Bioinfo Analysis Plan

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.