gene_universe contains 1,539 gene_symbols mapping to multiple Ensembl IDs
(3,033 excess). Non-canonical IDs lack data in some evidence tables, causing
weighted_sum/available_weight to inflate their composite scores.
Fix: after scoring SQL, keep one row per gene_symbol with the most
evidence_count (tiebreak by composite_score DESC). 22,604 → 19,555 genes.
Results: HIGH 82→4, all top 20 now have 5-6 layers with expression data.
Validation PASSED (CDH23 98.3rd percentile, median known gene 83.3%).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three bugs prevented gnomAD and expression data from contributing to scores:
1. gnomAD COLUMN_VARIANTS mapped "gene" (HGNC symbol) to gene_id instead of
gene_symbol, causing JOIN miss with gene_universe (Ensembl IDs)
2. Expression HPA data was fetched but never merged (lf_hpa unused)
3. GTEx versioned Ensembl IDs (ENSG*.5) didn't match gene_universe
Results: gnomAD 78.5% coverage, expression 87.4%, 19946 genes with ≥4 layers.
HIGH tier refined from 44 → 18 candidates. Validation PASSED (CDH23 96.5th pctl).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- scoring/integration.py: Use CTEs with GROUP BY gene_id to deduplicate
all evidence tables before LEFT JOIN (gnomAD had 211K rows for 18K
genes due to per-transcript data; annotation/localization also had dupes)
- literature/transform.py: Deduplicate gene_symbols before PubMed queries
to avoid querying the same symbol multiple times
- evidence_cmd.py: Fix Polars .alias() error in literature summary display
- Updated report results with full 6-layer scoring (44 HIGH, 7268 MEDIUM)
- Validation PASSED: known Usher genes median percentile 86.9%
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Revised .gitignore to ignore raw data/cache but track data/report/
(candidates TSV/Parquet, plots, reproducibility metadata)
- Added zh↔en cross-links between README.md and README.en.md
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
