用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/InternScience/Agents-A1 --skill molclaw-prolif-pdb命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | molclaw-prolif-pdb |
| description | ProLIF static complex analysis skill for a single protein-ligand structure. |
| license | MIT license |
| metadata | {"skill-author":"PJLab"} |
[!NOTE] Local files are not directly accessible by the server. Please upload them to the server using
drugsda-file-transferbefore execution. For PDB file inputs, it is recommended to preprocess them usingdrugsda-fix_pdbbefore execution.
Analyze interaction fingerprints from one static protein-ligand complex structure. Use this skill for fast assessment of crystal structures, top docking poses, or representative MD frames.
| Parameter | Source Guidance |
|---|---|
structure_path | Can come from PDB retrieval tools, best docking poses, MD frame extraction (e.g., openmm_extract_frames), or complex preparation tools (e.g., prepare_complex) outputting complex PDB files |
ligand_selection | User-defined ligand selection string that matches ligand identifiers in the structure file |
protein_selection | Defaults to protein; can be customized to limit the analyzed region |
prolif_pdbAnalyze a single complex structure and return ProLIF interaction fingerprints or counts with summary metrics.
Args:
structure_path (str): Path to the complex structure file (commonly PDB).
ligand_selection (str): Selection string identifying ligand atoms.
protein_selection (str): Selection string for protein atoms. Default: 'protein'.
interactions (List[str]|None): Optional interaction types to compute.
count (bool): If True, compute interaction counts instead of fingerprints. Default: False.
vicinity_cutoff (float|None): Optional distance cutoff for vicinity interactions.
params_json (str|None): Optional JSON parameter file path for ProLIF interaction settings.
Return:
status (str): 'success' or 'error'.
msg (str): Human-readable summary or error message.
command (str): The executed command label ('pdb').
output_dir (str|None): Run-specific directory under tool_result/prolif_result.
output_file (str|None): Path to the produced CSV file.
n_frames (int|None): Number of processed frames (typically 1 for static structures).
n_interactions (int|None): Number of interaction columns in output.
frequent_interactions (List[dict]|None): High-frequency interactions (>30%) with keys 'interaction' and 'frequency'.
result_summary (dict|None): Full summary dictionary from the wrapper.
prolif_pdbresponse = await client.session.call_tool(
"prolif_pdb",
arguments={
"structure_path": "relative/path/to/complex.pdb",
"ligand_selection": "resname LIG",
"protein_selection": "protein",
"interactions": ["Hydrophobic", "HBAcceptor"]
}
)
result = client.parse_result(response)
key_output = result["output_file"]
# 1) Main mode
{
"structure_path": "relative/path/to/complex.pdb",
"ligand_selection": "resname LIG",
"protein_selection": "protein",
"interactions": ["Hydrophobic", "HBDonor"]
}
# 2) Variant mode
{
"structure_path": "relative/path/to/complex.pdb",
"ligand_selection": "resname LIG",
"count": True,
"params_json": "relative/path/to/prolif_override.json"
}
ProLIF reports residue identifiers using the numbering of the input PDB file. If the PDB was generated by a prediction tool (ESMFold, Boltz-2, Chai-1), these numbers are tool-internal sequential numbers (1, 2, 3...) — NOT UniProt numbers.
Before interpreting ProLIF results when the task references specific residues (e.g., "confirm Met793 interaction"):
molclaw-residue-mapper to build a mapping table.Common catastrophic error: ProLIF reports "HBDonor at MET76" from a Boltz-2 structure. Agent searches for "MET793" in ProLIF output, does not find it, and concludes "Met793 interaction is absent." In reality, MET76 (Boltz-2 internal) IS Met793 (UniProt). Use residue_mapper with query="tool:76" to verify.
After ProLIF analysis, download ALL visualization outputs (interaction heatmaps, frequency barplots, etc.) from the output directory using server_file_to_base64. These are Category A files essential for result communication and user verification.
Predict the ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of the input molecules.
Predict binding affinity between target protein sequence and small molecule SMILES using Boltz-2.
Predict protein structures with Chai-1 from sequence or FASTA input and return model scoring summaries.
基于 SOC 职业分类