PGx validation on GSE165845: WGS-derived star truth, 2-fold masks, array-resolvable concordance

- scripts/pgx/wgs_star_truth.py: star alleles from the 1000G 30x phased panel with the DB definitions
  (wildcard-aware haplotype assignment; checked against GeT-RM on the demo CEPH individuals).
- validate_star_getrm.py: --all-genes, --individuals, --coverage (projection onto what the array can see);
  replicate_concordance.py.
- 120 individuals x 3 GSA v1 arrays: array-resolvable 1637/1710 (95.7 %), top 1223/1800 (chip lacks many
  defining variants), GeT-RM subset 28/35, replicates 99.2 % without the failed array (V2_pgx_star.md, K-029).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01814GKVzZVL3FuF3DeiZXhv

docs/known_unknowns.md

@@ -4,7 +4,7 @@ Plan §20 lists the open questions; this file records what the implementation ha
 
 | ID | Date | Topic | Finding | Status |
 |---|---|---|---|---|
-| K-029 | 2026-09-29 | PGx star alleles (v2) | (1) PGX-012 error model: ε(GS) = 0.5·(1 − GS)² was replaced by ε = 1e-3 (GS ≥ 0.15) / 0.05 — after probe masking 0 of 9 740 GSA v4 PGx calls disagreed with 1000G 30x at any GS, and the old model made correct low-GS calls lose to the prior (spec amended). (2) PGX-010 masks derived from 14 individuals (spec: ≥ 100); a variant absent from the filtered 1000G panel counts as hom-ref only if rare (CYP2B6 rs2279343 is filtered out, not absent). (3) CPIC core-allele definitions use IUPAC wildcards; alleles without definitions (NUDT15 *2, structural CYP2D6) are not callable from SNVs. (4) GeT-RM uses legacy UGT1A1/SLCO1B1 names. Result vs GeT-RM (16 CEPH): 100/109 gene calls, CYP2D6/2C19/2C9 41/45. (5) DPWG: no public machine-readable genotype→phenotype mapping; CPIC mappings are used for genes CPIC covers (as ClinPGx does), DPWG-only genes (CYP1A2, CYP3A4, F5) get no phenotype. (6) GeT-RM definitions are older than CPIC/PharmVar for UGT1A1, SLCO1B1, CYP1A2 and CYP2C8 (e.g. PharmVar CYP2C8 *19 = rs2071426 alone). Result vs GeT-RM (16 CEPH, 21 genes): 154/168 comparable gene calls. | Open: larger/independent truth set; KNMP DPWG tables if they become available (`docs/validation/V2_pgx_star.md`). |
+| K-029 | 2026-09-29 | PGx star alleles (v2) | (1) PGX-012 error model: ε(GS) = 0.5·(1 − GS)² was replaced by ε = 1e-3 (GS ≥ 0.15) / 0.05 — after probe masking 0 of 9 740 GSA v4 PGx calls disagreed with 1000G 30x at any GS, and the old model made correct low-GS calls lose to the prior (spec amended). (2) PGX-010 masks derived from 14 individuals (spec: ≥ 100); a variant absent from the filtered 1000G panel counts as hom-ref only if rare (CYP2B6 rs2279343 is filtered out, not absent). (3) CPIC core-allele definitions use IUPAC wildcards; alleles without definitions (NUDT15 *2, structural CYP2D6) are not callable from SNVs. (4) GeT-RM uses legacy UGT1A1/SLCO1B1 names. Result vs GeT-RM (16 CEPH): 100/109 gene calls, CYP2D6/2C19/2C9 41/45. (5) DPWG: no public machine-readable genotype→phenotype mapping; CPIC mappings are used for genes CPIC covers (as ClinPGx does), DPWG-only genes (CYP1A2, CYP3A4, F5) get no phenotype. (6) GeT-RM definitions are older than CPIC/PharmVar for UGT1A1, SLCO1B1, CYP1A2 and CYP2C8 (e.g. PharmVar CYP2C8 *19 = rs2071426 alone). Result vs GeT-RM (16 CEPH, 21 genes): 154/168 comparable gene calls. Independent check on GSE165845 (GSA v1 C2, 120 individuals × 3, WGS-derived truth, 2-fold masks): 95.7 % array-resolvable concordance, 67.9 % top-solution (the chip lacks many defining variants), replicate concordance 99.2 % without the failed array. | Open: KNMP DPWG tables if they become available (`docs/validation/V2_pgx_star.md`). |
 | K-028 | 2026-09-29 | PGx CN (v2) | Deviations from PCN-009: the per-probe reference is the training cohort's median stored in the model (a calling-batch median is not CN 2 where deletions are common, e.g. GSTM1 null ≈ 50 %); no GC-correction step; the CN mixture is ours: µ_c = δ + γ·log2((c+1)/3) per region with a free lower CN-0 component and Hardy–Weinberg-constrained weights (probe compression differs by region: γ ≈ 0.19 CYP2D6.p5 … 0.56 GSTT1). Interim validation on the GSA v4 PGx demo — whose 36 arrays turned out to be 30 CEPH/1000G individuals — against NYGC 30x SV RD_CN: 224/240 region calls (GSTM1, UGT2B17 30/30), but in-sample (trained on the same arrays, < 96 samples, priors chosen with this truth in view). | Open: validate on an independent GSA-PGx cohort; CYP2D6 via GeT-RM star alleles (`docs/validation/V2_pgx_cn.md`). |
 | K-027 | 2026-09-29 | CN/LOH (v2) | CYTO-014 noise gate: the summary `logr_dev` also measures cluster-file mismatch (GSE165845: 0.246, all samples `SampleQuality`) while the LRR used after the batch reference has SD ≈ 0.1; the gate now uses the SD of the normalised LRR (the JSON still reports the summary value). Spec: built-in mask/cytobands from UCSC; the annotation bundle uses UCSC ncbiRefSeq instead of RefSeq GFF + Ensembl GTF + HGNC. Real-data result: constitutional DEL/DUP calls 74–77 % in NYGC truth, mosaic calls mostly false positives. Pi tuning: mosaic prior / switch distance barely matter; CYTO-041 (noise-normalised BAF split, complete-loss promotion) cuts PASS mosaic calls by ~30 % without losing implanted ≥ 15 % mosaics. Two limits remain: (1) without phasing, < 20 % mosaic losses are not separable from regional BAF noise; (2) common CNVs (carrier frequency > 20 %) are almost never detected (1/236 DEL) because the median-LRR batch reference absorbs them; rare DEL (< 5 %) 38 %. | Open: rare-event sensitivity as primary metric; phasing-based mosaic BAF (`docs/validation/V2_cn.md`). |
 | K-026 | 2026-09-29 | Methylation QC (v2) | Spec deviations: vendor EPIC manifests put a 5th field on some control lines (≥ 4 fields accepted); detection p follows sesame's pOOBAH (1 − max(F(M), F(U))) rather than the ambiguous "1 − ECDF(M + U)"; minfi-style `Sentrix_ID`/`Sentrix_Position` sheet keys accepted. Against sesame 1.28.1 on minfiDataEPIC: β r = 0.99998 (max \|Δβ\| 0.011), detection calls 99.1 % identical, noob signals within 0.5 % (median) / 5 % (max). The systematic remainder is unexplained; it is not the negative controls (sesame did not attach the supplied control table). | Open; `docs/validation/V2_methylation.md`. Next step: dump sesame's background vectors and noob parameters to find the difference. |

docs/validation/V2_pgx_star.md

@@ -101,3 +101,54 @@ so the per-call error is 1e-3 (GS ≥ 0.15) and errors are handled by the probe
   for genes CPIC covers, so those genes reuse the CPIC tables (flag `dpwg_via_cpic_mapping`); DPWG-only genes
   (CYP1A2, CYP3A4, F5) have no public machine-readable mapping and get phenotype `null` with the flag
   `dpwg_mapping_not_available` (K-029).
+
+## Independent validation: GSE165845 (GSA v1 C2, 120 1000 Genomes individuals × 3 arrays)
+
+A second cohort, independent of the demo used for development (2026-09-29, all on the Pi, $0):
+
+* **Arrays:** the 360 GSE165845 GTCs (GSA-MD v1 chips, GSA v1 C2 manifest; K-024) → SNV VCFs; no targeted CN probes
+  on this chip, so star alleles are called with `--no-cn`. The chip carries little PGx content: 292 probe-to-variant
+  links (GSA v4 PGx: 5 280); `pgx coverage`: 495 alleles callable, 136 array-equivalent, 1 129 not covered — e.g.
+  no CYP2D6 *2/*4/*10 probes (rs16947, rs3892097, rs1065852 are absent from the manifest).
+* **Truth:** star alleles derived from the 1000G 30x phased panel with the same DB definitions
+  (`scripts/pgx/wgs_star_truth.py`; SNV/indel level — CYP2D6 *5, xN and hybrids invisible; CYP2B6 left out because
+  its common core variant rs2279343 is filtered from the panel). The derivation was checked against GeT-RM on the
+  demo's CEPH individuals (CYP2D6 14/15, CYP2C9/CYP3A4/CYP3A5 15/15, TPMT 16/16; the rest are nomenclature changes).
+* **Masks:** 2-fold — individuals split in halves; each half is called with the mask derived from the other half
+  (A → 0 probes masked, B → 1). After masking the calls agree with 1000G at 99 % or better in every GS bin.
+* **Metrics:** *top* = the top solution equals the WGS diplotype; *array-resolvable* = both diplotypes projected
+  onto the alleles this array can tell apart (`pgx coverage`: callable → itself, equivalent_to:X → X, not covered →
+  the reference's class) — the fair measure for a chip that lacks a gene's defining variants.
+
+| Gene | Top = WGS | Truth within feasible equivalents | Array-resolvable |
+|---|---|---|---|
+| CYP1A2 | 18/120 | 72/120 | 120/120 |
+| CYP2A13 | 97/120 | 119/120 | 117/120 |
+| CYP2A6 | 73/120 | 112/120 | 117/120 |
+| CYP2C19 | 100/120 | 116/120 | 110/120 |
+| CYP2C8 | 114/120 | 120/120 | 115/120 |
+| CYP2C9 | 95/120 | 108/120 | 116/120 |
+| CYP2D6 (SNV level) | 32/120 | 54/120 | 111/120 |
+| CYP3A4 | 112/120 | 117/120 | 117/120 |
+| CYP3A5 | 120/120 | 120/120 | 120/120 |
+| CYP4F2 | 43/120 | 86/120 | 96/120 |
+| NAT2 | 95/120 | 105/120 | 120/120 |
+| NUDT15 | 120/120 | 120/120 | 120/120 |
+| SLCO1B1 | 31/120 | 87/120 | 109/120 |
+| TPMT | 107/120 | 115/120 | 102/103 |
+| UGT1A1 | 66/120 | 120/120 | 47/47 |
+| **total** | **1223/1800 (67.9 %)** | | **1637/1710 (95.7 %)** |
+
+* Top-solution concordance is low where the chip lacks the defining variants: the call is then the most probable
+  member of an equivalence class the array cannot split (CYP2D6 *1/*2/*10, SLCO1B1 *1/*37, CYP1A2), so the prior
+  decides. Projected onto what the array can see, 95.7 % agree (half A 815/854, half B 822/856).
+* CYP4F2 (96/120) is the weakest: the array genotypes at its covered sites match 1000G exactly (rs3093200 180/180),
+  so the residue is definitional — *5 and *23 share the only covered variant, and the projection cannot separate
+  alleles whose other defining variants are off-array.
+* **GeT-RM** (7 of the 120 individuals, older consensus): 28/35 top, array-resolvable 29/30.
+* **Replicates:** the top genotype of all 70 genes agrees between replicate arrays of one individual in 24 916 of
+  25 200 pairs (98.9 %); 87 of the 284 discordant pairs come from NA20362's failed replicate (call rate 0.76, K-024);
+  without it 99.2 %.
+
+Files: `docs/validation/V2_pgx_star_gse165845.json`; scripts `wgs_star_truth.py`, `validate_star_getrm.py`
+(`--all-genes --individuals --coverage`), `replicate_concordance.py`.