation

Tertiary Structure-based RNA Inverse Folding with Generative Diffusion Models

Please refer torequirements.txt for the required packages.

Model checkpoint can be downloaded fromhere.Another checkpoint trained on the full dataset (with extra 0.1 Gaussian noise for coordinates) can be downloaded fromhere.

Download and put the checkpoint files in theckpts folder.

Inference demo notebook to get started:.

Run the following command to run the example for one sequence generation:

CUDA_VISIBLE_DEVICES=0 python main.py --PDB_file example/R1107.pdb

The generated sequence will be saved inexp_inf/fasta/R1107_0.fasta.

Multiple sequence generation can be run by:

CUDA_VISIBLE_DEVICES=0 python main.py --PDB_file example/R1107.pdb --config.eval.n_samples 10

For more sequence diversity, you can useexp_inf_large.pth or adjust the conditional scaling weight by--config.eval.cond_scale.

An example for adjusting the conditional scaling weight is as follows:

CUDA_VISIBLE_DEVICES=0 python main.py --PDB_file example/R1107.pdb --config.eval.n_samples 10 --config.eval.dynamic_threshold --config.eval.cond_scale 0.4

If you find this work useful, please cite:

@article{10.1093/bioinformatics/btae259,    author = {Huang, Han and Lin, Ziqian and He, Dongchen and Hong, Liang and Li, Yu},    title = {RiboDiffusion: tertiary structure-based RNA inverse folding with generative diffusion models},    journal = {Bioinformatics},    volume = {40},    number = {Supplement_1},    pages = {i347-i356},    year = {2024},    month = {06},    issn = {1367-4811}}

This project is licensed under theMIT License.