- Created 06-03-SUMMARY.md documenting plan execution - Updated STATE.md: - Current Position: Phase 6 COMPLETE (21/21 plans) - Performance Metrics: Phase 06 3/3 plans, 10 min total, 3.3 min/plan avg - Added decisions for comprehensive validation report and weight tuning recommendations - Session Continuity: Stopped at 06-03-PLAN.md completion Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
16 KiB
phase, plan, subsystem, tags, dependency_graph, tech_stack, key_files, decisions, metrics
| phase | plan | subsystem | tags | dependency_graph | tech_stack | key_files | decisions | metrics | ||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 06-validation | 03 | validation |
|
|
|
|
|
|
Phase 6 Plan 03: Comprehensive Validation Report & CLI Summary
CLI validate command orchestrating full validation pipeline (positive controls, negative controls, sensitivity analysis) with comprehensive Markdown report and weight tuning recommendations.
Tasks Completed
Task 1: Create comprehensive validation report and CLI validate command
Status: Complete
Commit: 10f19f8
Files: src/usher_pipeline/scoring/validation_report.py, src/usher_pipeline/cli/validate_cmd.py, src/usher_pipeline/cli/main.py, src/usher_pipeline/scoring/init.py
Created validation_report.py with comprehensive report generation:
-
generate_comprehensive_validation_report(): Multi-section Markdown report combining all three validation prongs
- Section 1: Positive Control Validation (median percentile, recall@k table, per-source breakdown, pass/fail)
- Section 2: Negative Control Validation (median percentile, top quartile count, in-HIGH-tier count, pass/fail)
- Section 3: Sensitivity Analysis (Spearman rho table by layer × delta, stability verdict, most/least sensitive layers)
- Section 4: Overall Validation Summary (all-pass/partial-fail/fail verdict with interpretation)
- Section 5: Weight Tuning Recommendations (targeted suggestions based on validation results)
-
recommend_weight_tuning(): Analyzes validation results and provides weight adjustment guidance
- All validations pass → "Current weights are validated. No tuning recommended."
- Positive controls fail → Suggest increasing weights for layers where known genes score highly
- Negative controls fail → Suggest examining layers boosting housekeeping genes, reducing generic layer weights
- Sensitivity unstable → Identify most sensitive layer, suggest reducing its weight
- CRITICAL WARNING: Documents circular validation risk (post-validation tuning invalidates controls)
- Provides best practices: independent validation set, document rationale, prefer a priori weights
-
save_validation_report(): Persists report to file with parent directory creation
Created validate_cmd.py CLI command following score_cmd.py pattern:
-
Click command
validatewith options:--force: Re-run even if validation checkpoint exists--skip-sensitivity: Skip sensitivity analysis for faster iteration--output-dir: Custom output directory (default: {data_dir}/validation)--top-n: Top N genes for sensitivity analysis (default: 100)
-
Pipeline steps:
- Load configuration and initialize store
- Check scored_genes checkpoint exists (error if not - must run
scorefirst) - Run positive control validation (validate_positive_controls_extended)
- Run negative control validation (validate_negative_controls)
- Run sensitivity analysis (unless --skip-sensitivity) - run_sensitivity_analysis + summarize_sensitivity
- Generate comprehensive validation report (generate_comprehensive_validation_report)
- Save report to output_dir/validation_report.md and provenance sidecar
-
Styled output with click.echo patterns (green for success, yellow for warnings, red for errors, bold for step headers)
-
Provenance tracking: record_step for each validation phase with metrics
-
Final summary: displays overall pass/fail, recall@10%, housekeeping median percentile, sensitivity stability
Updated main.py:
- Imported validate from validate_cmd
- Added
cli.add_command(validate)following existing pattern
Updated scoring.init.py:
- Added validation_report imports: generate_comprehensive_validation_report, recommend_weight_tuning, save_validation_report
- Added all 3 functions to all exports
Verification: Both verification commands passed:
python -c "from usher_pipeline.cli.validate_cmd import validate; print(f'Command name: {validate.name}'); print('OK')"→ OKpython -c "from usher_pipeline.cli.main import cli; assert 'validate' in cli.commands"→ OK
Task 2: Create unit tests for all validation modules
Status: Complete
Commit: 5879ae9
Files: tests/test_validation.py
Created test_validation.py with 13 comprehensive tests using synthetic DuckDB data:
Test helper:
- create_synthetic_scored_db(): Creates DuckDB with gene_universe (20 genes) and scored_genes table
- Designed scores ensure known cilia genes (MYO7A, IFT88, BBS1) get high scores (0.85-0.92)
- Housekeeping genes (GAPDH, ACTB, RPL13A) get low scores (0.12-0.20)
- Filler genes get mid-range scores (0.35-0.58)
- Includes all 6 layer scores and quality_flag
- Creates known_genes table with 3 genes (1 OMIM, 2 SYSCILIA)
Tests for negative controls (4 tests):
-
test_compile_housekeeping_genes_structure: Verifies compile_housekeeping_genes() returns DataFrame with 13 genes, correct columns (gene_symbol, source, confidence), all confidence=HIGH, all source=literature_validated
-
test_compile_housekeeping_genes_known_genes_present: Asserts GAPDH, ACTB, RPL13A, TBP are in gene_symbol column
-
test_validate_negative_controls_with_synthetic_data: Uses synthetic DB where housekeeping genes score low, asserts validation_passed=True, median_percentile < 0.5
-
test_validate_negative_controls_inverted_logic: Creates DB where housekeeping genes score HIGH (artificial scenario), asserts validation_passed=False
Tests for recall@k (1 test): 5. test_compute_recall_at_k: Uses synthetic DB, asserts recall@k returns dict with recalls_absolute and recalls_percentage keys. With 3 known genes in dataset (out of 38 total from compile_known_genes), recall@100 = 3/38 = 0.0789
Tests for weight perturbation (3 tests): 6. test_perturb_weight_renormalizes: Perturbs gnomad by +0.10, asserts weights still sum to 1.0 within 1e-6 tolerance
-
test_perturb_weight_large_negative: Perturbs by -0.25 (more than weight value), asserts weight >= 0.0 (clamped) and sum = 1.0
-
test_perturb_weight_invalid_layer: Asserts perturb_weight with layer="nonexistent" raises ValueError
Tests for validation report (5 tests): 9. test_generate_comprehensive_validation_report_format: Passes mock metrics dicts, asserts report contains expected sections ("Positive Control", "Negative Control", "Sensitivity Analysis", "Weight Tuning")
-
test_recommend_weight_tuning_all_pass: Passes metrics indicating all validations pass, asserts response contains "No tuning recommended"
-
test_recommend_weight_tuning_positive_fail: Passes metrics with positive controls failed, asserts response contains "Known Gene Ranking Issue" or "Positive Control"
-
test_recommend_weight_tuning_negative_fail: Passes metrics with negative controls failed, asserts response contains "Housekeeping" or "Negative Control"
-
test_recommend_weight_tuning_sensitivity_fail: Passes metrics with sensitivity unstable, asserts response contains "Sensitivity" or "gnomad"
Verification: All 13 tests passed:
tests/test_validation.py::test_compile_housekeeping_genes_structure PASSED
tests/test_validation.py::test_compile_housekeeping_genes_known_genes_present PASSED
tests/test_validation.py::test_validate_negative_controls_with_synthetic_data PASSED
tests/test_validation.py::test_validate_negative_controls_inverted_logic PASSED
tests/test_validation.py::test_compute_recall_at_k PASSED
tests/test_validation.py::test_perturb_weight_renormalizes PASSED
tests/test_validation.py::test_perturb_weight_large_negative PASSED
tests/test_validation.py::test_perturb_weight_invalid_layer PASSED
tests/test_validation.py::test_generate_comprehensive_validation_report_format PASSED
tests/test_validation.py::test_recommend_weight_tuning_all_pass PASSED
tests/test_validation.py::test_recommend_weight_tuning_positive_fail PASSED
tests/test_validation.py::test_recommend_weight_tuning_negative_fail PASSED
tests/test_validation.py::test_recommend_weight_tuning_sensitivity_fail PASSED
======================== 13 passed, 1 warning in 0.79s =========================
Deviations from Plan
None - plan executed exactly as written.
Verification Results
All verification checks passed:
python -c "from usher_pipeline.cli.validate_cmd import validate; print(f'Command name: {validate.name}'); print('OK')"→ OKpython -c "from usher_pipeline.cli.main import cli; assert 'validate' in cli.commands"→ OKpython -m pytest tests/test_validation.py -v→ 13 passed, 0 failed
Success Criteria
- CLI
validatecommand runs positive + negative + sensitivity validations and generates comprehensive report - Validation report includes all three prongs with pass/fail verdicts and weight tuning recommendations
- Unit tests cover negative controls, recall@k, perturbation, and report generation
- All tests pass with synthetic data
- validate command registered in main CLI
Key Files
Created
-
src/usher_pipeline/scoring/validation_report.py (410 lines)
- Comprehensive validation report generation combining all three validation prongs
- Exports: generate_comprehensive_validation_report, recommend_weight_tuning, save_validation_report
-
src/usher_pipeline/cli/validate_cmd.py (408 lines)
- CLI validate command orchestrating full validation pipeline
- Exports: validate (Click command)
-
tests/test_validation.py (478 lines)
- Unit tests for negative controls, recall@k, sensitivity, and validation report
- 13 tests with synthetic DuckDB fixture
Modified
-
src/usher_pipeline/cli/main.py (+2 lines)
- Added validate command import and registration
-
src/usher_pipeline/scoring/init.py (+7 lines)
- Added validation_report module exports
Integration Points
Depends on:
- Phase 06-01: Negative control validation (validate_negative_controls) and positive control validation (validate_positive_controls_extended, compute_recall_at_k)
- Phase 06-02: Sensitivity analysis (run_sensitivity_analysis, summarize_sensitivity)
- Phase 04-02: scored_genes checkpoint (validation requires scoring to be complete)
Provides:
- Comprehensive validation report combining all three validation prongs
- CLI
validatecommand for user-facing validation workflow - Unit test coverage for all validation modules
Affects:
- Phase 6 completion: This is the final plan in validation phase
- User workflow: Provides
usher-pipeline validatecommand for validation step
Technical Notes
Comprehensive Validation Report Design:
The report combines three complementary validation approaches:
- Positive controls (Plan 06-01): Known genes should rank high → validates sensitivity
- Negative controls (Plan 06-01): Housekeeping genes should rank low → validates specificity
- Sensitivity analysis (Plan 06-02): Rankings stable under perturbations → validates robustness
If all three pass: scoring system is sensitive, specific, and robust.
Overall Validation Verdict Logic:
- All pass → "ALL VALIDATIONS PASSED ✓" (scientifically defensible)
- Pos + Neg pass, Sensitivity fail → "PARTIAL PASS (Sensitivity Unstable)" (directionally correct but may need weight tuning)
- Pos pass, Neg fail → "PARTIAL PASS (Specificity Issue)" (sensitive but not specific)
- Pos fail → "VALIDATION FAILED ✗" (fundamental scoring issues)
Weight Tuning Recommendations Philosophy:
Recommendations are guidance, not automatic actions. They suggest:
- Which layers to adjust (increase/decrease weights)
- Why adjustments are needed (based on validation failures)
- How to interpret failures (specificity vs sensitivity vs stability)
CRITICAL WARNING included in all recommendations:
- Post-validation tuning introduces circular validation risk
- If weights are tuned based on validation results, those same controls cannot validate the tuned weights
- Best practices: independent validation set, document rationale, prefer a priori weights
This prevents the pitfall identified in 06-RESEARCH.md: "Tuning weights to maximize known gene recall, then using known gene recall as validation."
CLI validate Command Design:
Follows established pattern from score_cmd.py:
- Click command with options (--force, --skip-sensitivity, --output-dir, --top-n)
- Step-by-step pipeline with styled output (bold headers, colored status)
- Checkpoint-restart support (skips if validation_report.md exists unless --force)
- Provenance tracking for all steps (record_step for each validation phase)
- Final summary with overall status and key metrics
- Error handling with sys.exit(1) on failures
Test Design Philosophy:
All tests use synthetic DuckDB data with designed score patterns:
- Known genes get high scores (0.85-0.92) → positive controls should pass
- Housekeeping genes get low scores (0.12-0.20) → negative controls should pass
- Deterministic scores enable precise assertions
Tests cover:
- Happy path: Validations pass with expected data
- Inverted logic: Validations fail when data is wrong (housekeeping genes high)
- Edge cases: Large negative perturbations, invalid layer names
- Format verification: Report contains expected sections
- Recommendation logic: Different tuning suggestions for different failure modes
Usage Pattern:
# Full validation pipeline
usher-pipeline validate
# Skip sensitivity analysis (faster iteration)
usher-pipeline validate --skip-sensitivity
# Custom output directory
usher-pipeline validate --output-dir results/validation
# More genes for sensitivity (default 100)
usher-pipeline validate --top-n 200
# Force re-run
usher-pipeline validate --force
Expected Workflow:
- User runs
usher-pipeline score(Phase 04-03) - User runs
usher-pipeline validate(this plan) - User reviews validation report at {data_dir}/validation/validation_report.md
- If all pass: proceed to candidate prioritization
- If failures: review weight tuning recommendations, adjust weights with biological justification, re-run
Phase 6 Completion:
This plan completes Phase 6 (Validation). All three plans executed:
- 06-01: Negative controls and recall@k (2 min)
- 06-02: Sensitivity analysis (3 min)
- 06-03: Comprehensive validation report and CLI (5 min)
Total Phase 6 duration: 10 minutes across 3 plans.
Self-Check: PASSED
Created files verified:
- src/usher_pipeline/scoring/validation_report.py exists (410 lines)
- src/usher_pipeline/cli/validate_cmd.py exists (408 lines)
- tests/test_validation.py exists (478 lines)
Modified files verified:
- src/usher_pipeline/cli/main.py updated with validate import and registration
- src/usher_pipeline/scoring/init.py updated with validation_report exports
Commits verified:
10f19f8: Task 1 commit exists (comprehensive validation report and CLI validate command)5879ae9: Task 2 commit exists (unit tests for all validation modules)
Functionality verified:
- validate command imports correctly (Command name: validate, OK)
- validate registered in CLI (CLI commands includes 'validate', OK)
- All 13 tests pass (pytest reports 13 passed, 0 failed)
All claims in summary verified against actual implementation.