Feature Extraction for Non-Coding RNA: A Survey and Homology-Controlled Re-Evaluation

# ncRNA-FEB (reduced implementation) Pipeline for Section 10 of "Feature Extraction for Non-Coding RNA: A Survey and Homology-Controlled Re-Evaluation". ## Run order 1. `cluster.py` - greedy incremental clustering at 80% identity (edlib + 4-mer prefilter) 2. `features.py` - composition, 3-mer, thermodynamic (ViennaRNA), hybrid; records extraction cost 3. `evaluate.py` - evaluates all four under regimes A (published split), B (duplicates removed), C (cluster-disjoint 5-fold), with a fixed random-forest head 4. `stats.py` - repeats C with a second head (multinomial logistic regression) and computes 95% percentile bootstrap CIs over 2000 resamples of CLUSTERS (not sequences), plus paired differences between representations 5. `capacity.py` - capacity ladder: 10 models from regularised linear to 1-NN, testing whether leakage inflation scales with training-set fit