NA-MPNN

Pre-trained model for joint protein, DNA, and RNA sequence design and scoring from molecular backbones.

Disclaimer

This is an UNOFFICIAL implementation of the RNA sequence design and protein-DNA specificity prediction with NA-MPNN by Andrew Kubaney, et al.

The OFFICIAL repository of NA-MPNN is at baker-laboratory/NA-MPNN.

The team releasing NA-MPNN did not write this model card for this model so this model card has been written by the OFoldX team.

Model Details

NA-MPNN extends fixed-backbone inverse folding to protein, DNA, RNA, and mixed-polymer assemblies. A geometric message-passing graph represents polymer residues, and generation restricts identities to the residue's polymer type. Separate checkpoints support joint sequence design and protein-DNA specificity prediction.

Please refer to the Training Details section for more information on the training process.

Variants

oteam/nampnn: Sequence-design model for protein and nucleic-acid assemblies.
oteam/nampnn-specificity: Protein-DNA specificity model.

Model Specification

Vocabulary Size	Hidden Size	Encoder Blocks	Decoder Blocks	Neighbors	Atom Slots per Residue	Num Parameters (M)
72	128	3	3	32	16	2.293

Usage

The model file depends on the ofoldx library. You can install it using pip:

pip install ofoldx

Direct Use

Sequence Design

You can use this model directly with a pipeline for joint protein and nucleic-acid sequence design:

import ofoldx
from ofoldx.io import read_structure

structure, _ = read_structure("complex.cif")
pipeline = ofoldx.pipeline("design", model="oteam/nampnn")
result = pipeline(structure)

Each candidate contains the designed molecular target and named sequence scores. Shared nucleic tokens are the released default: RNA bases use DNA output classes during prediction and are reconstructed as RNA residues.

With shared nucleic tokens, global_omit_tokens uses DNA class names for both DNA and RNA: "DNA_T" excludes T and U, respectively. RNA_ names are rejected in this mode.

Sequence Scoring

You can use this model directly with a pipeline for sequence scoring on a mixed-polymer backbone:

import ofoldx
from ofoldx.io import read_structure

structure, _ = read_structure("complex.cif")
pipeline = ofoldx.pipeline("score", model="oteam/nampnn")
result = pipeline(structure, num_decoding_orders=8, seed=7)

The result contains designed and global sequence NLL in nats per residue.

See the pipeline guide for staged execution with prepare, forward, and postprocess.

Training Details

NA-MPNN is trained for structure-conditioned recovery of protein and nucleic-acid sequences.

Training Data

NA-MPNN was trained on protein, DNA, RNA, and mixed-polymer structures. The design and specificity variants target sequence recovery and protein-DNA specificity, respectively.

Training Procedure

NA-MPNN uses polymer-conditioned sequence recovery, with coordinate noise and optional protein-first decoding.

OFoldX exposes differentiable forward and the polymer-specific smoothed_loss for masked sequence recovery. Dataset construction, the upstream specificity-training dataset, and complete training orchestration remain the training application's responsibility.

Citation

@article{kubaney2025nampnn,
  author = {Kubaney, Andrew and Favor, Andrew and McHugh, Lilian and Mitra, Raktim and Pecoraro, Robert and Dauparas, Justas and Glasscock, Cameron and Baker, David},
  title = {RNA sequence design and protein-DNA specificity prediction with NA-MPNN},
  journal = {bioRxiv},
  year = {2025},
  doi = {10.1101/2025.10.03.679414},
}

The artifacts distributed in this repository are part of the OFoldX project. If OFoldX supports your research, please cite the OFoldX project as follows:

@software{chen2026ofoldx,
  author = {Chen, Zhiyuan},
  title  = {OFoldX},
  year   = {2026},
  url    = {https://github.com/OTeam-AI4S/OFoldX},
}

Contact

Please use GitHub issues of OFoldX for any questions or comments on the model implementation or model card.

Please contact the authors of the NA-MPNN paper for questions or comments on the paper/model.

License

The model weights, implementation, configuration, and documentation distributed by OFoldX are licensed under the GNU Affero General Public License.

For additional terms and clarifications, please refer to our License FAQ.

SPDX-License-Identifier: AGPL-3.0-or-later

Downloads last month: -; Downloads are not tracked for this model. How to track

Safetensors

Model size

2.3M params

Tensor type

F32

Inference Providers NEW

Other

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

oteam
/

nampnn