Use IgGM Online
Commercially Available IgGM No-Code Web Server
IgGM: Generative Antibody Design and Antibody-Antigen Complex Structure Prediction
IgGM is a generative model for de novo antibody design that co-designs CDR sequences and the full antibody structure for a target antigen. Given the antigen sequence and structure plus the framework-region (FR) sequences, it does not need FR structures, which are often unknown for a new antigen. It handles antibodies (two chains) and nanobodies (single heavy chain).
Antibody engineering is slow because binding is set by hypervariable CDR loops and their relation to the antigen. The authors argue that design tools should work from sequence-level information about the framework and predict structure jointly, rather than assuming a known antibody backbone. IgGM is built on a consistency model distilled from a diffusion model, so it can generate in a single step or a few steps.
Attribute | Details |
|---|---|
Task | CDR sequence and structure co-design plus complex prediction |
Inputs | Antigen sequence and structure, FR sequences, optional epitope |
Formats | Antibodies and nanobodies |
Core technique | Consistency model distilled from a diffusion model |
Sequence features | Frozen ESM-PPI protein language model |
Sampling | 10 steps chosen as a quality and speed balance |
How IgGM Works: Diffusion Training and Consistency Distillation
Multi-level encoder: an inter-chain embedding module, a structure encoder using discretized inter-residue distances, and a 16-block Sgformer fuse the features. Epitope information is encoded into single and pair representations.
Structure prediction module: 8 layers of invariant point attention refine coordinates, with a separate sequence head.
Two training phases: a diffusion phase learns to recover antibody structure from noised complexes, then a sequence phase trains CDR recovery with cross-entropy.
Consistency distillation: the final model is distilled for fast generation.
Candidate selection: generated candidates are ranked using DockQ. Initial structures can be replaced, for example with AlphaFold 3 predictions.
IgGM Benchmark Results on Antibodies and Nanobodies
The authors built time-split test sets from SAbDab with 60 antibodies and 27 nanobodies, removing sequences highly similar to the training data.
Metric | Result |
|---|---|
Antibody docking success rate (complex prediction) | 0.4667 for IgGM vs 0.067 for dyMEAN |
De novo antibody design docking success rate | 43.3% |
CDR-H3 sequence recovery | 36%, a 22.4% relative improvement over dyMEAN |
Nanobody complex prediction success rate | 44% |
Nanobody design success rate | 0.415 for IgGM vs 0.346 for DiffAb (AF3) |
Using AlphaFold 3 structures as initialization improved docking success by nearly 20%, with a small decline in CDR sequence and structure metrics. IgGM was the only method with iRMS below 8 and LRMS below 20 in the design benchmark.
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 IgGM on Tamarind Bio
De novo antibody design: generate CDR sequences and structures for a new antigen when no reference antibody exists.
Nanobody design: design single-domain binders, including against different epitopes of the same antigen as the authors illustrated.
Complex structure prediction: predict antibody-antigen complexes, including correcting inaccurate AlphaFold 3 predictions as the authors show.
Targeted CDR design: design specific CDRs, such as CDR H3, while keeping the framework fixed.
How to Use IgGM on Tamarind Bio
Open the tool: log in to tamarind.bio and open IgGM.
Provide the antigen: upload the antigen structure and sequence.
Provide framework sequences: enter the antibody or nanobody FR sequences with the CDRs to be designed marked, if exposed.
Define the epitope: specify the binding region, if exposed.
Set generation options: choose the number of samples and sampling steps, if exposed. The authors used 10 steps.
Run the job: submit and wait for completion.
Download and rank: download the designed complexes and review the DockQ-based ranking if provided.
Validate: take top candidates to experimental binding assays.
Things to Keep in Mind
Existing in silico metrics do not reliably show whether designed antibodies bind their antigens, so wet-lab validation is essential.
Evaluation relies on comparison with existing antibodies, which are unavailable for entirely new antigens.
AlphaFold 3 initialization improves docking but slightly reduces CDR sequence and structure metrics.
Nanobody CDR recovery and docking are weaker than for antibodies, which the authors attribute to longer CDRs and flexible binding modes.
Framework sequences are not designed.
Source: IgGM preprint on bioRxiv, DOI 10.1101/2024.09.19.613838 (https://www.biorxiv.org/content/10.1101/2024.09.19.613838v2). No code repository was listed on the page read.