SKILL.md — Paper2Protocol Skill Definition

Version: 1.2
Created: 2026-03-20
License: CC BY-NC 4.0

Overview

From published high-impact primary literature, reverse-engineer complete experimental validation plans — transforming scientific discoveries into executable research protocols.

Core Principle: Only use primary sources (PMC full-text, journal PDFs), never abstracts or second-hand reviews.

---

Input Requirements

✅ Accepted

• PMC full-text (NCBI PubMed Central, Open Access)
• Journal website PDFs (Nature/Science/Cell, peer-reviewed)
• DeepReader-generated full-text analysis documents

❌ Rejected

• Abstracts only
• News articles / media interpretations
• Review articles (as primary input)
• AI-generated summaries (not based on primary sources)

Input Formats

1. PMC URL → Auto-fetch full text
2. PDF file → Direct analysis
3. Paper title → Search PMC for full text

---

Workflow (5 Stages)

Stage 1: Source Acquisition & Quality Assessment

1. Validate input as primary source
2. Fetch full text (PMC API / PDF parsing)
3. Quality rating:
   • Journal tier (CNS / sub-journal / field-top / other)
   • Research type (basic / clinical / translational)
   • Data completeness (supplementary materials, raw data links)
   • Reproducibility (method detail, sample size)

Stage 2: Scientific Logic Deconstruction

Extract complete scientific logic:

1. Core Scientific Question: What problem does this paper solve?
2. Research Strategy: Hypothesis, models (in vivo/in vitro/in silico/clinical), key techniques
3. Validation Chain:

   Hypothesis → Key Experiment 1 → Key Experiment 2 → ... → Conclusion

   Annotate purpose and expected outcome at each node.
4. Innovation Analysis: Methodological, conceptual, and application innovations.

Stage 3: Executable Experimental Paths

3.1 Experiment Layering

• Must-do: Core experiments validating the hypothesis
• Should-do: Supporting experiments
• Nice-to-do: Mechanism deep-dives or scope extensions

3.2 Per-Experiment Details

Field	Content
Experiment Name	Specific name
Purpose	Role in validation chain
Method	Detailed protocol (paper Methods + best practices)
Samples/Materials	Cell lines, animal models, clinical samples
Sample Size	Statistically required minimum
Key Reagents	Brand, catalog reference, concentration
Equipment	Required instruments + alternatives
Expected Results	Positive/negative controls, data type
Timeline	Per-experiment duration + replicates
Budget	Reagents + consumables + services
Risk Assessment	Failure causes + backup plans

3.3 Bioinformatics Analysis (if applicable)

Field	Content
Analysis Goal	Specific task
Data Source	Public databases (TCGA/GEO) or generated data
Tools	Recommended pipeline (R/Python/online)
Key Parameters	Standard settings
Expected Output	Figure types, statistics
Compute Resources	Local/server/cloud requirements

3.4 Bioinformatics Code (REQUIRED when analysis involves bioinformatics)

When experiments involve bioinformatics, complete runnable code MUST be provided.

Requirements:

• Language: R (Bioconductor) or Python (R preferred)
• Completeness: End-to-end, data download to publication figures
• Comments: Key steps annotated in English
• Data Sources: Prioritize public databases (TCGA, GEO, Beat-AML)
• Standard Tools: ssGSEA/GSEA, DESeq2, CIBERSORTx/xCell, survival, ComplexHeatmap
• Statistical Rigor: Multiple testing correction (BH), power analysis

Coverage:

1. Subtype Classification: ssGSEA + K-means/Hierarchical clustering
2. Differential Expression: DESeq2/edgeR → volcano plot
3. Survival Analysis: Kaplan-Meier + Cox regression + ROC (timeROC)
4. Gene Enrichment: GSEA + ssGSEA + Hallmark/Immunologic gene sets
5. Immune Microenvironment: CIBERSORTx/xCell deconvolution
6. Heatmaps: ComplexHeatmap / pheatmap
7. Prognostic Models: LASSO Cox + glmnet + Nomogram (rms)
8. Flow Cytometry: FlowJo export → Python statistical analysis
9. Panel Selection: LASSO + Random Forest intersection → minimal gene set
10. Automation: Bash shell script to chain all analysis steps

3.5 Budget Summary

Phase 1 (Core Validation): $XX,XXX
  - Reagents: $X,XXX
  - Consumables: $X,XXX
  - Services (sequencing): $XX,XXX
  - Animals: $X,XXX

Phase 2 (Mechanism): $XX,XXX
...

Total: $XXX,XXX – $XXX,XXX

Stage 4: Extension Projects (2-3 proposals)

Each includes:

• Project Name
• Scientific Question
• Innovation vs original paper
• Feasibility: ⭐ rating (technical difficulty, resources, timeline)
• Expected Outcomes: Paper tier, patent potential, clinical value
• Risk Assessment: Bottlenecks and failure risks

Stage 5: Multi-Paper Synthesis (Accumulation Mode)

Triggered when ≥3 papers accumulate per topic:

• By Scientific Question: Group papers by shared research questions
• By Method: Rank techniques by frequency → prioritize platform setup
• Integrated Roadmap: Deduplicate protocols, consolidate budgets
• Research Timeline: 12-month plan based on synthesis

---

Output Format

Standard Structure

# 📋 [Paper Title] → Experimental Validation Plan

## 📄 Paper Information
## 🔬 Part 1: Validation Logic
## 🧪 Part 2: Executable Experimental Paths
## 💻 Part 3: Bioinformatics Code (if applicable)
## 🚀 Part 4: Extension Projects
## 📝 Execution Recommendations

Output Formats

• Markdown (default)
• PDF Report (HTML → browser print, all tables and code blocks)
• Any document platform (Feishu, Notion, etc.)

---

Storage & Indexing

literature-to-experiment/
├─ index.json
├─ by_project/
│  └─ [Project Name]/
│     └─ PMCxxxxxx_protocol.md
├─ by_topic/
│  └─ [Topic Name]/
└─ summaries/
   └─ [Topic]_synthesis.md

---

Notes

1. Pricing: Based on 2025-2026 market rates, marked "reference price"
2. Sample Size: Follows statistical principles, power analysis recommended
3. Ethics: Mark IRB/IACUC requirements for human/animal studies
4. Timeliness: Flag methods >5 years old for verification
5. Code: Must provide complete runnable code for bioinformatics analyses

---

Dependencies

• DeepReader: Full-text analysis (pre-requisite step)
• academic-paper: If integrating plans into papers

License

CC BY-NC 4.0 — Free for academic use with attribution. No commercial use without permission.

Authors

• Jiacheng Lou (GitHub)
• 🦞 Claw (AI Research Assistant)