🔥 Web Server | 📄 Paper
We provide the environment configurations for cuda 12.4 + pytorch 2.5.1 (env_cuda124.yaml), which can be installed by the following command using conda:
conda update conda # prevent conda from stucking in "resolving environment"
conda env create -f ./env_cuda124.yamlAfter installation, please activate the environment by the following command:
conda activate AnewOmniThe trained model checkpoint (model.ckpt) is provided at github release. Please download it into the checkpoints folder.
cd checkpoints
wget https://github.com/bytedance/AnewOmni/releases/download/init/model.ckptWe provide a universal entry point to run inference on different binder modalities by specifying the configuration (.yaml) and the output folder as follows:
python -m api.generate_with_rank --config /path/to/config.yaml --save_dir /path/for/outputThe example of a config.yaml is as below:
dataset:
# Multiple references can be provided. You just need to add the corresponding information in a new line under each of pdb_paths, tgt_chains, and lig_chains.
pdb_paths:
- ./demo/data/8u4r_chothia.pdb # Path to the reference complex. Both pdb files and mmcif files are supported
tgt_chains:
- R # The target proteins only include one chain, that is R.
lig_chains:
- HL # The reference binder, which is an antibody, includes two chains, H and L. For antibody cdr design, the program assumes the heavy chain goes before the light chain, which means it will regard H as the heavy chain and L as the light chain in this case.
templates:
# Here we specify the type of binders we want to design
- class: LinearPeptide # Linear peptides with at lengths between 10 to 12, left inclusive and right exclusive.
size_min: 10
size_max: 12
- class: Molecule # Small molecules. By default the number of blocks is sampled according to the spatial size of the binding site. You can also specify "size_min" and/or "size_max" to control the threshold of the sampled number of blocks.
batch_size: 8 # Batch size. Adjust it according to your GPU memory available. Batch size = 8 can be run under 12G memory for most cases.
n_samples: 20 # Number of generations. On each target provided, the model will generate 20 candidates for each template.
patience: 100 # the generation stops if 100 samples are generated yet the top-k have not changedIdeally, you can include a reference binder (such as the native protein-protein interaction partner) in the provided structure file. In this case, the algorithm will choose the residues within 10
dataset:
pdb_paths:
- ./demo/data/8u4r_chothia.pdb
tgt_chains:
- R
hotspots:
- [[R, 188, ''], [R, 288, ''], [R, 192, ''], [R, 196, '']] # each hotspot is represented as [chain id, residue number, insertion code]We have also provided demo configurations for common binder modalities in the Demo section.
| Binder Modality | Template Class | Configuration | Description |
|---|---|---|---|
| Small Molecule | Molecule |
demo/small_molecule.yaml |
The range of number of blocks can be either specified or automatically sampled based on the spatial volume of the binding site. |
| Linear Peptide | LinearPeptide |
demo/linear_pep.yaml |
Minimum and maximum of lengths need to be specified. |
| Disulfide Cyclic Peptide | DiSulfidePeptide |
demo/disulfide_cyc_pep.yaml |
Prompting generations with cysteines at head and tail, as well as a disulfide bond between them. |
| Head-to-Tail Cyclic Peptide | HeadTailPeptide |
demo/headtail_cyc_pep.yaml |
Prompting generations with amide bond between head and tail. |
| Antibody Single CDR | AntibodySingleCDR |
demo/antibody_single_cdr.yaml |
Antibody frameworks should be provided with Chothia numbering system, and the CDR type to be designed need to be specified. |
| Antibody Multiple CDRs | AntibodyMultipleCDR |
demo/antibody_multiple_cdrs.yaml |
Antibody frameworks should be provided with Chothia numbering system, and the CDR type to be designed need to be specified. Change of the CDR lengths is also supported. |
Arguments
Molecule:
size_min(default:None): Minimum number of blocks in the generated small molecule.size_max(default:None): Maximum number of blocks in the generated small molecule.
LinearPeptide:
-
size_min($\geq 4$ , default: 8): Minimum number of residues in the generated peptide, inclusive of this value. -
size_max($\leq 25$ , default: 13): Maximum number of residues in the generated peptide, exclusive of this value.
DiSulfidePeptide and HeadTailPeptide:
-
size_min($\geq 4$ , default: 8): Minimum number of residues in the generated peptide, inclusive of this value. -
size_max($\leq 25$ , default: 13): Maximum number of residues in the generated peptide, exclusive of this value. -
w(default: 1.0): Strength of the graph prompts implemented as the weight for classifier-free guidance. Larger value indicates better alignment with the control, but a value$<3.0$ is recommended.
AntibodySingleCDR
cdr_type(default:HCDR3): The type of CDR to design. Combinations of heavy (H) / light (L) and the number of CDRs (1/2/3) are supported.
AntibodyMultipleCDR
cdr_types(required): The type of CDR to design. Combinations of heavy (H) / light (L) and the number of CDRs (1/2/3) are supported (e.g.,[LCDR3, HCDR3]).length_ranges(default:{}): Range of lengths for each CDR type (e.g.,{ "HCDR3": [10, 12] }), with both end inclusive. If not specified, the length will be set to the same as the native CDR on the framework.
In addition to the config-driven command-line entry points, AnewOmni also provides an interactive local UI for prompt-based generation and result inspection.
The UI is intended as a lightweight proof-of-concept layer on top of the existing backend. It is useful when you want to:
- prototype small-molecule, peptide, cyclic-peptide, and antibody prompts interactively
- inspect prompt state before launch
- submit generation tasks from a browser
- review filtered results, 2D SVGs, and 3D structures in one place
- iterate on prompt constraints without writing YAML configs first
- built-in demos for molecule, peptide, cyclic peptide, and antibody workflows
- prompt-programming interface instead of pure config editing
- task submission and task switching in the browser
- predefined physiochemical filtering with quota retries and fallback selection
- result inspection with top-record summaries, 2D rendering, and Mol* 3D viewing
Start the web UI from the repository root:
ANEW_UI_PORT=8766 python -m api.ui.webThen open:
http://127.0.0.1:8766
If ANEW_UI_PORT is not set, the default port is 8765.
You can follow these instructions to run simple demos:
- Select a demo modality from the dropdown menu (molecule, peptide, cyclic peptide, antibody).
- Adjust the prompt program in the Program panel if needed (see UI User Guide and Prompt Language Design for detailed grammar and architecture of the programming language).
- Click Execute to preview/validate the current prompt state.
- If the preview looks correct, click Run Generation to submit a generation task.
- Monitor progress in the Program output box (it shows the current task log), and review results in the Latest Result panel after the task finishes.
For command-line interaction without graphic user interface (python-interpreter-like), start the REPL from the repository root:
python -m api.ui.replFor the detailed UI documentation, see:
If you find the models useful in your research, please cite the relevant paper:
@article{kong2026programming,
title={Programming Biomolecular Interactions with All-Atom Generative Model},
author={Kong, Xiangzhe and Chen, Junwei and Zhang, Ziting and Li, Gaodeng and Zhu, Qingyuan and Wei, Lei and Li, Mingyu and Shi, Yan and Dai, Weiyang and Zhang, Zishen and others},
journal={bioRxiv},
pages={2026--03},
year={2026},
publisher={Cold Spring Harbor Laboratory}
}@inproceedings{
kong2025unimomo,
title={UniMoMo: Unified Generative Modeling of 3D Molecules for De Novo Binder Design},
author={Kong, Xiangzhe and Zhang, Zishen and Zhang, Ziting and Jiao, Rui and Ma, Jianzhu and Liu, Kai and Huang, Wenbing and Liu, Yang},
booktitle={Forty-second International Conference on Machine Learning},
year={2025}
}This project is licensed under the MIT License.
Thank you for your interest in our work!
Please feel free to ask about any questions about the algorithms, codes, as well as problems encountered in running them so that we can make it clearer and better. You can either create an issue in the github repo or contact us at anewbt_mind@bytedance.com.