10  Extended: Full Paper Section Workflow

Building on the citation recovery demo, let’s explore how Claude Code can assist with broader paper writing tasks.

10.1 Overview

This section covers:

  1. Outlining a methods section
  2. Generating figure legends
  3. Consistency checking
  4. Formatting for submission

10.2 Outlining a Methods Section

Methods sections follow predictable patterns. Claude can help structure them:

> I need to write a methods section for a single-cell RNA-seq analysis.
> Help me create an outline covering all the typical subsections.

Claude suggests a structure:

## Methods

### Sample collection and preparation
- Source of samples
- Ethical approvals
- Tissue processing

### Single-cell RNA sequencing
- Library preparation method (10x, Smart-seq, etc.)
- Sequencing platform and parameters
- Number of cells/samples

### Data preprocessing
- Read alignment (tool, reference)
- Quality control (metrics, filtering thresholds)
- Normalization approach

### Clustering and annotation
- Dimensionality reduction
- Clustering algorithm and parameters
- Cell type annotation method

### Differential expression analysis
- Statistical method
- Multiple testing correction
- Significance thresholds

### Data and code availability
- Repository links
- Accession numbers

10.2.1 Filling in Details

> For the "Data preprocessing" section, I used:
> - CellRanger 7.0 for alignment to GRCh38
> - Filtered cells with <200 genes or >20% mitochondrial reads
> - Normalized with scran
>
> Write this as a proper methods paragraph.

Claude generates:

### Data preprocessing

Raw sequencing reads were processed using Cell Ranger v7.0 (10x Genomics)
with alignment to the GRCh38 human reference genome. Quality control was
performed to remove low-quality cells, excluding cells with fewer than 200
detected genes or greater than 20% mitochondrial reads. Data normalization
was performed using the scran package (Lun et al., 2016), which computes
size factors by pooling cells to address composition biases in single-cell
data.

10.2.2 Checking Completeness

> Review my methods draft and flag anything missing that reviewers typically ask about

Common gaps Claude might identify: - Missing software versions - Unclear parameter choices - No justification for filtering thresholds - Missing accession numbers

10.3 Generating Figure Legends

Figure legends have a specific format: statement of what’s shown, then panel-by-panel descriptions.

> I have a figure with:
> - Panel A: UMAP showing all cells colored by cell type
> - Panel B: Dot plot of marker genes by cluster
> - Panel C: Violin plots of key genes across conditions
>
> Write a figure legend.
**Figure 2. Single-cell transcriptomic characterization of tumor samples.**
(A) UMAP visualization of 45,000 cells colored by annotated cell type,
revealing 12 distinct populations. (B) Dot plot displaying expression of
canonical marker genes across identified clusters; dot size represents
the percentage of cells expressing each gene, color intensity indicates
mean expression level. (C) Violin plots showing expression distributions
of key differentially expressed genes (CD8A, FOXP3, IL2RA) across treatment
conditions; significance determined by Wilcoxon rank-sum test (*p < 0.05,
**p < 0.01, ***p < 0.001).

10.3.1 Refining Legends

> This is for a Nature paper. Make it more concise and follow their style.
> The figure actually has 4 panels. Add panel D: a survival curve.

10.4 Consistency Checking

Claude Code can check your paper for internal consistency:

10.4.1 Terminology Consistency

> Check my draft for inconsistent terminology. I might have used
> "single-cell RNA-seq", "scRNA-seq", and "single-cell transcriptomics"
> interchangeably.

Claude identifies inconsistencies and suggests standardizing.

10.4.2 Number Consistency

> Verify that all numbers mentioned are consistent:
> - Number of cells
> - Number of samples
> - Number of genes
> - Percentages that should sum to 100

10.4.3 Reference Consistency

> Check that every citation in the text appears in the reference list,
> and every reference is cited at least once.

10.4.4 Figure/Table References

> Verify that:
> - All figures and tables mentioned exist
> - They're referenced in order (Fig 1 before Fig 2)
> - Supplementary figures are correctly numbered

10.5 Formatting for Submission

Different journals have different requirements:

10.5.1 Converting Citation Style

> Convert all citations from Author (Year) format to numbered references
> in order of appearance.

10.5.2 Word Count Check

> Count words in each section and compare to the journal limits:
> - Abstract: 150 words
> - Main text: 3000 words
> - Methods: no limit but should be concise

10.5.3 Generating Highlights/Summary Points

Many journals require bullet-point highlights:

> Based on this abstract, generate 3-5 highlights (max 85 characters each)
> for journal submission.

10.5.4 Cover Letter Draft

> Draft a cover letter for submitting this paper to [Journal].
> Emphasize the novelty and fit for the journal's scope.

10.6 Working with LaTeX

If your paper is in LaTeX:

> My paper is in LaTeX. Help me:
> 1. Fix the BibTeX formatting
> 2. Ensure figures are referenced correctly
> 3. Check that all packages are necessary
> Convert this Word document to LaTeX while preserving formatting and citations.

10.7 Responding to Reviews

Claude Code can help with revisions:

> Here are the reviewer comments. Help me create a response document with:
> 1. Each comment quoted
> 2. Our response
> 3. Changes made (with page/line numbers)

Template:

## Reviewer 1

> Comment 1: "The authors should provide more details about..."

**Response**: We thank the reviewer for this suggestion. We have expanded
the methods section to include [specific details]. See Methods, page X,
lines Y-Z.

**Changes made**: Added paragraph describing [specific content].

---

> Comment 2: "It is unclear why..."

**Response**: [...]

10.8 Example: Methods Section From Scratch

Let’s write a complete methods section:

> I'm writing a methods section for a paper about spatial transcriptomics
> analysis of mouse brain. Here's what I did:
>
> - Used Visium platform
> - 3 replicates per condition (control vs treated)
> - Processed with Space Ranger
> - Analyzed with Seurat and cell2location for deconvolution
> - Compared regions using SPARK-X for spatial patterns
>
> Write a complete methods section.

Claude drafts each subsection, you refine:

> This is good but too detailed for this journal. Condense each section
> to 2-3 sentences while keeping essential information.
> Add the specific versions of all software used.
> A reviewer might ask why we chose cell2location over other deconvolution
> methods. Add a brief justification.

10.9 Checklist: Paper Writing Workflow

Use this when writing/revising papers:

  • Draft outline of each section
  • Fill in methods with specific details
  • Write figure legends for all figures
  • Check terminology consistency
  • Check number consistency
  • Verify all citations
  • Verify figure/table references
  • Check against journal requirements
  • Generate supplementary materials list
  • Prepare response to reviewers template

10.10 Tips for Working with Claude Code on Writing

10.10.1 Be Specific About Style

> Write this in a formal academic style, not conversational.
> Make this more concise. Remove hedging language.

10.10.2 Provide Context

> This is for a high-impact journal with strict formatting requirements.
> My audience is bioinformaticians, so technical terms don't need explanation.

10.10.3 Iterate on Drafts

Don’t expect perfection on first try:

  1. Generate initial draft
  2. Identify what’s wrong/missing
  3. Request specific changes
  4. Repeat until satisfied

10.10.4 Keep the Human in the Loop

Claude helps but doesn’t replace: - Scientific judgment - Interpretation of results - Ethical considerations - Final responsibility for accuracy

10.11 What You’ve Learned

By exploring this extended workflow, you’ve seen how Claude Code assists with:

  • Structuring paper sections
  • Writing technical prose
  • Maintaining consistency
  • Formatting for submission
  • Organizing revision responses

10.12 Next Steps

Continue to Part 5: Beyond the Demos to see more applications of Claude Code.