Use NA-MPNN Online
Commercially Available NA-MPNN No-Code Web Server
NA-MPNN: Inverse Folding and Specificity Prediction for Proteins, DNA and RNA
NA-MPNN is a message-passing neural network that designs and predicts sequences involving nucleic acids, handling protein, DNA and RNA nodes in a single architecture. It extends ProteinMPNN and LigandMPNN, and supports both nucleic acid sequence design from backbone structure and protein-DNA binding specificity prediction.
The authors note that RNA sequence design from structure and protein-DNA specificity prediction are both nucleic acid inverse-folding problems, yet existing methods are task-specific and no single architecture had addressed both. Experimental RNA structures are also scarce: as of January 2025, the PDB held 8,961 RNA-containing entries versus 235,538 protein-containing entries.
Property | Detail |
|---|---|
Capability | Sequence design for DNA and RNA, and protein-DNA specificity prediction |
Based on | ProteinMPNN and LigandMPNN |
Input | Protein, DNA and/or RNA backbone coordinates |
Output | Per-position distributions over 21 amino acid and 5 nucleic acid tokens |
Models | A design model and a specificity model |
How NA-MPNN Works
Unified graph: nodes can be protein residues, DNA bases or RNA bases, marked by a polymer-type embedding.
Neighbors: each node connects to its 32 nearest neighbors, using distance features over phosphate, sugar and virtual side-chain atoms.
Shared tokens: DNA and RNA share five nucleic acid tokens, and the model does not see side-chain chemistry.
Design model: trained against label-smoothed crystal sequences; sampled with 10 sequences per backbone at temperature 0.1.
Specificity model: trained against experimental position probability matrices (PPMs); predictions average 30 random-order decoding runs at temperature 0.6.
NA-MPNN Results for Nucleic Acid Sequence Design and DNA Binding Specificity
Sequence recovery
Context | Median recovery |
|---|---|
DNA-only | 57.4% |
RNA-only | 60.5% |
DNA in protein context | 58.6% |
RNA in protein context | 55.4% |
RNA pseudoknots and specificity
On a pseudoknot set, NA-MPNN had a predicted OpenKnot score of 83.2 versus 81.7 for gRNAde and 72.9 for RhoDesign, while RhoDesign had higher sequence recovery (67.7% versus 58.9%). In OpenKnot Round 6 (12 puzzles), the median experimental score was 89.9 for NA-MPNN versus 80.7 for gRNAde. For protein-DNA specificity, overall median MAE was 0.53 versus 0.86 for DeepPBS.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with NA-MPNN on Tamarind Bio
RNA design: propose sequences for RNA backbones, including pseudoknots.
DNA design: design DNA sequences, alone or in complex with a protein.
Specificity prediction: predict the DNA motif a protein binds as a PPM.
Context-aware design: hold the protein sequence fixed while designing the nucleic acid.
How to Use NA-MPNN on Tamarind Bio
Open the tool: log in to tamarind.bio and open NA-MPNN.
Select the task: design or specificity prediction, if exposed.
Provide a structure: upload protein, DNA and/or RNA backbone coordinates.
Fix positions: keep the protein sequence fixed as context or fix some design positions, if exposed.
Set sampling: the paper uses temperature 0.1 for design and 0.6 for specificity.
Run: submit the job.
Download and interpret: review designed sequences or the predicted PPM.
Validate: check designs with a structure predictor or experiments.
Things to Keep in Mind
RhoDesign had higher sequence recovery on the RNA-monomer and pseudoknot sets.
The pseudoknot benchmark is small (10 clusters).
The model ignores side-chain chemistry by design, and DNA and RNA share tokens.
Specificity training relies on public PPM databases and distilled structures.
Source: bioRxiv preprint describing NA-MPNN, https://www.biorxiv.org/content/10.1101/2025.10.03.679414v1.