OncoBulk Research Copilot

Bilingual, privacy-aware skill for tumor-focused medical graduate students working with GEO, TCGA, and local expression matrices.

面向肿瘤医学研究生的双语 AI skill：先做隐私与路径分流，再做转录组分析、免疫微环境规划与方法写作。

Why this repo exists | 为什么做这个项目

Many transcriptome skills are powerful but hard to reuse: they mix private paths, giant monolithic workflows, and reporting logic into one bundle.

很多转录组 skill 虽然功能全，但常见问题也很明显：路径写死、流程过重、隐私边界不清、论文方法与分析执行脱节。

OncoBulk Research Copilot is designed as a research copilot, not a script warehouse.

What makes it different | 它和常见流水线 skill 的区别

Generic pipeline style	OncoBulk Research Copilot
Start from tools	Start from study intent
Force one long workflow	Route by lane: intake / DEG / immune / methods
Often ignores privacy	Explicit privacy and de-identification gate
Focus on execution only	Also builds manifests, methods text, and captions
Easy to accumulate path debt	Portable scaffold with no personal absolute paths

Core capabilities | 核心能力

1. Study intake router
   GEO / TCGA / local matrix / results-only 四类入口分流。

2. Privacy and safety gate
   优先识别本地临床文件、潜在 PHI、不可公开资源。

3. Project scaffold
   统一创建可复现目录、manifest 和报告出口。

4. Decision-first analysis lanes
   不强迫所有任务都走完整 preprocess -> DEG -> WGCNA -> GSEA。

5. Methods and report pack
   自动生成 dataset card、methods paragraph、parameter summary、figure caption。

Quick-start examples | 快速使用示例

Example 1: GEO bulk expression study

Use $oncobulk-research-copilot to inspect GSE19804, decide whether it is microarray or bulk RNA-seq, and prepare a safe project scaffold.

Example 2: TCGA tumor-vs-normal route

Use $oncobulk-research-copilot to route a TCGA-STAD processed matrix into a tumor-vs-normal DEG plan and generate a methods summary.

Example 3: Tumor immune microenvironment lane

Use $oncobulk-research-copilot to plan an immune microenvironment workflow for a gastric cancer cohort and document the required external resources without redistributing them.

Example 4: Paper-ready support only

Use $oncobulk-research-copilot to turn my finished DEG and immune results into a methods paragraph, dataset card, and figure captions.

Usage snapshot | 实际调用示意

One natural-language request can be routed into the right lane, then turned into scaffolded analysis outputs and paper-ready writing support.

你不需要先翻完整个 skill 文件夹，直接把输入类型、研究目标和是否涉及本地敏感数据说清楚，skill 就会先分流、再组织输出。

Repository structure | 仓库结构

oncobulk-research-copilot/
├─ SKILL.md
├─ agents/openai.yaml
├─ references/
├─ scripts/
├─ assets/
└─ LICENSE

References

• references/intake-routing.md
• references/privacy-safety.md
• references/analysis-lanes.md
• references/immune-lane.md
• references/methods-pack.md
• references/publication-checklist.md
• references/release-process.md

Scripts

• scripts/preflight_check.py
• scripts/build_dataset_manifest.py
• scripts/make_project_scaffold.py
• scripts/methods_pack_builder.py

Minimal workflow map | 最小工作流

flowchart LR
    A["Input: GEO / TCGA / local / results"] --> B["Privacy gate"]
    B --> C["Study intake router"]
    C --> D["Project scaffold + manifest"]
    D --> E["Analysis lane"]
    E --> F["Methods + caption pack"]

Loading

Privacy note | 隐私说明

This repo is not intended for raw PHI, direct patient identifiers, or redistribution of restricted third-party deconvolution assets.

本项目不用于公开原始患者身份信息、敏感临床台账，或不明确授权的第三方免疫浸润资源再分发。

Best use cases | 最适合的场景

• medical graduate students doing bulk tumor transcriptome projects
• GEO/TCGA route planning before full execution
• local de-identified matrix organization
• immune/TME workflow framing
• methods and figure writing support for theses and papers

Project maintenance | 项目维护入口

• CHANGELOG.md
• references/release-process.md
• .github/release-template.md

License

This repository is released under the MIT License for the original content in this repo.