Comprehensive QC checklists and guidelines for experimental design. --- ## Power Analysis QC ### Before Calculating Power - [ ] **Verify CV estimates match your biological system** - Cell lines: CV = 0.1-0.2 - Inbred mice: CV = 0.2-0.3 - Outbred mice/primary cells: CV = 0.3-0.4 - Human samples: CV = 0.3-0.5 - Check cv\_tissue\_database.csv for your specific tissue - [ ] **Use pilot data if available** - Pilot-based power is more accurate than literature CV estimates - Requires at least 2-3 samples per condition in pilot - Use `extract_dispersion_deseq2()` to get accurate parameters - [ ] **Sequencing depth is realistic** - Bulk RNA-seq: 15-30 million reads typical - scRNA-seq: 50,000-100,000 reads per cell typical - ATAC-seq: 25-50 million reads typical - Don't assume excessive depth to compensate for low sample size - [ ] **Effect size is biologically meaningful** - Can you measure this fold-change with your assay? - Is this effect size relevant to your biological question? - Smaller effects require more samples ### After Calculating Power - [ ] **Power ≥0.80 (80%)** - Lower power wastes resources (high false negative rate) - For critical experiments, aim for power ≥0.90 - [ ] **Test multiple scenarios** - Calculate power for range of effect sizes - Test different sample sizes - Evaluate trade-offs (depth vs. replicates) - [ ] **Document assumptions** - Record CV source (pilot data or literature) - Document sequencing depth assumptions - Note effect size justification --- ## Sample Size QC ### Minimum Requirements (CRITICAL) - [ ] **At least 3 biological replicates per condition** - n=2 is insufficient for statistics - n=3 is absolute minimum - n≥6 recommended for robust results - [ ] **Add 10-20% buffer for failures** - Sample failures happen (extraction, QC, etc.) - Plan for: `n_planned = n_required × 1.15` - Budget accordingly - [ ] **Biological > technical replicates** - Prioritize more biological samples over technical replicates - Technical replicates don't capture biological variation - Exception: Method validation studies ### Budget and Feasibility - [ ] **Total cost calculated** - Cost = (samples × depth) + prep costs - Verify budget supports proposed design - [ ] **Sample availability confirmed** - Are enough samples obtainable? - Consider recruitment timeline (human studies) - Animal breeding timelines (mouse studies) - [ ] **Sequencing capacity available** - Check facility capacity and timeline - Plan for potential delays ### Paired Design Considerations - [ ] **Verify pairing reduces variance** - Paired designs assume within-pair correlation - Use pilot data to verify benefit - Calculate power assuming reduced CV (×0.7-0.8) - [ ] **Account for attrition in paired studies** - Missing one timepoint loses entire pair - Buffer needs to be higher (~20-30%) --- ## Batch Design QC (MOST CRITICAL) ### Pre-Assignment Checks - [ ] **Understand batch structure** - What defines a "batch"? (day, plate, run, operator?) - How many samples can be processed per batch? - Are there multiple batch levels? (plate + day + sequencing run) - [ ] **Identify all variables to balance** - Primary: Experimental condition (MUST balance) - Secondary: Sex, age, genetic background, collection site - Consider hidden variables (time of day, operator) ### Critical Confounding Check (REQUIRED) - [ ] **Batch NOT confounded with condition** ⚠️ CRITICAL - Each batch MUST contain all experimental conditions - Never process all samples from one condition in single batch - Chi-square test p-value > 0.05 - **If confounded:** Regenerate assignment immediately - [ ] **Visual inspection passed** - Create contingency table: `table(batch, condition)` - Each cell should have samples - Roughly balanced counts ### Balance Checks - [ ] **Primary covariates balanced** - Sex balanced across batches (chi-square p > 0.05) - Age groups balanced if relevant - Genetic background balanced (mouse studies) - [ ] **Sample sizes approximately equal** - All batches have similar n - Deviations <20% acceptable - Last batch can be smaller if needed - [ ] **Controls included in each batch** - Same control sample(s) processed in every batch - Enables batch effect correction - Use "split pool" controls if possible ### Practical Considerations - [ ] **Randomization applied** - Randomize sample order within batches - Use `add_processing_order(randomize_within=TRUE)` - Avoid systematic patterns - [ ] **Plate positions randomized** - Avoid edge wells when possible (edge effects) - Randomize within plates - Document well positions - [ ] **Metadata recording plan ready** - Template for recording batch variables - Fields: date, operator, reagent lots, instrument, plate/well - Lab team trained on importance --- ## Metadata Documentation Requirements ### Essential Batch Metadata (ALWAYS RECORD) - [ ] **Processing date** - Format: YYYY-MM-DD - Include time if relevant (morning vs. afternoon effects) - [ ] **Operator/Technician** - Record who performed each step - Operator effects are real - [ ] **Reagent lot numbers** - Record ALL reagent lots used - Lot-to-lot variation can create batch effects - Keep labels/receipts - [ ] **Instrument/Equipment ID** - Which machine used for each step? - Calibration status - Maintenance logs - [ ] **Plate/Chip information** - Plate barcode or ID - Well positions for each sample - Plate position effects (edge, center) ### Additional Metadata (Recommended) - [ ] **Library prep batch** - If separate from extraction batch - Kit lot numbers - [ ] **Sequencing run information** - Flowcell ID - Lane assignments - Run date - [ ] **Protocol deviations** - ANY deviation from standard protocol - Which samples affected - Reason for deviation - [ ] **Environmental conditions** - Room temperature (if relevant) - Humidity (RNA stability) - Storage conditions during delays --- ## Multiple Testing QC ### Method Selection Verification - [ ] **Method appropriate for data type** - Genome-wide (>5000 tests): BH-FDR or IHW - Candidate genes (<100 tests): Bonferroni - GWAS (millions): Bonferroni with standard threshold (5×10⁻⁸) - [ ] **Consider effect heterogeneity** - Heterogeneous effects: Use IHW if covariate available - Homogeneous effects: BH-FDR sufficient - [ ] **Sample size adequate for method** - Small n (<6): Consider permutation-based FDR - Parametric assumptions may be violated ### Post-Correction Checks - [ ] **Number of significant results reasonable** - Too many (>50%): Check for technical issues - Too few (0-1): Power may be insufficient - [ ] **QQ plot checked (if applicable)** - Verify calibration of p-values - Check for inflation/deflation - [ ] **Top hits make biological sense** - Known genes in pathway? - Consistent with literature? - Effect sizes reasonable? --- ## Grant Proposal QC Checklist When preparing experimental design for grant proposals: - [ ] **Power analysis documented** - CV source clearly stated - Power calculations shown - Sample size justified - [ ] **Budget justified** - Cost per sample calculated - Sequencing depth justified - Why this n and not less? - [ ] **Batch design planned** - Statement that batches will be balanced - Description of batch effect mitigation - Reference to statistical analysis plan - [ ] **Statistical analysis plan included** - Which software/methods - Multiple testing correction method - Batch effect correction approach - [ ] **Citations included** - Power analysis method papers - Statistical method papers - Best practices references - [ ] **Preliminary data supports assumptions** - Pilot data shows expected variability - Effect size estimates reasonable - Technical feasibility demonstrated --- ## Red Flags to Watch For ### Power Analysis Red Flags - ⚠️ **Power <0.60**: Study severely underpowered - ⚠️ **Very large n (>20/group) for large effects**: Check assumptions - ⚠️ **CV unusually low (<0.1) for complex samples**: Likely unrealistic - ⚠️ **Depth >100M reads for RNA-seq**: Diminishing returns, better to add samples ### Batch Design Red Flags - 🚨 **CRITICAL: Batch confounded with condition**: DO NOT PROCEED - ⚠️ **All samples of one condition in first batch**: Bad randomization - ⚠️ **No controls across batches**: Can't correct batch effects - ⚠️ **Processing timeline >6 months**: Long timeline increases batch risk - ⚠️ **Different operators for each condition**: Operator confounded ### Sample Size Red Flags - ⚠️ **n=2 per group**: Insufficient for statistics - ⚠️ **No buffer for failures**: Will likely be underpowered - ⚠️ **Technical replicates counted as biological**: Wrong variance estimate - ⚠️ **Paired design treated as independent**: Wrong power calculation --- **Last Updated:** 2026-01-28 **Maintained by:** Knowhows Repository