New training: Transcriptome complexity analysis with TSENAT

topics/transcriptomics/tutorials/transcriptome_complexity_tsenat/faqs/index.md

@@ -0,0 +1,138 @@
+# Frequently Asked Questions about TSENAT and Transcriptome Complexity Analysis
+
+## Conceptual Questions
+
+### What's the difference between Tsallis entropy and Shannon entropy?
+
+**Shannon entropy** (q=1 in Tsallis framework) treats all isoforms equally. **Tsallis entropy** adds a parameter `q` that acts as a sensitivity dial:
+- When q < 1: emphasizes **rare isoforms**
+- When q = 1: recovers Shannon entropy (balanced)
+- When q > 1: emphasizes **abundant isoforms**
+
+This allows you to examine complexity at different biological scales in a single analysis.
+
+### Why is transcriptome complexity important if DESeq2 already detects changes?
+
+DESeq2 and edgeR measure **total gene abundance** changes. They miss situations where:
+- A gene maintains constant total expression but **reorganizes its isoforms**
+- Rare isoforms become more common (or vice versa) without affecting overall counts
+- The **diversity** of the isoform landscape changes
+
+These are valid biological signals (isoform switching via splicing regulation) that complement abundance-based methods.
+
+## Technical Questions
+
+### What sample size do I need?
+
+TSENAT assumes **≥6-8 samples per group** for paired/longitudinal designs:
+- Smaller samples are underpowered to detect entropy shifts
+- Larger samples increase power to detect subtle complexity changes
+- Biological replicates are essential (not technical replicates)
+
+### How do I choose the q-spectrum (range and granularity)?
+
+**Standard configuration**: q = 0 to 2.0 with Δq = 0.05 (41 q-values)
+- Range: 0-2 covers extreme rare-emphasis to abundant-emphasis
+- Granularity: Δq=0.05 provides good resolution without computational burden
+- For visualization: Can downsample to q = (0, 0.5, 1, 1.5, 2) for quick exploration
+
+### Does read depth affect entropy estimates?
+
+Yes. TSENAT uses **automatic pseudocount selection** to handle this:
+- Pseudocounts scale with library size
+- Ensures entropy estimates are comparable across samples with different sequencing depths
+- Breaks ties in rank-based validation methods
+- Standard practice in RNA-seq analysis
+
+### What's the difference between "consistent" vs. "scale-dependent" transcript switching?
+
+- **Consistent switching**: Same transcripts show high delta-influence values across all q-values
+  - Robust, scale-independent splicing shift
+  - Suggests coordinated regulatory mechanism
+
+- **Scale-dependent switching**: Different transcripts matter at different q-values
+  - Regulatory complexity varies by scale
+  - Different mechanisms for rare vs. abundant isoforms
+
+## Methodological Questions
+
+### What does "q × condition interaction" mean?
+
+It means the effect of **condition on entropy depends on which q-value** you examine:
+- At q=0.5, one condition might have higher entropy
+- At q=2, the pattern might reverse
+- This reveals **scale-dependent** biological signals
+
+### What's the advantage of Tsallis divergence over fold-change?
+
+**Fold-change**:
+- Simple 1:1 ratio between conditions
+- Misses complexity patterns
+- Unbounded (can be arbitrarily large)
+
+**Tsallis divergence**:
+- Information-theoretic distance between distributions
+- Captures pattern type (rare-driven, balanced, abundant-driven)
+- Symmetric [0] to asymmetric, bounded interpretability
+- Directly quantifies isoform complexity differences
+
+## Practical Workflow Questions
+
+### Where do I get transcript-level counts for this analysis?
+
+TSENAT expects **SALMON or Kallisto output** including:
+- NumReads: raw fragment counts per transcript
+- TPM: length- and library-normalized estimates
+- EffectiveLength: read-length corrected transcript length
+
+These are standard outputs from pseudoalignment-based quantification.
+
+
+## Interpretation Questions
+
+### A gene shows significant q×condition interaction but small effect size—should I report it?
+
+Guidelines for meaningful results:
+- **Effect size (Tsallis divergence) D > 0.1**: ~10% information divergence (notable)
+- **D > 0.2**: Substantial divergence (recommended threshold)
+- **Small D with p < 0.05**: May reflect statistical noise; prioritize high-effect genes
+
+Report both for completeness:
+- High-effect genes (D > 0.2): Strong biological signal
+- Borderline genes (D ~ 0.1, p < 0.05): Mention as candidates for validation
+
+
+## Troubleshooting
+
+### My results show no significant genes. What could be wrong?
+
+1. **Sample size too small**: Need ≥6-8 per group (check design)
+2. **Entropy differences too subtle**: Try visualizing q-curves manually
+3. **Filtering too stringent**: Relaxing filters may reveal additional patterns
+4. **Method choice**: Try alternative statistical method (GAM vs. LMM vs. SRH)
+5. **Data quality**: Run M-estimation QC; check for outlier samples
+
+### Some q-values give NA or NaN results. Why?
+
+Possible causes:
+1. **Zero counts**: Pseudocount selection should handle; check filtering
+2. **Numerical instability**: Very small divergence values can underflow
+3. **Method-specific**: SRH (rank test) may have issues with many tied values
+
+Solution: Ensure pseudocount regularization is enabled; try alternative statistical method.
+
+### Q-curves show erratic patterns. Is this normal?
+
+**Expected**: Small fluctuations in entropy across q-values
+**Concerning**: Large spikes or reversals
+- May indicate normalization issues
+- Check: Are housekeeping genes (GAPDH, ACTB) showing balanced patterns?
+- If not: Revisit pseudocount and filtering parameters
+
+### Can I use TSENAT with very long reads (PacBio, Nanopore)?
+
+Conceptually yes, but:
+- TSENAT expects pseudoalignment-based quantification (SALMON, Kallisto)
+- Long reads typically use alignment-based tools (Minimap2, etc.)
+- Workflow would need adaptation to parse alignment-based counts
+- Overall approach (entropy framework) remains valid