README.md

---
license: cc-by-nc-sa-4.0
tasks:
  - protein-complex-structure-prediction
frameworks:
  - jax
language:
  - en
  - zh
tags:
  - OneScience
  - Life Sciences
  - Protein Structure Prediction
  - Biomolecular Interactions
  - Complex Structure Prediction
  - AlphaFold3
datasets:
  - OneScience/AlphaFold3_dataset
---

<p align="center">
  <strong>
    <span style="font-size: 30px;">AlphaFold3</span>
  </strong>
</p>

# Model Introduction

AlphaFold3 is a biomolecular structure prediction model proposed by Google DeepMind and Isomorphic Labs. It can predict the three-dimensional structures and interactions of molecules and their complexes, including proteins, DNA, RNA, and small-molecule ligands.

Paper: Accurate structure prediction of biomolecular interactions with AlphaFold 3  
https://www.nature.com/articles/s41586-024-07487-w

# Model Description

AlphaFold3 uses Pairformer and diffusion models to predict biomolecular complex structures. This model package provides a JAX / Flax inference project and data search scripts, and is released together with the ModelScope dataset `OneScience/AlphaFold3_dataset`.

# Applicable Scenarios

| Scenario | Description |
| :---: | :--- |
| Direct inference with existing features | Input an AlphaFold3 JSON containing features such as MSA / template, and output structure prediction results |
| Protein structure prediction | Input a protein sequence, generate features together with search databases, and predict the structure |
| Biomolecular complex modeling | Input multi-component objects such as proteins, DNA, RNA, and ligands, and predict the spatial conformation of the complex |
| Data search workflow verification | Use Jackhmmer / Nhmmer or MMseqs workflows to check database paths and search tool connectivity |
| ModelScope / OneCode runtime | After downloading the model project and complete dataset, quickly verify script connectivity in a biology-domain runtime environment |



# Usage Instructions

## 1. OneCode Usage

You can experience intelligent one-click AI4S programming through the OneCode online environment:

[Click to experience intelligent one-click AI4S programming](https://web-2069360198568017922-iaaj.ksai.scnet.cn:58043/home)

## 2. Manual Installation and Usage

**Hardware Requirements**

- GPU or DCU runtime is recommended.
- CPU can be used for import checks and small-configuration connectivity verification, but full training and inference are slow.
- DCU users need to install DTK in advance. DTK 25.04.2 or later is recommended, or the OneScience-recommended version matching the current cluster.





**Environment Check**

- NVIDIA GPU:

```bash
nvidia-smi
```

- Hygon DCU:

```bash
hy-smi
```

### Download the Model Package

```bash
modelscope download --model OneScience/AlphaFold3 --local_dir ./AlphaFold3
cd AlphaFold3
```

### Install the Runtime Environment

**DCU Environment**

```bash
# First activate DTK and Conda
conda create -n onescience311 python=3.11 -y
conda activate onescience311
# Supports uv installation
pip install onescience[bio-dcu] -i http://mirrors.onescience.ai:3141/pypi/simple/  --trusted-host mirrors.onescience.ai
```

After installation, return to the model package directory:

```bash
cd ./AlphaFold3
```

If the current environment has not yet built the AlphaFold3 C++ extension and runtime data files, execute:

```bash
python -m onescience.flax_model.alphafold3.build_extension
python -m onescience.flax_models.alphafold3.build_data
```

### Training and Inference Data Introduction

The OneScience community has uploaded the complete data required for AlphaFold3 inference and data search to ModelScope: [OneScience/AlphaFold3_dataset](https://modelscope.cn/datasets/OneScience/AlphaFold3_dataset). This model package does not include a training entry point; this dataset is mainly used for MSA / template feature construction and pre-inference data search.

```bash
modelscope download --dataset OneScience/AlphaFold3_dataset --local_dir ./data/alphafold3
```
### Training Weights

Weights will be uploaded soon.

### Prepare Weights

Place the AlphaFold3 model weights in the following directory, or specify them through an environment variable:

```text
weight/
  AlphaFold3/
    ...
```

The default lookup order is:

- `ALPHAFOLD3_MODEL_DIR`
- `${ONESCIENCE_MODELS_DIR}/AlphaFold3`
- `weight/AlphaFold3`

Example:

```bash
export ALPHAFOLD3_MODEL_DIR=/path/to/AlphaFold3
```

### Direct Inference

When the input JSON already contains features such as MSA and template, you can run directly:

```bash
bash scripts/infer.sh
```

Equivalent Python command example:

```bash
python scripts/run_alphafold.py \
  --json_path inputs/7r6r_data.json \
  --model_dir weight/AlphaFold3 \
  --output_dir outputs \
  --run_data_pipeline=false \
  --flash_attention_implementation=triton
```

The output will be written to `outputs/`, including the best structure, structure results for different seeds / samples, the ranking score CSV, and a copy of the input JSON.

### Jackhmmer / Nhmmer Data Search

When the input JSON contains only sequences and requires local database search, use:

```bash
bash scripts/infer_jackhmmer.sh
```

Common environment variables:

```bash
export ALPHAFOLD3_DATASET_ROOT=/path/to/alphafold3
export ALPHAFOLD3_MODEL_DIR=/path/to/AlphaFold3
export ALPHAFOLD3_JSON_PATH=inputs/t1119_search.json
export ALPHAFOLD3_OUTPUT_DIR=outputs
export ALPHAFOLD3_RUN_INFERENCE=false
```

Here, `ALPHAFOLD3_DATASET_ROOT` is expected by default to contain database directories such as `public_databases/`, `jackhmmer_split/`, and `mmseqsDB/`.

### MMseqs Data Search

If the runtime environment provides the MMseqs program and MMseqs database, use:

```bash
bash scripts/infer_mmseqs.sh
```

Common environment variables:

```bash
export ALPHAFOLD3_MMSEQS_HOME=/path/to/mmseqs
export ALPHAFOLD3_DATASET_ROOT=/path/to/alphafold3
export ALPHAFOLD3_MMSEQS_DB_DIR=/path/to/alphafold3/mmseqsDB
export ALPHAFOLD3_RUN_INFERENCE=false
```

To continue inference after searching, set `ALPHAFOLD3_RUN_INFERENCE` to `true` and make sure the weight directory is available.

# Data Format

AlphaFold3 input uses JSON format. The basic structure is as follows:

```json
{
  "dialect": "alphafold3",
  "version": 1,
  "name": "example",
  "sequences": [
    {
      "protein": {
        "id": "A",
        "sequence": "..."
      }
    }
  ],
  "modelSeeds": [100],
  "bondedAtomPairs": null,
  "userCCD": null
}
```

This repository provides two examples:

- `inputs/7r6r_data.json`: Contains information such as sequence, MSA, and template, and is suitable for direct inference.
- `inputs/t1119_search.json`: Contains only sequence and is suitable for data search workflow verification.

The complete ModelScope dataset `OneScience/AlphaFold3_dataset` is recommended to be downloaded to `data/alphafold3/` under the model package. The relative structure read by the data search workflow by default is as follows:

```text
data/
  alphafold3/
    public_databases/
      mmcif_files/
      pdb_seqres_2022_09_28.fasta
      ...
    jackhmmer_split/
      bfd-first_non_consensus_sequences.fasta@64
      mgy_clusters_2022_05.fa@512
      uniprot_cluster_annot_2021_04.fa@256
      uniref90_2022_05.fa@128
    mmseqsDB/
      small_bfd_db
      mgnify_db
      uniprot_cluster_annot_db
      uniref90_db
```

# Verification

Static import check:

```bash
python tests/check_import_boundaries.py
```

# Official OneScience Information

| Platform | OneScience Main Repository | Skills Repository |
| --- | --- | --- |
| Gitee | https://gitee.com/onescience-ai/onescience | https://gitee.com/onescience-ai/oneskills |
| GitHub | https://github.com/onescience-ai/OneScience | https://github.com/onescience-ai/oneskills |

# Citations and License

- This repository is based on the AlphaFold3 open-source model and provides DCU adaptation.
- AlphaFold3 source code uses the CC BY-NC-SA 4.0 license; model parameters are subject to separate terms of use.
- For scientific research, cite the original paper: [Accurate structure prediction of biomolecular interactions with AlphaFold 3](https://www.nature.com/articles/s41586-024-07487-w).