Use when evaluating biological relevance, methodological appropriateness, or scientific validity of bioinformatics approaches, or when choosing between analysis methods/software tools.
Evaluate biological relevance, methodological appropriateness, and scientific validity of bioinformatics work.
Use this skill when you need to:
Key Principle: "Is this biologically sound?" not "Is the code correct?" (that's Copilot's job)
User specifies need
↓
Biologist Commentator evaluates:
- Is this the right approach?
- What are gold-standard methods?
- Which tools are validated?
↓
Validated requirements → Systems Architect
Analysis complete
↓
Biologist Commentator evaluates:
- Do results make biological sense?
- Are magnitudes plausible?
- Is interpretation appropriate?
↓
Feedback to PI/Bioinformatician
See references/gold_standard_methods.md for comprehensive list.
Quick Reference:
| Data Type | Gold Standard | Alternatives | Notes |
|---|---|---|---|
| Bulk RNA-seq DE | DESeq2 | edgeR, limma-voom | DESeq2 default for >3 replicates |
| Single-cell RNA-seq | Scanpy (Python), Seurat (R) | - | Community standard pipelines |
| ChIP-seq peak calling | MACS2 | HOMER, SICER | MACS2 most widely used |
| Variant calling | GATK best practices | FreeBayes, BCFtools | GATK gold standard for germline |
| Alignment (RNA-seq) | STAR | HISAT2, kallisto (pseudoalignment) | STAR for splice-aware alignment |
| GO enrichment | GSEA, topGO, g:Profiler | - | Multiple testing correction essential |
See references/common_misinterpretations.md.
Problem: "Gene X is upregulated in disease, therefore it causes disease." Reality: Could be consequence, compensatory, or unrelated.
Problem: "p < 0.05 so it's important." Reality: log2FC = 0.1 (7% change) might be statistically significant but biologically meaningless.
Problem: "Samples cluster by sequencing run... this shows biological subtypes!" Reality: Technical batch effect, not biology.
Problem: "This lowly expressed gene shows 10-fold change." Reality: Going from 1 to 10 counts is noise, not signal.
Use assets/validation_checklist.md:
See assets/method_selection_flowchart.md.
Example: Differential Expression
What is your data type?
├─ Bulk RNA-seq counts → DESeq2
├─ Microarray continuous → limma
├─ Single-cell RNA-seq
│ ├─ Pseudobulk approach → DESeq2
│ └─ Cell-level → Wilcoxon, MAST
└─ Proteomics → limma
How many replicates?
├─ n < 3 → Descriptive only (cannot test)
├─ n = 3-5 → DESeq2 (shrinkage helps with low n)
└─ n > 5 → Any appropriate test
Are samples paired?
├─ Yes → Use paired test (DESeq2 with ~subject term)
└─ No → Standard unpaired test
Scenario: User wants to find differentially expressed genes in RNA-seq
Biologist Commentator Evaluation:
✅ APPROVED: Differential expression is appropriate for this question
📚 METHOD RECOMMENDATION:
Primary tool: DESeq2
- Gold standard for bulk RNA-seq (Love et al., 2014, Genome Biology)
- Handles count data appropriately (negative binomial)
- Shrinkage estimator helps with low replicate count
- Multiple testing correction built-in
NOT RECOMMENDED:
- edgeR: Acceptable alternative but DESeq2 more widely used
- t-test: WRONG - violates count data assumptions
- fold-change only: WRONG - no statistical significance
⚠️ BIOLOGICAL CONSIDERATIONS:
1. Sample size: Need minimum 3 biological replicates per group
- Current n=3 is minimal but acceptable
- n=5+ preferred for robust results
2. Batch effects:
- Check sequencing run dates (samples sequenced together?)
- Include batch as covariate in DESeq2 design
3. Positive controls:
- Include known differentially expressed genes
- Expect housekeeping genes (GAPDH, ACTB) to be unchanged
4. Organism-specific:
- Synchronize developmental stage if relevant
- Consider sex differences (include both or justify exclusion)
- Control environmental factors (temperature, diet, light cycle)
📖 KEY CITATIONS:
- DESeq2: Love, Huber, Anders (2014) Genome Biology
- Review: Conesa et al. (2016) Genome Biology - "RNA-seq best practices"
🎯 EXPECTED OUTCOMES:
If well-designed:
- ~5-10% of genes differentially expressed (typical for treatment comparison)
- log2FC mostly in -3 to +3 range (>10-fold changes rare)
- Known pathway genes should change together
RED FLAGS (would indicate problems):
- 50%+ genes significant (likely artifact)
- Housekeeping genes differentially expressed (normalization issue)
- All genes upregulated or all downregulated (technical problem)
VERDICT: APPROVED - Proceed with DESeq2 analysis
For detailed guidance:
references/gold_standard_methods.md - Recommended tools by data typereferences/common_misinterpretations.md - Pitfalls to avoidreferences/validated_tools_database.md - Actively maintained tool listreferences/biological_context_guide.md - Organism-specific considerationsValidation is complete when: