DS3dRNA: De novo design of functional RNAs through higher-order interactions

De novo design of 3D RNAs via higher-order interactions

Read the bioRxiv preprint · Posted September 26, 2026

De Novo 3D RNA Design Using Higher-Order Interactions
Oral presentation · 14th National Conference on Soft Matter and Biological Physics

Quick start · Method · Examples · Datasets · Citations

Fixed-backbone sequence design and high-throughput ranking for single-state and multi-state 3D RNA/DNA scaffolds.

DS3dRNA searches nucleotide sequence space for sequences compatible with one or more fixed three-dimensional nucleic-acid scaffolds. It combines coarse-grained three-body TriRNASP energies, local nearest-neighbor thermodynamic screening, optional secondary-structure and frozen-site constraints, and Monte Carlo sampling. The same public entry point provides DS3dRank for scoring externally generated FASTA candidates.

Research preprint

De novo design of functional RNAs through higher-order interactions — bioRxiv, September 26, 2026 (preprint; not peer reviewed).

The manuscript evaluates single-state and multi-state RNA design and reports functional assays in two RNA systems:

• Mango II: all three tested designs retained fluorogenic activity and sub- to low-nanomolar apparent affinity, with larger relative fluorescence increases at saturation than wild type.
• Twister ribozymes: five designs yielded mean 20-second endpoint cleavage fractions of 37.7–50.6%, compared with 23.5% for wild type, at 25 °C (four replicate experiments).

See the preprint for benchmark protocols, design constraints, experimental conditions, and limitations, and CITATIONS.md for the full citation and BibTeX.

At a glance

🧬 Design	🔗 Constrain	⚡ Rank
Single target, PDB batch, or residue-matched multi-state ensemble	Automatic, explicit, or zero-contact secondary structure; optional frozen residues	Fine-energy scoring of FASTA candidates on single, batch, or multi-state scaffolds
RNA and DNA energy models	Pseudoknot-aware dot-bracket and chain breaks	CPU support and CUDA acceleration

The model represents each residue with P, C4′, and N1/N9 sites, preserving backbone and base-anchor geometry without exposing complete nucleotide-specific heavy-atom patterns to the design procedure. See the method overview for the scientific workflow.

Quick start

bash DS3dRNA_Installer.sh
conda activate DS3dRNA
python DS3dRNA.py --help

The installer first restores the four model tensors from Energy.zip, then prepares the Conda/PyTorch environment. Re-running it safely skips extraction when all tensors are already present and valid.

Task	Command
RNA design (default)	python DS3dRNA.py Examples/Design/inputs/8VY0.pdb --ss none --batch 10
DNA design	python DS3dRNA.py target_DNA.pdb --mol DNA --batch 10
Sequence ranking	python DS3dRNA.py -rank --str target.pdb --fa candidates.fasta

Note

--mol RNA is the default; --mol DNA selects the DNA energy tensors, thermodynamic backend, pairing rules, and T-based output alphabet. See RNA and DNA molecule modes.

Multi-state design with explicit structural and frozen-site constraints

python DS3dRNA.py -m Examples/MultiState_Design/inputs \
  --ss Examples/MultiState_Design/inputs/ss.dbn \
  --frz Examples/MultiState_Design/inputs/Frz.fasta \
  --batch 10

Rank the bundled FASTA candidates

python DS3dRNA.py -rank \
  --str 'Examples/Seq_Rank/R1138_7PTL_A:720.pdb' \
  --fa Examples/Seq_Rank/candidates.fasta \
  --rank_out ranked_candidates.csv

Important

Omitted --ss, --ss auto, and --ss none have deliberately different behavior. Read Secondary structure before production runs.

Seed replay must preserve the original execution mode: replay batch-generated results with the saved --seed_batch archive, keeping the complete seed list and its order unchanged; replay standalone results with --seed. See Reproduce a run.

Documentation

Get started	Science and data	Reference and policy
Installation Environment and CUDA verification	Method overview Scientific workflow	Citation guide DS3dRNA and dependencies
Quick start Design and ranking recipes	Energy tensors RNA/DNA model assets	Third-party notices Redistributed materials
Inputs and constraints RNA/DNA, PDB, DBN, FASTA, frozen masks	Datasets Training and benchmark collections	License Academic/non-commercial terms
Outputs Summary, ensemble, and trajectories	Examples Reproducible workflows	Optional DSSR utility Acquisition and citation

The Mode/ directory contains the internal computational core. Users should invoke DS3dRNA.py and follow the documented examples rather than calling developer modules directly.

✉️ Contact

Prof. Zhi-jie Tan
Wuhan University
Email: zjtan@whu.edu.cn

Citation

If DS3dRNA contributes to published work, cite both the DS3dRNA manuscript and the TriRNASP energy model. Ready-to-copy entries are available in CITATIONS.md and CITATION.cff.

License

DS3dRNA-authored materials are distributed under the academic and non-commercial terms in LICENSE.txt. The GNU LGPL 2.1 text is provided separately in LICENSES/LGPL-2.1.txt; other third-party files remain governed by their respective terms in THIRD_PARTY_NOTICES.md.

Commercial licensing: tongwei.tovi.yuan@gmail.com · tovi_yuen@whu.edu.cn