Dyna-1

Predict micro-millisecond protein dynamics from sequence and/or structure

Paper | GitHub | Model | Datasets | Colab

Dyna-1 predicts the probability that each residue experiences micro-millisecond motions. You can provide either a protein sequence, a structure (PDB ID or file), or both for the best performance.

Instructions

Authorize access to ESM-3 by logging in to HuggingFace (required for ESM-3)
Enter a protein sequence using standard amino acid letters (optional)
Provide a PDB ID (e.g., "1ubq") or upload a PDB file (optional)
Specify the chain ID if using a structure (default: A)
Choose whether to use the sequence from the PDB structure

You can toggle between using the ESM-3 and ESM-2 versions of the Dyna-1 model. To run with ESM-3, make sure you already have access to the EvolutionaryScale/esm3-sm-open-v1 weights here.

Note: The model will automatically set up the required environment on first run.

Use of this HF Space is subject to a Non-Commercial Use License.

Choose model version

Job Name (optional)

Protein Sequence

PDB ID

Use sequence from PDB

Or upload PDB file

Chain ID

Examples

Label	Protein Sequence	PDB ID	Chain ID	Use sequence from PDB

Download Results (.csv)

Download PDB

This HuggingFace Space was created by Gina El Nesr @ginaelnesr.

If you are using our code, datasets, or model, please use the following citation:

@article {Dyna-1,
    author = {Wayment-Steele, Hannah K. and El Nesr, Gina and Hettiarachchi, Ramith and Kariyawasam, Hasindu and Ovchinnikov, Sergey and Kern, Dorothee},
    title = {Learning millisecond protein dynamics from what is missing in NMR spectra},
    year = {2025},
    doi = {10.1101/2025.03.19.642801},
    journal = {bioRxiv}
}