rfdiffusion-nim

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

RFDiffusion NIM

Design protein backbone PDBs for de novo proteins, motif scaffolds, and binders. Use this

SKILL.md

for first-pass hosted/local usage; load supplemental files only when needed:

```
references/api.md
```
: exact endpoints, schemas, Docker flags, response fields.
```
references/science.md
```
: design modes, strengths, limits, and handoffs.
```
references/parameters.md
```
: contigs, hotspots, steps, and seeds.
```
references/validation.md
```
: PDB, contig, and artifact sanity checks.
```
references/examples.md
```
: compact hosted/local request patterns.

为从头设计的蛋白质、基序支架和结合体设计蛋白质骨架PDB文件。使用本

SKILL.md

进行首次托管/本地使用；仅在需要时加载补充文件：

```
references/api.md
```
：精确的端点、模式、Docker参数、响应字段。
```
references/science.md
```
：设计模式、优势、局限性及交接说明。
```
references/parameters.md
```
：contigs、热点残基、扩散步骤及随机种子。
```
references/validation.md
```
：PDB文件、contigs及产物合理性检查。
```
references/examples.md
```
：简洁的托管/本地请求示例。

Choose Mode

选择模式

Ask only when context is unclear:

Hosted NVIDIA API or local Docker NIM?

Hosted:

https://health.api.nvidia.com/v1/biology/ipd/rfdiffusion/generate

Local:

http://localhost:8000/biology/ipd/rfdiffusion/generate

Local inference paths do not include

/v1/

. Hosted requests use

Authorization: Bearer $NGC_API_KEY

. Supported local Docker startup uses

NGC_API_KEY

(or

NVIDIA_API_KEY

via the preflight) for registry login, entitlement checks, and first-run model downloads; pass it into the container with

-e NGC_API_KEY

. Local inference requests use no auth header after readiness. Warm-cache key-free startup varies by image/version and should not be assumed.

仅在上下文不明确时询问：

使用NVIDIA托管API还是本地Docker NIM？

托管版：

https://health.api.nvidia.com/v1/biology/ipd/rfdiffusion/generate

本地版：

http://localhost:8000/biology/ipd/rfdiffusion/generate

本地推理路径不包含

/v1/

。托管请求需使用

Authorization: Bearer $NGC_API_KEY

。本地Docker启动需使用

NGC_API_KEY

（或通过预检查使用

NVIDIA_API_KEY

）进行镜像仓库登录、权限验证及首次运行时的模型下载；需通过

-e NGC_API_KEY

将其传入容器。本地推理请求在就绪后无需认证头。无缓存密钥的预热启动方式因镜像/版本而异，不应默认依赖。

Local Docker

本地Docker部署

For local setup answers, copy the preflight below exactly before

docker login

docker run

, readiness, and the no-auth local request. Do not replace it with a simple

: "${NGC_API_KEY:?Set NGC_API_KEY}"

check, do not invent a cache default, and do not drop the

NVIDIA_API_KEY

fallback. Default setup is single GPU

device=0

bash

set -a
[ -f .env ] && . ./.env
set +a

if [ -z "${NGC_API_KEY:-}" ] && [ -n "${NVIDIA_API_KEY:-}" ]; then
  export NGC_API_KEY="$NVIDIA_API_KEY"
fi
: "${NGC_API_KEY:?Set NGC_API_KEY or NVIDIA_API_KEY}"
: "${LOCAL_NIM_CACHE:?Set LOCAL_NIM_CACHE}"

echo "$NGC_API_KEY" | docker login nvcr.io --username '$oauthtoken' --password-stdin

mkdir -p "${LOCAL_NIM_CACHE}"
chmod 755 "${LOCAL_NIM_CACHE}"

docker run -it \
  --runtime=nvidia \
  --gpus "device=0" \
  -e NGC_API_KEY \
  -v "${LOCAL_NIM_CACHE}:/opt/nim/.cache" \
  -p 8000:8000 \
  nvcr.io/nim/ipd/rfdiffusion:2

Readiness:

bash

until curl -sf http://localhost:8000/v1/health/ready; do sleep 5; done

如需本地设置相关解答，请在

docker login

、

docker run

、就绪检查及无认证本地请求前，严格复制以下预检查脚本。请勿将其替换为简单的

: "${NGC_API_KEY:?Set NGC_API_KEY}"

检查，请勿自行设置缓存默认值，请勿省略

NVIDIA_API_KEY

回退选项。默认设置为单GPU

device=0

。

bash

set -a
[ -f .env ] && . ./.env
set +a

if [ -z "${NGC_API_KEY:-}" ] && [ -n "${NVIDIA_API_KEY:-}" ]; then
  export NGC_API_KEY="$NVIDIA_API_KEY"
fi
: "${NGC_API_KEY:?Set NGC_API_KEY or NVIDIA_API_KEY}"
: "${LOCAL_NIM_CACHE:?Set LOCAL_NIM_CACHE}"

echo "$NGC_API_KEY" | docker login nvcr.io --username '$oauthtoken' --password-stdin

mkdir -p "${LOCAL_NIM_CACHE}"
chmod 755 "${LOCAL_NIM_CACHE}"

docker run -it \
  --runtime=nvidia \
  --gpus "device=0" \
  -e NGC_API_KEY \
  -v "${LOCAL_NIM_CACHE}:/opt/nim/.cache" \
  -p 8000:8000 \
  nvcr.io/nim/ipd/rfdiffusion:2

就绪检查：

bash

until curl -sf http://localhost:8000/v1/health/ready; do sleep 5; done

Contigs DSL

Contigs领域特定语言（DSL）

contigs

defines what to keep and what to generate. For the full pattern syntax (fixed length, ranges, kept chain segments, chain breaks), see

references/api.md

under Contigs Language Reference.

Design modes:

De novo:
```
contigs="80-120"
```
; live hosted validation requires a non-empty
```
input_pdb
```
or
```
input_pdb_asset
```
, so inline requests should include the dummy PDB below.
Motif scaffolding: read
```
target.pdb
```
, pass
```
input_pdb
```
, use a contig like
```
"A25-35/0 50-80"
```
.
Binder design: pass target
```
input_pdb
```
, contig with target and binder segment, and
```
hotspot_res=["A50", "A51", ...]
```
in ChainResidue string format.

python

DUMMY_PDB = (
    "CRYST1    1.000    1.000    1.000  90.00  90.00  90.00 P 1           1\n"
    "ATOM      1  CA  ALA A   1       0.000   0.000   0.000  1.00  0.00           C\n"
    "END\n"
)

contigs

用于定义保留和生成的区域。完整的模式语法（固定长度、范围、保留链段、链断裂）请参见

references/api.md

中的Contigs语言参考部分。

设计模式：

从头设计：
```
contigs="80-120"
```
；托管版实时验证要求非空的
```
input_pdb
```
或
```
input_pdb_asset
```
，因此在线请求需包含以下虚拟PDB文件。
基序支架设计：读取
```
target.pdb
```
，传入
```
input_pdb
```
，使用类似
```
"A25-35/0 50-80"
```
的contigs参数。
结合体设计：传入目标
```
input_pdb
```
，包含目标和结合体片段的contigs参数，以及ChainResidue字符串格式的
```
hotspot_res=["A50", "A51", ...]
```
。

python

DUMMY_PDB = (
    "CRYST1    1.000    1.000    1.000  90.00  90.00  90.00 P 1           1\n"
    "ATOM      1  CA  ALA A   1       0.000   0.000   0.000  1.00  0.00           C\n"
    "END\n"
)

Request Pattern

请求示例

python

import os
from pathlib import Path
import requests

HOSTED = True
url = (
    "https://health.api.nvidia.com/v1/biology/ipd/rfdiffusion/generate"
    if HOSTED else "http://localhost:8000/biology/ipd/rfdiffusion/generate"
)
headers = {"Content-Type": "application/json"}
if HOSTED:
    headers["Authorization"] = f"Bearer {os.getenv('NGC_API_KEY')}"

payload = {
    "input_pdb": DUMMY_PDB,
    "contigs": "80-120",
    "diffusion_steps": 50,
}
response = requests.post(url, headers=headers, json=payload, timeout=300)
response.raise_for_status()
result = response.json()
Path("designed_backbone.pdb").write_text(result["output_pdb"])

Motif scaffold:

python

payload = {
    "input_pdb": Path("target.pdb").read_text(),
    "contigs": "A25-35/0 50-80",
    "diffusion_steps": 50,
}

Binder design:

python

payload = {
    "input_pdb": Path("target.pdb").read_text(),
    "contigs": "A1-100/0 50-100",
    "hotspot_res": ["A50", "A51", "A52", "A53", "A54"],
    "diffusion_steps": 50,
}

python

import os
from pathlib import Path
import requests

HOSTED = True
url = (
    "https://health.api.nvidia.com/v1/biology/ipd/rfdiffusion/generate"
    if HOSTED else "http://localhost:8000/biology/ipd/rfdiffusion/generate"
)
headers = {"Content-Type": "application/json"}
if HOSTED:
    headers["Authorization"] = f"Bearer {os.getenv('NGC_API_KEY')}"

payload = {
    "input_pdb": DUMMY_PDB,
    "contigs": "80-120",
    "diffusion_steps": 50,
}
response = requests.post(url, headers=headers, json=payload, timeout=300)
response.raise_for_status()
result = response.json()
Path("designed_backbone.pdb").write_text(result["output_pdb"])

基序支架设计请求：

python

payload = {
    "input_pdb": Path("target.pdb").read_text(),
    "contigs": "A25-35/0 50-80",
    "diffusion_steps": 50,
}

结合体设计请求：

python

payload = {
    "input_pdb": Path("target.pdb").read_text(),
    "contigs": "A1-100/0 50-100",
    "hotspot_res": ["A50", "A51", "A52", "A53", "A54"],
    "diffusion_steps": 50,
}

Save And Interpret Output

保存与解读输出

Save

result["output_pdb"]

as a PDB artifact and report

elapsed_ms

when present. Generated backbones are not final proteins; feed them to ProteinMPNN for sequence design, then validate sequences/structures with Boltz2 or OpenFold3. For PDB and contig checks, read

references/validation.md

将

result["output_pdb"]

保存为PDB产物，若存在

elapsed_ms

则报告该值。生成的骨架并非最终蛋白质；需将其输入ProteinMPNN进行序列设计，随后使用Boltz2或OpenFold3验证序列/结构。有关PDB文件和contigs的检查，请参阅

references/validation.md

。

Limits And Troubleshooting

限制与故障排除

```
diffusion_steps
```
: 1-50; 50 is maximum quality, fewer is faster.
Single GPU; minimum GPU VRAM is about 12 GB.
```
hotspot_res
```
uses strings like
```
"A50"
```
, not tuples.
```
422
```
usually means chain IDs in
```
contigs
```
/
```
hotspot_res
```
do not match
```
input_pdb
```
, a malformed contig, or omitted
```
input_pdb
```
for hosted de novo.
Local URL 404 usually means an accidental
```
/v1/
```
prefix.

```
diffusion_steps
```
：取值1-50；50为最高质量，数值越少速度越快。
仅支持单GPU；最低GPU显存约为12 GB。
```
hotspot_res
```
需使用类似
```
"A50"
```
的字符串格式，而非元组。
错误码
```
422
```
通常意味着
```
contigs
```
/
```
hotspot_res
```
中的链ID与
```
input_pdb
```
不匹配、contigs格式错误，或托管版从头设计时遗漏了
```
input_pdb
```
。
本地URL返回404通常是因为意外添加了
```
/v1/
```
前缀。