Use IgDesign Online
Commercially Available IgDesign No-Code Web Server
IgDesign: Antibody CDR Design with Generative Inverse Folding, Validated In Vitro
IgDesign is a generative inverse folding method for antibody design that redesigns the complementarity-determining regions (CDRs) of an antibody given an antibody-antigen complex structure. It is described in a bioRxiv preprint and tested in vitro against eight therapeutic antigens.
IgDesign pairs IgMPNN, an antibody-specific inverse folding model, with the ESM2-3B protein language model, linked by a bottleneck adapter with cross-attention; the two sets of logits are summed.
Antibody binding is determined largely by the CDR loops, and the paper tests whether an antibody-specific inverse folding model can design them against a given antigen better than sequences sampled from known antibody-antigen data.
IgDesign tackles this by fine-tuning on antibody-antigen complexes and then validating designed sequences experimentally by surface plasmon resonance (SPR).
Item | Detail |
|---|---|
Task | Design heavy-chain CDR sequences (HCDR3, or all three HCDRs) for a given antigen complex |
Model | IgMPNN plus ESM2-3B with a bottleneck adapter |
Training | Pretrained on PDB proteins, fine-tuned on SAbDab antibody-antigen complexes; each antigen held out |
Validation | SPR on eight therapeutic antigens |
Library size | 1 million generated sequences per antigen, filtered to 100 HCDR3s and 100 HCDR123s |
Preprint DOI | 10.1101/2023.12.08.570889 |
How IgDesign Generates and Filters Antibody CDRs
Structure-conditioned design: each antigen has a reference antibody and complex structure that conditions sequence generation.
Held-out evaluation: models were trained per antigen with that antigen excluded from training.
Ranking: from 1 million sequences per antigen, those with the lowest cross-entropy loss were kept.
Baseline: each library also contained 100 HCDR3s sampled from SAbDab.
IgDesign Results: In Silico Recovery and Experimental Binding
Evaluation | Result |
|---|---|
Amino acid recovery | IgMPNN and IgDesign beat ProteinMPNN in all cases (p < 2e-4) |
HCDR3 binding (SPR) | IgDesign outperformed the baseline on all 8 antigens, significantly on 7 of 8 |
HCDR123 binding | Outperformed the HCDR3-only baseline on 7 of 8 antigens, significantly on 4 of 8 |
Baseline binding | 0% on antigens 2, 4, 5, 6, 7 and 8; a single baseline HCDR3 bound for two antigens |
The authors note that the HCDR123 comparison is more demanding because the baseline varies only HCDR3 while IgDesign redesigns all three CDRs.
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 IgDesign on Tamarind Bio
Therapeutic antibody discovery: generate candidate CDRs against known antigens.
Affinity-oriented redesign: redesign CDRs of a reference antibody on an existing complex structure.
Library design: build focused libraries of HCDR3 or HCDR123 variants for screening.
Comparison with natural diversity: benchmark designs against sequences sampled from SAbDab.
How to Use IgDesign on Tamarind Bio
Log in and open the tool: go to tamarind.bio and select IgDesign.
Provide the complex: upload an antibody-antigen structure, if exposed, and indicate the antibody chains and antigen.
Choose the regions to design: HCDR3 only or all heavy-chain CDRs (HCDR123), if exposed.
Set the number of sequences to generate and keep, if exposed.
Run the job and wait for completion.
Download outputs: the designed CDR sequences.
Interpret: rank by model loss and prioritize diverse candidates.
Use downstream: express and test designs, for example by SPR, as the authors did.
Things to Keep in Mind
The preprint text reviewed does not include a dedicated limitations section; the points below follow from the reported setup.
Designs depend on a high-quality antibody-antigen complex structure for the target.
Binding was validated for eight antigens only, and the authors trained new models per antigen with that antigen held out.
Affinity values are reported in the paper's appendix, so check the preprint before quoting them.
Computational designs still require experimental validation.
Source: IgDesign bioRxiv preprint, DOI 10.1101/2023.12.08.570889 (https://www.biorxiv.org/content/10.1101/2023.12.08.570889v1).