OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Dymean

Dymean is an open-source Python research codebase for end-to-end full-atom antibody design, with tasks for CDR design, complex structure prediction and affinity optimization.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Dymean — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Dymean when an antibody research project needs code and checkpoints for a defined design, structure-prediction or affinity-optimization task and can support local data and scorer setup.

Who it is for

  • Researchers exploring antibody sequence and structure design.
  • Python users building experiments around antibody CDR design, complex prediction or affinity optimization.

Documented uses

  • Prototype CDR-H3 or multi-CDR antibody design.
  • Predict antibody–antigen complex structures.
  • Explore affinity optimization for antibody CDRs.
  • Use design and optimize APIs from Python code.
Before you choose

Requirements & limits

The repository is public and specifies an MIT license. Its documented workflow uses a local Conda environment.

  • Setup requires a Conda environment, antibody PDB data and TMscore and DockQ scorers.
  • The README identifies IMGT-renumbered SAbDab structures as the data used in the paper.
  • Released checkpoints cover CDR-H3 design, complex structure prediction, affinity optimization, multi-CDR design and full-variable-domain design.
  • Optional pipeline baselines require their own dependencies and configured paths.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare dyMEAN prerequisites and antibody structure data

Set up the documented runtime environment, evaluation scorers, and IMGT-renumbered antibody PDB data for dyMEAN experiments.

Before you start

  • The dyMEAN repository, including its env.yml, evaluation/TMscore.cpp, and configs.py.
  • Conda and a C++ compiler.
  • The DockQ project and its prerequisites, prepared according to DockQ's instructions.
  • Antibody structure data downloaded from SAbDab.

Steps

  1. Create the conda environment with conda env create -f env.yml.

  2. Compile TMscore with g++ -static -O3 -ffast-math -lm -o evaluation/TMscore evaluation/TMscore.cpp.

  3. Clone DockQ, compile its prerequisites as instructed by that project, and set DOCKQ_DIR in configs.py to the DockQ directory.

  4. Download and decompress the SAbDab structure archive; use the all_structures/imgt PDB data for the documented IMGT workflow.

Expected outcome

The documented environment, scorers, and IMGT antibody structure files are prepared as prerequisites for dyMEAN experiments.

Guide limitations

  • The repository describes these as prerequisites, not a complete experiment or model execution workflow.
  • The DockQ prerequisite compilation details are delegated to DockQ's own instructions.
  • SAbDab data is updated weekly; the repository-provided summary is from November 12, 2022.
Documentation 1 ↗
Source-based practical guide

Generate processed benchmark data from IMGT antibody structures

Run the repository's documented preprocessing script to transform IMGT-renumbered PDB data into processed benchmark data.

Before you start

  • Prepared IMGT-renumbered PDB files in all_structures/imgt/.
  • The dyMEAN repository and the prerequisites described in its Setup section.

Steps

  1. Ensure the IMGT-renumbered PDB files are located at all_structures/imgt/.

  2. Run the command below.

    Documented command
    bash scripts/data_preprocess.sh all_structures/imgt all_data

  3. Use the generated data under all_data as the processed dataset output.

Expected outcome

The script processes SAbDab, RAbD, the Igfold test set, and SKEMPI V2.0 into the specified output directory; the repository estimates about 5 GB of processed data.

Guide limitations

  • The README estimates preprocessing takes about one hour.
  • Errors may be reported for incorrectly annotated or formatted antibody structures; these are dropped during data cleaning.
  • The guide documents preprocessing only, not training or inference.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
THUNLP-MT/dyMEAN · Official homepage title · retained in its source language
How we review sourcesSuggest a correction