All-atom molecular design

PartiDesign

Joint molecular structure
and identity generation

Design proteins, peptides, antibodies, DNA, RNA and small molecules — with sequences or chemical identities and three-dimensional atomic structures generated together.

Start with pipeline Agent Skill GitHub Web server

Compose your conditionsFixed sequencesMotifsSecondary structureHotspotsExplicit bonds

Identity and geometry, designed together.

Generate molecules on their own or in protein, nucleic-acid and small-molecule environments. Combine molecular context with the conditions that define your task.

Model overview: molecular inputs and sequence, motif, secondary-structure and bond constraints pass through an encoder and flow matcher to a decoder that jointly produces identities and atomic coordinates. — The PartiDesign framework connects molecular context, composable conditions and joint structure–identity generation.
View full figure

Capabilities

Design across
molecular modalities.

From new molecular identities to fixed-identity docking, use one framework with the context and constraints your design needs.

01 / PROTEIN

Proteins, peptides & antibodies

Generate monomers and binders, scaffold residue or atomic motifs, and guide secondary structure. Design linear, cyclic and crosslinked peptides, or selected CDRs in paired VH/VL and VHH scaffolds. Fixed-identity peptides support docking.

Protein examples Peptide topologies Antibody examples

Protein capability examples: protein binders with secondary-structure guidance, protein design in RNA, DNA and ligand contexts, motif scaffolding, peptide topologies, Fv and VHH antibody design, and peptide docking. — Protein generation, peptide chemistry, antibody CDR design and docking.
View full figure

02 / NUCLEIC ACIDS

DNA & RNA

Design aptamers around proteins or small molecules, or generate structures for fixed sequences. Combine paired/unpaired, specified-pair and stem-loop guidance with sequence and motif conditions. Fixed-identity ssRNA supports docking.

Nucleic-acid examples

Nucleic-acid capability examples: RNA and DNA design around proteins and ligands, ssRNA docking, paired and unpaired preferences, specified base pairs, stem-loop guidance, and fixed-sequence structure prediction. — Aptamer design, pairing guidance and fixed-sequence structure generation.
View full figure

03 / SMALL MOLECULES

Ligands & molecular fragments

Generate pocket-conditioned ligands, grow or link retained fragments, and explore scaffold changes. Fixed-identity ligands support docking in protein or nucleic-acid context.

Ligand examples Docking examples

Small-molecule capability examples: de novo ligands in protein, RNA and DNA pockets, docking, pose-conditioned and atomic-motif redesign, fragment growing and linking, connector replacement, and scaffold hopping. Colors distinguish retained, editable and generated atoms. — Pocket-conditioned generation, molecular editing and docking.
View full figure

The iterative pipeline

One command for de novo design and structure-guided redesign.

Connect generation, partial diffusion, MPNN sequence redesign, refolding and final validation in one runner. Start from a task YAML or an existing complex.

Pipeline progression: a de novo or existing binder complex passes through three rounds of MPNN redesign, refolding and filtering, with partial diffusion between rounds, followed by final validation and accepted candidates. — Three rounds of sequence–structure refinement, followed by final structure validation.
View full figure

Start from your task

Use a design YAML, or describe starting complexes and design chains in annotation.csv.

Keep native constraints

Fixed identities, motifs and chemical connections remain part of the task through refinement.

Collect accepted candidates

Refold and filter between rounds, then apply the selected preset’s final validation.

Your first insulin receptor binder

partidesign pipeline configs/examples/pipeline_tasks.yaml \
  --task protdbench_insulin_receptor --preset ppi \
  --config configs/pipeline/demo.yaml --run-name insulin_receptor_demo

The lighter demo runs three rounds with ESMFold2-fast, relaxed pipeline screening and partial diffusion, followed by five Mol-1 samples for final validation. Prepare the environment and demo models below.

Refine an existing peptide complex

partidesign pipeline configs/examples/optimization \
  --config configs/pipeline/optimize.yaml --preset ppi \
  --run-name peptide_optimization

Optimize the bundled linear and cyclic peptide complexes. The directory’s annotation.csv describes each starting complex. This workflow begins with partial diffusion and uses the full selected preset for final validation; prepare its evaluation models.

Optimize a GLP1R VHH

partidesign pipeline configs/examples/antibody_opt \
  --preset antibody --run-name glp1r_vhh_optimization

Use the antibody preset with the bundled annotated starting complexes. Native CDR/framework permissions and the paratope declaration guide optimization. See the antibody pipeline guide for model requirements and final validation.

For all command examples, open the pipeline guide. JavaScript enables example tabs and copy controls.

Add --plan to preview a command without starting model jobs.

Accepted candidates: inference/<run>/<task>/final/success/

Explore presets, annotation fields and resume

Get started

From your first run
to your own target.

Recommended · Agent Skill

Bring MilkyFrog to your coding agent.

The PartiDesign MilkyFrog skill helps set up the environment, prepare model downloads, adapt task YAML to your molecules and constraints, and run or troubleshoot design workflows.

Open Agent Skill

Give your agent this prompt

Use $partidesign-milkyfrog to set up PartiDesign, download the models needed for the insulin receptor pipeline demo, and help me adapt a task to my target.

Set up the environment

Use Linux with an NVIDIA GPU and Conda or a compatible environment manager. Follow the commands below from your working directory.

01 Clone and install

git clone https://github.com/MegaGatlingPea/PartiDesign-official.git
cd PartiDesign-official
conda env create -f partidesign_cu12.yaml
conda activate partidesign

For CUDA 13 and alternative setups, see the environment guide.

02 Download the models

Download the PartiDesign generation bundle and save it as ckpts/partidesign.ckpt. This checkpoint is a separate download from the environment installation.

Then prepare the additional models for the lighter insulin receptor pipeline demo:

Additional demo models

partidesign download esmfold2-fast mol1 solublempnn

The download guide covers terminal downloads, model locations and predictors required by full presets.

Documentation

Make the next step yours.

Open the user guide

Prefer visual task setup?

Explore PartiDesign through the web server.

Open web server