Source: https://github.com/aipoch/medical-research-skills

Bioinformatics Analysis Workflow Extraction and Plan Design Skill

Core Task

The core objectives of this skill are:

1. Systematically extract the complete analysis workflow of a bioinformatics paper (at the macro pipeline level)
2. Design customized analysis plans for new diseases/phenotypes

Workflow

Step 1: Confirm Literature Source

When the user expresses intent to extract an analysis workflow:

• If a PDF file already exists in the workspace (check the directory for .pdf files via Glob): use it directly, inform the user which file is being used
• If no file exists: ask the user to upload/provide the PDF literature file
  • Prompt the user that the file should be placed in the current working directory
  • If necessary, remind the user they can provide it via drag-and-drop or upload

Step 2: Read and Extract Analysis Workflow

This skill includes a built-in scripts/extract_pdf.py Python script for PDF text extraction (depends on the PyMuPDF library, pre-installed in the environment).

PDF Reading Strategy (two complementary methods):

Method A — Direct Read tool (preferred): Use the Read tool to directly read the PDF file. Some environments support PDF rendering and can directly obtain text content.

• Advantage: No extra steps needed, completed in one action
• Limitation: Chart text may be missing or misaligned (this is normal)
• Focus on the Methods and Results sections

Method B — Python script full-text extraction (recommended/fallback): If Read tool output is insufficient, or more complete text is needed, use Bash to run the extraction script:

python3 scripts/extract_pdf.py <pdf_path> <output_txt_path>

The script extracts text from all PDF pages and saves it as a .txt file with page number markers. Then use the Read tool to read the generated txt file.

Recommended workflow:

1. First use Method A (Read tool) for a quick overview of the PDF to get the overall structure
2. If key information is incomplete (e.g., methods section is truncated, chart text is missing), use Method B to extract full text
3. Cross-reference the txt file text with the Read tool content to ensure completeness

The extracted analysis workflow should be at the macro pipeline level, identifying the major analysis stages in the paper, with each stage including:

• Stage name (e.g., data preprocessing, differential expression analysis, functional enrichment analysis, prognostic model construction, immune infiltration analysis, etc.)
• Purpose (why this step is performed)
• Methods/tools/databases used (e.g., limma package, clusterProfiler, LASSO regression, TCGA database, etc.)
• Key parameters or filtering criteria (e.g., |log2FC|>1, adj.P<0.05, P<0.05, etc.)
• Input and output (what data goes in at each step, what results come out)

Notes during extraction:

• Focus on the logical relationships and sequential order between analysis methods (which step depends on the output of which)
• Identify whether the paper uses single-dataset analysis or multi-dataset validation, and their validation strategies
• Distinguish between primary analyses and supplementary analyses (Supplementary)
• If the paper contains multiple independent analysis tracks (e.g., transcriptomics + single-cell + epigenomics), summarize each separately

Step 3: Output Analysis Workflow Summary

The output process is: first generate a Markdown intermediate file, then convert to .docx format using scripts/generate_docx.py for user delivery.

3a. Generate Markdown Intermediate File

Write a structured summary document in Chinese Markdown format, saved as analysis_workflow_summary.md (as an intermediate file).

Document structure as follows (adjust according to actual paper content; it is not necessary to strictly follow this structure, but ensure it is structured):

# Literature Analysis Workflow Summary

**Literature Information**: [Title], [Journal/Year] (note if identifiable from the PDF)

## Analysis Pipeline Overview
Use a flowchart or bullet points to briefly describe the overall analysis pipeline, giving readers an at-a-glance understanding.

## Stage 1: [Stage Name]
### Purpose
### Data Source
### Methods/Tools
### Key Parameters/Filtering Criteria
### Main Output Results

## Stage 2: [Stage Name]
...

## Key Validation Strategies
- Internal validation:
- External validation:
- Other validation:

## Reusable Analysis Patterns
Extract generalizable analysis approaches (e.g., "differential screening → LASSO dimensionality reduction → multivariate regression model building → multi-dataset validation" as a universal pattern)

3b. Convert to Word Document

Use the built-in script to convert Markdown to a professionally formatted .docx file:

python3 scripts/generate_docx.py analysis_workflow_summary.md .

The script generates analysis_workflow_summary.docx, including:

• Correct Chinese/English font settings (SimSun/Times New Roman)
• Hierarchical heading formats (SimHei bold, Level 1 18pt centered, Level 2 15pt, Level 3 13pt)
• Bold, italic, code inline format rendering
• Markdown tables rendered as Word tables (headers with blue background)
• Code blocks in monospaced font
• List rendering (ordered/unordered)
• Professional page setup (2.54cm margins)

After outputting the .docx, also display a core content summary in the conversation (no need to show the full text; provide the pipeline overview and key findings, guiding the user to view the .docx file).

Step 4: Ask Whether a Customized Plan is Needed

After outputting the summary, proactively ask the user:

Would you like a customized analysis plan designed for another disease or phenotype using this analysis workflow? If so, please provide the disease/phenotype name you are interested in.

Step 5: Design Customized Analysis Plan

If the user provides a disease/phenotype name:

Based on the workflow extracted in Step 2, design an adapted plan stage by stage for the new disease/phenotype.

Output process: first generate custom_analysis_plan_[disease_name].md intermediate file, then convert to .docx:

python3 scripts/generate_docx.py custom_analysis_plan_[disease_name].md .

Adaptation principles:

• Maintain the core logic of each analysis stage, but select appropriate databases, parameters, and tools for the new disease
• Data sources: Recommend relevant public databases for the new disease (TCGA, GEO, ICGC, GTEx, etc.) or specify what data needs to be collected independently
• Sample size considerations: Different diseases have varying sample availability; adjust analysis strategies accordingly (e.g., rare diseases may require different modeling approaches)
• Clinical relevance: Combine with the clinical characteristics of the new disease to suggest clinically meaningful analysis entry points
• Feasibility assessment: Provide actionable recommendations for each stage, noting potential difficulties

Output format similar to Step 3, but with added "Adaptation Notes":

# Customized Analysis Plan: [New Disease/Phenotype Name]

## Based on Literature Workflow: [Original Paper Analysis Pipeline Name]

### Stage 1: [Stage Name] → Adapted Plan
- **Original method**:
- **Adaptation recommendations**:
- **Recommended databases/tools**:
- **Notes**:

...

Notes and Principles

1. Chinese interaction: Communicate with the user in Chinese throughout; output is also in Chinese Markdown
2. Macro granularity: Do not go down to line-by-line code or specific syntax level; stay at the method/tool level
3. Honest labeling: If PDF content is insufficient to extract information about certain stages, explicitly note "not clearly stated in the paper" rather than fabricating
4. No preset templates: Dynamically construct structure based on actual paper content; do not force-fit a fixed template
5. File management: First generate .md intermediate file (can be cleaned before output), then use scripts/generate_docx.py to convert to .docx format. The final deliverable to the user is the .docx file. Intermediate .md files can be kept or deleted as needed
6. PDF reading strategy: Prefer using the Read tool to directly read PDFs; if content is incomplete (e.g., chart text missing, methods section truncated), immediately use scripts/extract_pdf.py via Bash to extract full text. The extraction script depends on PyMuPDF, pre-installed in the environment. Be especially careful when extracting information from the Methods and Results sections — these are the essence of the analysis workflow
7. Multiple files scenario: If there are multiple PDFs in the workspace, ask the user to specify which paper to analyze
8. Emphasize universal patterns: When outputting "Reusable Analysis Patterns", distill transferable strategies that can be applied to other similar studies — this is the most valuable part for the user

Error Handling

• If required inputs are missing, state exactly which fields are missing and request only the minimum additional information.
• If the task goes outside the documented scope, stop instead of guessing or silently widening the assignment.
• If execution fails, report the failure point, summarize what can still be completed safely, and provide a manual fallback.
• Do not fabricate files, citations, data, search results, or execution outcomes.

Input Validation

This skill accepts requests that match the documented purpose of bioinfo_analysis_plan and include enough context to complete the workflow safely.

Do not continue the workflow when the request is out of scope, missing a critical input, or would require unsupported assumptions. Instead respond:

bioinfo_analysis_plan only handles its documented workflow. Please provide the missing required inputs or switch to a more suitable skill.