Proposed Field Guide Outline

Audience: Researchers with similar omics datasets (potentially different species or stressors) who want to learn from this project’s lessons, methods, and pipelines for biomarker discovery.

0. Start Here (Landing Page — revised to be succinct)

Format: Abstract-style, ~250–300 words total.

Content:

  1. Purpose — This field guide documents considerations for building a biomarker discovery pipeline from RNA-seq meta-analysis, with an emphasis on lessons learned, reproducibility, and practical recommendations for researchers working with noisy multi-study omics data.

  2. Grant mission — One to two sentences drawn from the project narrative: developing molecular resilience biomarkers to support selective breeding and management of shellfish aquaculture under disease pressure. (Full narrative: ProjectSummaryandNarrative.pdf)

  3. What was done — Two to three sentences: integrated RNA-seq datasets from Crassostrea virginica exposed to Perkinsus marinus, conducted differential abundance analyses, and developed a two-step classifier pipeline.

  4. Key results — One to two sentences: identified a validated 6-gene classifier panel that distinguishes tolerant from sensitive oyster phenotypes, confirmed via Leave-One-Study-Out (LOSO) cross-validation.

  5. Intended application — One sentence: this guide is a reusable template for researchers facing batch effects, weak signals, and overfitting risks in multi-study biomarker discovery.

  6. How to navigate this guide — A short table or bullet list pointing to the main sections below.

1. Research Context & Problem Framing

Why this study, why these data, and what “resilience” means here.

  • Background: Dermo disease (P. marinus) and its impact on oyster aquaculture
  • Biological question: Can gene expression predict tolerance/resistance phenotypes?
  • Dataset overview: study descriptions, species, stressors, sample sizes
  • Phenotype definition (tolerant vs. sensitive) and how it evolved

2. Process Narrative (replaces “Year in Review”)

A chronological account of decisions, pivots, and surprises — the honest story of what happened.

  • Phase 1 (Dec 2024 – Jan 2025): Integrated analysis attempt; why it failed
  • Phase 2 (Feb 2025): Confronting batch effects; COMBAT and RemoveBatchEffect trials
  • Phase 3 (Apr – Jun 2025): Shift to per-dataset independent analysis; TAG-seq parameter discovery
  • Phase 4 (Jul 2025): Normalization and study-combination experiments
  • Phase 5 (Aug 2025): Stepwise differential abundance; discovery of innate-signal problem
  • Phase 6 (Aug – Sep 2025): Two-step classifier development; 6-gene panel validation

3. Big Lessons Learned

Distilled, numbered insights for researchers adapting this work to new species or stressors.

  1. Integrated analysis fails with noisy, weak-signal data → use post-data integration instead
  2. Trait definitions must be specific and comparable → generic “stress vs. control” is insufficient
  3. Innate biomarkers live in control groups → stepwise filtering removes them by design
  4. Training-set leakage inflates accuracy → prevent leakage at every step (feature selection, normalization, tuning)
  5. Pipeline internals matter → PCAs in nf-core/differentialabundance are generated before normalization
  6. Technology differences require different parameters → TAG-seq ≠ standard RNA-seq

4. Methods & Pipelines

Practical, decision-first guidance for implementing the analyses.

4a. Pipeline Decision Guide

  • Flowchart: Which approach to choose and when (box diagram/flowchart graphic)

4b. Stepwise Differential Abundance Pipeline

  • Step 1: control vs. treated (stress-responsive genes)
  • Step 2: resistant vs. sensitive (from Step 1 genes)
  • Normalization approach; known failure modes

4c. Two-Step Classifier Pipeline (primary validated approach)

  • Step 1: Reproducibility and directionality scoring across datasets (box diagram/flowchart graphic)
  • Step 2: Logistic regression to minimize feature set
  • How the 6-gene panel was identified

4d. Validation & Pitfalls

  • Avoiding overfitting; train/test hygiene
  • Leave-One-Study-Out (LOSO) protocol
  • Detecting and handling batch effects
  • Cross-study generalizability considerations

5. Analysis Code & Source Code

Entry point for reproducibility — where to find and how to run everything.

  • GitHub repository structure overview
  • Key analysis scripts with brief descriptions
  • How to run pipelines from scratch (environment setup, inputs, outputs)
  • Links to relevant notebook entries (organized by pipeline/phase, not just chronologically)

6. Glossary

Terms defined in the context of this project.

  • Domain terms (e.g., Dermo, tolerance phenotype)
  • Statistical/computational methods (DESeq2, VST, LOSO, logistic regression)
  • Acronyms and abbreviations

7. Sources & References

Primary source material underlying the guide.

  • Notebook posts index (analysis logs)
  • GitHub issues index (decision trail)
  • External publications cited
Current Section Proposed Section Change
Start Here Start Here Rewritten as concise abstract
Problem Framing Research Context & Problem Framing Expanded with dataset details
Year in Review Process Narrative Renamed; framed as narrative
(none) Big Lessons Learned New section
Pipelines (3 sub-pages) Methods & Pipelines (4 sub-pages) Restructured; decision guide added
(none) Analysis Code & Source Code New section
Glossary Glossary Unchanged
Sources (posts + issues) Sources & References Moved to end; scoped as references

This outline is a proposal for review. Full implementation will proceed after approval.