Network Toxicology + Molecular Docking Research Planner
Generates a complete network toxicology + molecular docking study design from a user-provided toxicant and disease/phenotype. Always outputs four workload configurations and a recommended primary plan.
Input Validation
This skill accepts: a toxicant (environmental chemical, endocrine disruptor, heavy metal, food contaminant, pharmaceutical residue, or consumer product chemical) paired with a disease or phenotype, for which the user wants to generate a network toxicology + molecular docking research design.
If the user's request does not involve a toxicant–disease pair for network toxicology research design — for example, asking to execute a STRING query, download GEO datasets, write production code, answer a clinical pharmacology question, or design a non-toxicology study — do not proceed with the workflow. Instead respond:
"Network Toxicology + Molecular Docking Research Planner is designed to generate computational research designs for toxicant–disease mechanism studies. Please provide a toxicant and a disease or phenotype. If you want to run the analysis directly, use a data-execution tool; if you need a different study type, use the appropriate planner skill."
Minimum required input: one toxicant + one disease or phenotype.
If workload is unspecified, default to: Standard as primary · Lite as minimal · Advanced as upgrade.
Step 1 — Infer Study Context
Read → INLINECODE0
Identify: toxicant class · disease type · whether docking is central or supportive · validation feasibility · resource constraints · publication ambition · whether input involves multiple toxicants (→ Pattern F in Step 2).
Step 2 — Select Study Pattern
Read → INLINECODE1
Match to one of six canonical design styles (A–F). State which pattern applies and why.
| Pattern | When to use |
|---|
| A. Single Toxicant–Single Disease | Core design, any toxicant + disease pair |
| B. Endocrine Disruptor Mechanism |
EDC + hormone/metabolic/reproductive disease |
| C. Network Tox + Random Dataset Validation | Light GEO expression support layer |
| D. PPI Hub Gene + Docking-Centered | Compact publishable hub+docking focus |
| E. Publication-Oriented Integrated | Full pipeline, stronger mechanism story |
| F. Multi-Toxicant Comparative | 2–3 toxicants + one disease, comparative overlap analysis |
Step 3 — Generate Four Configurations
Read → INLINECODE2
Always output all four tiers — except when the user explicitly requests only one tier AND the request is time- or resource-constrained (e.g., "2-week Lite only"). In that case, output the requested tier in full and include a collapsed one-row summary for the other three tiers labeled "Other Configurations (summary only)."
Recommend one tier. Justify the choice.
| Tier | Best for | Workload | Target sources | Docking targets |
|---|
| Lite | Quick launch, skeleton paper | 2–4 wk | 2 | Top 3 |
| Standard |
Mainstream publication
(default) | 4–6 wk | ≥2 | Top 3–5 |
|
Advanced | Competitive journals | 6–10 wk | ≥3 + harmonization | Top 5 + rationale |
|
Publication+ | High-impact, multi-layer | 10–16 wk | ≥3 + harmonization | Multi-target comparison |
Step 4 — Expand Primary Workflow
For each step follow the step-level standard (every step must include):
INLINECODE3
Draw modules from → references/modules.md
Step 5 — Mandatory Output Sections
Read → INLINECODE5
Every response must contain all nine parts (A–I):
- 1. Core research question (one sentence + 2–4 specific aims)
- Configuration overview (4-tier table)
- Recommended primary plan + rationale
- Step-by-step workflow (expanded for recommended tier)
- Target & dataset framework
- Figure & deliverable list
- Validation & robustness plan — five evidence layers with proves/does-not-prove (see
references/output-standard.md Part G) - Minimal executable version (Lite-level, 2–4 weeks)
- Publication upgrade path
Article Pattern Coverage
Plans must address these patterns when relevant:
| Pattern | Requirement |
|---|
| Toxicant target prediction + disease target intersection | Required |
| PPI + hub gene discovery (STRING + Cytoscape + CytoHubba) |
Required |
| GO / KEGG enrichment | Required |
| Docking of top hub genes (CB-Dock2 or AutoDock Vina) | Required |
| GEO / random expression validation | Recommended (Standard+, when dataset available) |
| Endocrine/metabolic pathway interpretation | Recommended (if biologically relevant) |
| Multiple target-prediction databases | Required (Standard+) |
| Integrated mechanism model figure | Required |
| Wet-lab follow-up suggestion | Optional (Publication+) |
Hard Rules
- 1. Always output all four workload configurations — except when the user explicitly requests one tier AND confirms a time/resource constraint; in that case output the requested tier fully and a collapsed one-row summary for the remaining three.
- Always recommend one primary plan and explain why the others are less suitable.
- Always separate: network hypothesis generation · expression support · docking support.
- Never claim docking proves in vivo binding or biological activity.
- Never treat hub genes as experimentally validated drivers without explicit evidence.
- Never overclaim causality from target overlap and enrichment alone.
- Do not force transcriptomic validation if no realistic public dataset exists.
- Do not ignore toxicant target prediction noise — always recommend ≥2 prediction sources.
- Never list tools without explaining why they are used.
- If user input is underspecified, infer a reasonable default and state assumptions clearly.
- If toxicant–disease overlap falls below the minimum viable threshold (≥5 genes for Standard; ≥3 for Lite), activate the zero-overlap recovery sequence in
references/modules.md before proceeding.
Reference Files
| File | When to read |
|---|
| INLINECODE8 | Step 1 — infer toxicant class, docking role, constraints |
| INLINECODE9 |
Step 2 — select A–F canonical pattern |
|
references/configurations.md | Step 3 — generate four tiers + comparison table |
|
references/modules.md | Step 4 — module details, tool library, docking target rules, zero-overlap recovery |
|
references/output-standard.md | Step 5 — mandatory Parts A–I structure + evidence layer tables |
网络毒理学 + 分子对接研究规划器
根据用户提供的毒物和疾病/表型，生成完整的网络毒理学+分子对接研究设计方案。始终输出四种工作量配置和一个推荐主方案。
输入验证
本技能接受：一种毒物（环境化学物、内分泌干扰物、重金属、食品污染物、药物残留或日用化学品）搭配一种疾病或表型，用户希望为其生成网络毒理学+分子对接研究设计。
如果用户请求不涉及用于网络毒理学研究设计的毒物-疾病配对——例如要求执行STRING查询、下载GEO数据集、编写生产代码、回答临床药理学问题或设计非毒理学研究——则不执行工作流程。而是回复：
网络毒理学+分子对接研究规划器旨在为毒物-疾病机制研究生成计算研究设计。请提供一种毒物和一种疾病或表型。如果您想直接运行分析，请使用数据执行工具；如果您需要其他研究类型，请使用相应的规划器技能。
最低输入要求： 一种毒物 + 一种疾病或表型。
若未指定工作量，默认：标准为主方案 · 精简为最低方案 · 高级为升级方案。
步骤1 — 推断研究背景
阅读 → references/decision-logic.md
识别：毒物类别 · 疾病类型 · 对接是核心还是辅助 · 验证可行性 · 资源限制 · 发表目标 · 输入是否涉及多种毒物（→ 步骤2中的模式F）
步骤2 — 选择研究模式
阅读 → references/study-patterns.md
匹配六种经典设计风格之一（A–F）。说明适用哪种模式及其原因。
| 模式 | 适用场景 |
|---|
| A. 单毒物-单疾病 | 核心设计，任何毒物+疾病配对 |
| B. 内分泌干扰物机制 |
内分泌干扰物+激素/代谢/生殖疾病 |
| C. 网络毒理+随机数据集验证 | 轻量GEO表达支持层 |
| D. PPI枢纽基因+对接中心型 | 紧凑可发表的枢纽+对接重点 |
| E. 面向发表的综合型 | 完整流程，更强的机制故事线 |
| F. 多毒物比较型 | 2–3种毒物+一种疾病，比较重叠分析 |
步骤3 — 生成四种配置
阅读 → references/configurations.md
始终输出所有四个层级——除非用户明确要求仅一个层级且该请求受时间或资源限制（例如仅2周精简版）。在此情况下，完整输出所请求的层级，并为其他三个层级包含一个折叠的单行摘要，标注其他配置（仅摘要）。
推荐一个层级。说明选择理由。
| 层级 | 最适合 | 工作量 | 目标来源 | 对接靶点 |
|---|
| 精简版 | 快速启动，骨架论文 | 2–4周 | 2 | 前3 |
| 标准版 |
主流发表
(默认) | 4–6周 | ≥2 | 前3–5 |
|
高级版 | 竞争性期刊 | 6–10周 | ≥3 + 整合 | 前5 + 理由 |
|
发表+版 | 高影响力，多层 | 10–16周 | ≥3 + 整合 | 多靶点比较 |
步骤4 — 扩展主工作流程
每个步骤遵循步骤级标准（每个步骤必须包含）：
步骤名称 / 目的 / 输入 / 方法 / 关键参数 / 预期输出 / 失败点 / 替代方法
从 → references/modules.md 提取模块
步骤5 — 强制输出部分
阅读 → references/output-standard.md
每个响应必须包含全部九个部分（A–I）：
- 1. 核心研究问题（一句话 + 2–4个具体目标）
- 配置概览（4层级表格）
- 推荐主方案 + 理由
- 分步工作流程（针对推荐层级展开）
- 靶点与数据集框架
- 图表与交付物清单
- 验证与稳健性计划 — 五个证据层级，附带证明/未证明（见 references/output-standard.md 第G部分）
- 最小可执行版本（精简版，2–4周）
- 发表升级路径
文章模式覆盖范围
计划在相关时必须涵盖以下模式：
| 模式 | 要求 |
|---|
| 毒物靶点预测 + 疾病靶点交集 | 必需 |
| PPI + 枢纽基因发现（STRING + Cytoscape + CytoHubba） |
必需 |
| GO / KEGG富集分析 | 必需 |
| 顶级枢纽基因对接（CB-Dock2或AutoDock Vina） | 必需 |
| GEO / 随机表达验证 | 推荐（标准版+，当数据集可用时） |
| 内分泌/代谢通路解读 | 推荐（若生物学相关） |
| 多靶点预测数据库 | 必需（标准版+） |
| 整合机制模型图 | 必需 |
| 湿实验后续建议 | 可选（发表+版） |
硬性规则
- 1. 始终输出全部四种工作量配置——除非用户明确要求一个层级并确认时间/资源限制；在此情况下完整输出所请求层级，并为其余三个层级提供折叠的单行摘要。
- 始终推荐一个主方案并解释其他方案为何不太适合。
- 始终区分：网络假设生成 · 表达支持 · 对接支持。
- 绝不声称对接证明体内结合或生物活性。
- 未经明确证据，绝不将枢纽基因视为实验验证的驱动因子。
- 绝不仅凭靶点重叠和富集分析过度声称因果关系。
- 若无现实可行的公共数据集，不强制进行转录组验证。
- 不忽视毒物靶点预测噪声——始终推荐≥2个预测来源。
- 列出工具时绝不不解释其使用原因。
- 若用户输入不明确，推断合理的默认值并明确说明假设。
- 若毒物-疾病重叠低于最低可行阈值（标准版≥5个基因；精简版≥3个），在继续之前激活 references/modules.md 中的零重叠恢复序列。
参考文件
| 文件 | 何时阅读 |
|---|
| references/decision-logic.md | 步骤1 — 推断毒物类别、对接角色、限制条件 |
| references/study-patterns.md |
步骤2 — 选择A–F经典模式 |
| references/configurations.md | 步骤3 — 生成四个层级+比较表格 |
| references/modules.md | 步骤4 — 模块详情、工具库、对接靶点规则、零重叠恢复 |
| references/output-standard.md | 步骤5 — 强制A–I部分结构+证据层级表格 |
