Use ODesign Antibody Online
Commercially Available ODesign Antibody No-Code Web Server
ODesign: All-Atom Generative Model for Epitope-Targeted Binder Design
ODesign is an all-atom generative "world model" for all-to-all biomolecular interaction design: given a target and an optional epitope, it generates binding partners such as proteins, nucleic acids and small molecules. It was introduced in an arXiv preprint (2510.22304) by a team led from Lingang Laboratory in Shanghai, with Zhejiang University, The Chinese University of Hong Kong, Shanghai Artificial Intelligence Laboratory and other institutions. A web server is available at odesign.lglab.ac.cn.
Most design methods are built for one modality, such as protein binders or small-molecule ligands, and give limited control over where on the target the partner binds. ODesign aims to unify these tasks in a single framework with fine-grained control over the interaction.
On computational benchmarks, the authors report that it outperforms modality-specific baselines in controllability and performance, and extends to protein-binding RNA and DNA and to nucleic-acid-binding ligands.
Property | Detail |
|---|---|
Model type | AlphaFold3-like all-atom generative model with a 48-block Pairformer and a conditional diffusion decoder |
Design control | User-specified epitopes (hotspots) on arbitrary targets |
Modes | Rigid-target mode and flexible-target mode |
Design outputs | Protein (-P), nucleic acid (-N) and ligand (-L) partners |
Training data | PDB structures released before 30 September 2021 plus AlphaFold2 and OpenFold distilled structures |
Why Epitope-Controlled Binder Design Matters
Directing a binder to a specific epitope is central to designing antibodies and binding proteins that block, activate or detect a target at the right site. A model that accepts epitope information and works across modalities lets one pipeline cover protein binders, motif scaffolds and ligand design.
How ODesign Works
Architecture and tokens
Unified tokens: each modality's minimal chemical unit becomes a generative token, with a suffix marking protein, nucleic acid or ligand design.
Conditioning: a conditional module encodes the target's 3D structure, and a diffusion decoder generates all-atom coordinates.
Two-stage design: backbone generation first, followed by sequence design with an inverse-folding module (OInvFold).
Hotspot control: hotspots are residues within 8 Å of the partner across entities, flagged as an input feature; during training between 5 and 20 hotspots are retained so the model does not depend on a full epitope.
Rigid and flexible targets
In rigid mode the target keeps its true coordinates. In flexible mode the target and binder are co-generated.
Benchmark Results for Protein, RNA, DNA and Ligand Design
Benchmark | Reported result |
|---|---|
Protein binders (800 designs per target) | 2,672 successful designs per GPU-day vs 555 for RFDiffusion, 61 for BoltzDesign and 22 for BindCraft |
Motif scaffolding (MotifBench, 30 cases) | 18.4% average success vs 14.3% for RFDiffusion-AA; successes in 27 of 30 cases |
Interface design (4 targets) | Pocket RMSD 2.24 Å lower than PocketGen |
Protein-binding RNA (10 complexes) | 77.95% success at a 5 Å RMSD threshold |
Protein-binding small molecules | 9.0-fold more successful molecules than TargetDiff |
These results come from computational refolding with AlphaFold3 and Chai-1 and associated filters; the paper text we reviewed does not present wet-lab validation.
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 ODesign on Tamarind Bio
Protein binder design: generate binders against a target with a chosen epitope.
Motif scaffolding: build scaffolds around functional motifs.
Nucleic acid binders: design protein-binding RNA or DNA.
Ligand design: generate small molecules for protein, DNA or RNA targets.
How to Use ODesign on Tamarind Bio
Open the tool: log in to tamarind.bio and select ODesign Antibody.
Provide the target: upload the 3D structure of the target, since the model conditions on it.
Define the epitope: specify hotspot residues on the target, if exposed.
Choose the design mode: rigid or flexible target, and the binder type, if exposed.
Set design size: choose binder length and the number of designs, if exposed.
Run the job: submit and wait for generation.
Download and filter: retrieve structures and sequences, then refold and rank them; the authors rely on AlphaFold3 and Chai-1 refolding with iPAE-type metrics.
Validate: test top designs experimentally.
Things to Keep in Mind
Performance drops under some out-of-distribution conditions.
DNA aptamer designability falls off faster with length than RNA.
There is no fine-grained control over properties such as secondary-structure content or small-molecule aromaticity.
Training data are dominated by protein and ligand examples, and RNA shows the highest loss.
Source: Zhang et al., "ODesign: A World Model for Biomolecular Interaction Design", arXiv:2510.22304 (2025), arxiv.org/abs/2510.22304. Web server: odesign.lglab.ac.cn.