pen-stack 6.9.0__tar.gz → 6.9.1__tar.gz

{pen_stack-6.9.0 → pen_stack-6.9.1}/CHANGELOG.md

@@ -3,6 +3,32 @@
 All notable changes to PEN-STACK are documented here. This file follows
 [Keep a Changelog](https://keepachangelog.com/) and the program's phase structure.
 
+## [6.9.1] - 2026-06-20 - Real NetMHCIIpan-4.0 / NetMHCpan-4.1 MHC epitope load (replaces the v6.9.0 proxy)
+
+**PATCH — rigour upgrade.** v6.9.0's MHC-II axis was a documented *heuristic proxy* (P1-anchor density). The
+gold-standard licensed predictors were already on the VM — so v6.9.1 computes the MHC-II epitope load with **real
+NetMHCIIpan-4.0** (and MHC-I with **NetMHCpan-4.1**) over a frequent HLA panel, and re-grounds the axis on the real
+values. Honest: the licensed binaries are **never committed/distributed** — only the **derived fractions** are
+cached (`configs/mhc_epitope_oracle.yaml`), exactly like the v5.3 MHCflurry cache.
+
+### Changed
+- `pen_stack/planner/immune_mhc2.py` — `mhc2_epitope_load(seq, name)` now uses the **real NetMHCIIpan-4.0 EL
+  %Rank≤2** cache (over 7 frequent HLA-II alleles) when the antigen is cached; the documented promiscuous-binder
+  proxy remains only as the **offline/CI fallback** for uncached sequences. `real_mhc2_load(name)` exposes the cache.
+- `ada_risk` + `immune_profile` (writer-as-antigen) + `benchmarks/immuno` thread the antigen name → real values.
+- `configs/mhc_epitope_oracle.yaml` (committed) — derived epitope fractions for the writer + capsid antigens.
+
+### Result (real tool is more discriminating than the proxy)
+- Real NetMHCIIpan-4.0 strong-binder fraction: SpCas9 **0.153**, Bxb1 **0.152**, ISCro4 **0.112**, AAV2 0.114 vs
+  **human albumin (self) 0.066 — the lowest**. The gold-standard tool shows the self protein has a genuinely lower
+  MHC-II load and the foreign writers higher — a signal the heuristic proxy flattened (~0.08–0.10 for all). The
+  immunogenic-vs-tolerated recovery is unchanged (foreign ≫ self) but now on real predictions.
+
+### How it ran (no host install; per the VM Docker rule)
+- `penmhc:tools` (debian + tcsh + gawk + perl) with `~/netmhc` (the licensed tools) **mounted**, NMHOME fixed at
+  runtime; `scratch/v691_mhc_compute.py` runs both predictors and writes the derived cache. The axis stays a 🟡
+  population-level proxy (frequent-HLA panel, not a patient-HLA magnitude — a known-unknown).
+
 ## [6.9.0] - 2026-06-20 - PEN-IMMUNE: MHC-II/CD4 + ADA + writer-as-antigen
 
 **MINOR feature release.** Extends the immune profile from CD8/MHC-I-only to a full T-cell profile — MHC-I +

{pen_stack-6.9.0 → pen_stack-6.9.1}/CITATION.cff

@@ -1,7 +1,7 @@
 cff-version: 1.2.0
 message: "If you use PEN-STACK, please cite it as below."
 title: "PEN-STACK: open infrastructure for genome writing"
-version: 6.9.0
+version: 6.9.1
 date-released: 2026-06-20
 authors:
   - family-names: "Mahaboob Ali"

{pen_stack-6.9.0 → pen_stack-6.9.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pen-stack
-Version: 6.9.0
+Version: 6.9.1
 Summary: Open infrastructure for genome writing: the Writable Genome atlas, the Writer Atlas, and the Write Planner.
 Author-email: Anees Ahmed Mahaboob Ali <ahmedaneesm@gmail.com>
 License: MIT
@@ -90,7 +90,7 @@ every design against rule-grounded mechanism, reports calibrated confidence, cit
 [![codecov](https://codecov.io/gh/ahmedanees-m/pen-stack/branch/main/graph/badge.svg)](https://codecov.io/gh/ahmedanees-m/pen-stack)
 [![License: MIT](https://img.shields.io/badge/License-MIT-informational.svg)](LICENSE)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/)
-[![Version](https://img.shields.io/badge/version-6.9.0-blue.svg)](CHANGELOG.md)
+[![Version](https://img.shields.io/badge/version-6.9.1-blue.svg)](CHANGELOG.md)
 [![Status](https://img.shields.io/badge/status-1.0%20First%20Stable-success.svg)](docs/STABILITY.md)
 [![Tests](https://img.shields.io/badge/tests-378%20passing-success.svg)](tests/)
 [![Lint: ruff](https://img.shields.io/badge/lint-ruff-purple.svg)](https://github.com/astral-sh/ruff)
@@ -155,9 +155,12 @@ driver is **MHC-II / CD4 help → anti-drug antibodies (ADA)**, and the **writer
 (Cas9 elicits MHC-II-presented CD4 responses, Simhadri *Nat Commun* 2021; bridge recombinases / serine integrases
 are bacterial/phage) — yet Stage G scored only the capsid. v6.9 closes that gap, **never collapsing** the axes:
 
-- **A CD4/MHC-II epitope-load axis** (`planner/immune_mhc2.py`) — a grounded, dependency-free **promiscuous MHC-II
-  binder density** over the documented P1 hydrophobic anchor (Stern & Wiley 1994) + secondary pockets, scored over
-  **capsid AND writer** sequences. Population-level proxy (🟡), never a patient-HLA magnitude.
+- **A CD4/MHC-II epitope-load axis** (`planner/immune_mhc2.py`) — **v6.9.1 computes it with the real, licensed
+  NetMHCIIpan-4.0** (EL %Rank≤2 over a frequent HLA-II panel; MHC-I via NetMHCpan-4.1), scored over **capsid AND
+  writer** sequences; only the derived fractions are cached (`configs/mhc_epitope_oracle.yaml`, the binaries are
+  never distributed), with a documented promiscuous-binder proxy as the offline/CI fallback. Population-level proxy
+  (🟡), never a patient-HLA magnitude. *(v6.9.0 used the heuristic proxy; the real tool is more discriminating —
+  human-self albumin 0.066 vs foreign writers 0.11–0.15.)*
 - **An ADA-risk axis** (`planner/ada_risk.py`) — **ADA-risk = MHC-II epitope density × foreignness**, with a
   **self-tolerance filter** (JanusMatrix-style: self epitopes are tolerated; foreign drive ADA). It **recovers
   immunogenic-vs-tolerated**: the foreign writers (real UniProt SpCas9 / ISCro4 / Bxb1) score **above** the human

{pen_stack-6.9.0 → pen_stack-6.9.1}/README.md

@@ -15,7 +15,7 @@ every design against rule-grounded mechanism, reports calibrated confidence, cit
 [![codecov](https://codecov.io/gh/ahmedanees-m/pen-stack/branch/main/graph/badge.svg)](https://codecov.io/gh/ahmedanees-m/pen-stack)
 [![License: MIT](https://img.shields.io/badge/License-MIT-informational.svg)](LICENSE)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/)
-[![Version](https://img.shields.io/badge/version-6.9.0-blue.svg)](CHANGELOG.md)
+[![Version](https://img.shields.io/badge/version-6.9.1-blue.svg)](CHANGELOG.md)
 [![Status](https://img.shields.io/badge/status-1.0%20First%20Stable-success.svg)](docs/STABILITY.md)
 [![Tests](https://img.shields.io/badge/tests-378%20passing-success.svg)](tests/)
 [![Lint: ruff](https://img.shields.io/badge/lint-ruff-purple.svg)](https://github.com/astral-sh/ruff)
@@ -80,9 +80,12 @@ driver is **MHC-II / CD4 help → anti-drug antibodies (ADA)**, and the **writer
 (Cas9 elicits MHC-II-presented CD4 responses, Simhadri *Nat Commun* 2021; bridge recombinases / serine integrases
 are bacterial/phage) — yet Stage G scored only the capsid. v6.9 closes that gap, **never collapsing** the axes:
 
-- **A CD4/MHC-II epitope-load axis** (`planner/immune_mhc2.py`) — a grounded, dependency-free **promiscuous MHC-II
-  binder density** over the documented P1 hydrophobic anchor (Stern & Wiley 1994) + secondary pockets, scored over
-  **capsid AND writer** sequences. Population-level proxy (🟡), never a patient-HLA magnitude.
+- **A CD4/MHC-II epitope-load axis** (`planner/immune_mhc2.py`) — **v6.9.1 computes it with the real, licensed
+  NetMHCIIpan-4.0** (EL %Rank≤2 over a frequent HLA-II panel; MHC-I via NetMHCpan-4.1), scored over **capsid AND
+  writer** sequences; only the derived fractions are cached (`configs/mhc_epitope_oracle.yaml`, the binaries are
+  never distributed), with a documented promiscuous-binder proxy as the offline/CI fallback. Population-level proxy
+  (🟡), never a patient-HLA magnitude. *(v6.9.0 used the heuristic proxy; the real tool is more discriminating —
+  human-self albumin 0.066 vs foreign writers 0.11–0.15.)*
 - **An ADA-risk axis** (`planner/ada_risk.py`) — **ADA-risk = MHC-II epitope density × foreignness**, with a
   **self-tolerance filter** (JanusMatrix-style: self epitopes are tolerated; foreign drive ADA). It **recovers
   immunogenic-vs-tolerated**: the foreign writers (real UniProt SpCas9 / ISCro4 / Bxb1) score **above** the human

pen_stack-6.9.1/configs/mhc_epitope_oracle.yaml

@@ -0,0 +1,239 @@
+version: 6.9.1
+method:
+  mhc2: NetMHCIIpan-4.0 EL %Rank<=2
+  mhc1: NetMHCpan-4.1 %Rank<=0.5
+  hla2_panel:
+  - DRB1_0101
+  - DRB1_0301
+  - DRB1_0401
+  - DRB1_0701
+  - DRB1_1101
+  - DRB1_1301
+  - DRB1_1501
+  hla1_panel:
+  - HLA-A01:01
+  - HLA-A02:01
+  - HLA-A03:01
+  - HLA-A11:01
+  - HLA-A24:02
+  - HLA-A26:01
+  - HLA-B07:02
+  - HLA-B08:01
+  - HLA-B15:01
+  - HLA-B27:05
+  - HLA-B40:01
+  - HLA-B58:01
+provenance_dois:
+- 10.1093/nar/gkac1029
+- 10.1093/nar/gkaa379
+note: population-level epitope load over a frequent HLA panel (NOT patient-HLA-specific);
+  licensed NetMHC binaries run locally, only derived fractions cached.
+mhc2:
+  SpCas9:
+    epitope_fraction_strong: 0.1529
+    immune_score: 0.8471
+    n_windows: 1354
+    n_strong: 207
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  ISCro4:
+    epitope_fraction_strong: 0.1122
+    immune_score: 0.8878
+    n_windows: 312
+    n_strong: 35
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  Bxb1:
+    epitope_fraction_strong: 0.152
+    immune_score: 0.848
+    n_windows: 487
+    n_strong: 74
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  HumanAlbumin:
+    epitope_fraction_strong: 0.0655
+    immune_score: 0.9345
+    n_windows: 595
+    n_strong: 39
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  AAV2_VP1:
+    epitope_fraction_strong: 0.1137
+    immune_score: 0.8863
+    n_windows: 721
+    n_strong: 82
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  Ad5_hexon:
+    epitope_fraction_strong: 0.1471
+    immune_score: 0.8529
+    n_windows: 938
+    n_strong: 138
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  VSVg_Indiana:
+    epitope_fraction_strong: 0.0684
+    immune_score: 0.9316
+    n_windows: 497
+    n_strong: 34
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  HSV1_gD:
+    epitope_fraction_strong: 0.1184
+    immune_score: 0.8816
+    n_windows: 380
+    n_strong: 45
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+  HSV1_gB:
+    epitope_fraction_strong: 0.1067
+    immune_score: 0.8933
+    n_windows: 890
+    n_strong: 95
+    alleles_used:
+    - DRB1_0101
+    - DRB1_0301
+    - DRB1_0401
+    - DRB1_0701
+    - DRB1_1101
+    - DRB1_1301
+    - DRB1_1501
+mhc1:
+  AAV2_VP1:
+    epitope_fraction_strong: 0.0935
+    immune_score: 0.9065
+    n_windows: 727
+    n_strong: 68
+    alleles_used:
+    - HLA-A01:01
+    - HLA-A02:01
+    - HLA-A03:01
+    - HLA-A11:01
+    - HLA-A24:02
+    - HLA-A26:01
+    - HLA-B07:02
+    - HLA-B08:01
+    - HLA-B15:01
+    - HLA-B27:05
+    - HLA-B40:01
+    - HLA-B58:01
+  Ad5_hexon:
+    epitope_fraction_strong: 0.1367
+    immune_score: 0.8633
+    n_windows: 944
+    n_strong: 129
+    alleles_used:
+    - HLA-A01:01
+    - HLA-A02:01
+    - HLA-A03:01
+    - HLA-A11:01
+    - HLA-A24:02
+    - HLA-A26:01
+    - HLA-B07:02
+    - HLA-B08:01
+    - HLA-B15:01
+    - HLA-B27:05
+    - HLA-B40:01
+    - HLA-B58:01
+  VSVg_Indiana:
+    epitope_fraction_strong: 0.1074
+    immune_score: 0.8926
+    n_windows: 503
+    n_strong: 54
+    alleles_used:
+    - HLA-A01:01
+    - HLA-A02:01
+    - HLA-A03:01
+    - HLA-A11:01
+    - HLA-A24:02
+    - HLA-A26:01
+    - HLA-B07:02
+    - HLA-B08:01
+    - HLA-B15:01
+    - HLA-B27:05
+    - HLA-B40:01
+    - HLA-B58:01
+  HSV1_gD:
+    epitope_fraction_strong: 0.1218
+    immune_score: 0.8782
+    n_windows: 386
+    n_strong: 47
+    alleles_used:
+    - HLA-A01:01
+    - HLA-A02:01
+    - HLA-A03:01
+    - HLA-A11:01
+    - HLA-A24:02
+    - HLA-A26:01
+    - HLA-B07:02
+    - HLA-B08:01
+    - HLA-B15:01
+    - HLA-B27:05
+    - HLA-B40:01
+    - HLA-B58:01
+  HSV1_gB:
+    epitope_fraction_strong: 0.115
+    immune_score: 0.885
+    n_windows: 896
+    n_strong: 103
+    alleles_used:
+    - HLA-A01:01
+    - HLA-A02:01
+    - HLA-A03:01
+    - HLA-A11:01
+    - HLA-A24:02
+    - HLA-A26:01
+    - HLA-B07:02
+    - HLA-B08:01
+    - HLA-B15:01
+    - HLA-B27:05
+    - HLA-B40:01
+    - HLA-B58:01

{pen_stack-6.9.0 → pen_stack-6.9.1}/docs/immune_profiler.md

@@ -6,7 +6,15 @@ adaptive axis was **CD8/MHC-I only** (capsid epitope load via MHCflurry). v6.9 a
 
 ## The MHC-II/CD4 axis (`planner/immune_mhc2.py`)
 
-A grounded, dependency-free **promiscuous MHC-II binder density**: MHC-II presents a 9-mer core in an open groove
+> **v6.9.1 — real predictor.** The MHC-II epitope load is computed with the **licensed NetMHCIIpan-4.0** (EL
+> %Rank≤2 over a frequent HLA-II panel; MHC-I via **NetMHCpan-4.1**), run inside `penmhc:tools` Docker on the VM
+> with the licensed tools **mounted** (never committed). Only the **derived fractions** are cached
+> (`configs/mhc_epitope_oracle.yaml`), exactly like the v5.3 MHCflurry cache. The documented proxy below remains as
+> the offline/CI fallback for sequences not in the cache. Real result: human-self albumin **0.066** vs foreign
+> writers **0.11–0.15** (the gold-standard tool discriminates self from foreign — the v6.9.0 proxy gave ~0.08–0.10
+> for all). Still a 🟡 population-level proxy (frequent-HLA panel, not a patient-HLA magnitude).
+
+A grounded, dependency-free **promiscuous MHC-II binder density** (the offline/CI fallback): MHC-II presents a 9-mer core in an open groove
 whose **P1 pocket is deep and hydrophobic** — the single dominant anchor (M/F/Y/W/L/I/V; Stern & Wiley, *Nature*
 1994), with secondary pockets at P4/P6/P9. A *promiscuous* epitope (binds many HLA-DR) has a strong P1 anchor +
 favorable secondaries (Southwood 1998). We count promiscuous-binder cores → an epitope-density proxy, computed over

{pen_stack-6.9.0 → pen_stack-6.9.1}/pen_stack/__init__.py

@@ -1,2 +1,2 @@
 """PEN-STACK v3.0 - open infrastructure for genome writing."""
-__version__ = "6.9.0"
+__version__ = "6.9.1"

{pen_stack-6.9.0 → pen_stack-6.9.1}/pen_stack/planner/ada_risk.py

@@ -52,10 +52,12 @@ def self_tolerance(seq: str, human_kmers: frozenset | None = None) -> dict:
             "reference": "bundled human self proteins (full human proteome substitutable on VM)"}
 
 
-def ada_risk(seq: str, origin: str | None = None, human_kmers: frozenset | None = None) -> dict:
-    """ADA-risk = MHC-II epitope density x foreignness, with the self-tolerance filter. Higher ada_risk_score =
-    MORE anti-drug-antibody risk; `ada_immune_score = 1 - ada_risk_score` (higher = safer) for profile parity."""
-    el = mhc2_epitope_load(seq)
+def ada_risk(seq: str, origin: str | None = None, human_kmers: frozenset | None = None,
+             name: str | None = None) -> dict:
+    """ADA-risk = MHC-II epitope density x foreignness, with the self-tolerance filter. Uses the REAL NetMHCIIpan-4.0
+    epitope density when the antigen `name` is cached (else the documented proxy). Higher ada_risk_score = MORE
+    anti-drug-antibody risk; `ada_immune_score = 1 - ada_risk_score` (higher = safer) for profile parity."""
+    el = mhc2_epitope_load(seq, name)
     st = self_tolerance(seq, human_kmers)
     sm = st["self_match_fraction"] or 0.0
     if origin == "self":
@@ -81,4 +83,5 @@ def ada_risk_named(name: str) -> dict:
     rec = writer_sequences().get(name)
     if not rec:
         return {"available": False, "note": f"no bundled sequence {name!r}"}
-    return {"available": True, "name": name, "family": rec.get("family"), **ada_risk(rec["seq"], rec.get("origin"))}
+    return {"available": True, "name": name, "family": rec.get("family"),
+            **ada_risk(rec["seq"], rec.get("origin"), name=name)}

{pen_stack-6.9.0 → pen_stack-6.9.1}/pen_stack/planner/immune_mhc2.py

@@ -53,8 +53,39 @@ def mhc2_binder_cores(seq: str) -> list[tuple[int, str, float]]:
     return out
 
 
-def mhc2_epitope_load(seq: str) -> dict:
-    """Sequence-intrinsic MHC-II promiscuous-binder density + an immune score (1 = least presentable), 🟡 proxy."""
+@lru_cache(maxsize=1)
+def _real_cache() -> dict:
+    """The REAL NetMHCIIpan-4.0 EL %Rank epitope-load cache (configs/mhc_epitope_oracle.yaml), computed over a
+    frequent HLA-II panel on the VM. Only the derived fractions are shipped (the licensed binary is never
+    distributed). Empty dict if the cache is absent (then the documented proxy is used)."""
+    try:
+        import yaml
+        return yaml.safe_load(resource("configs/mhc_epitope_oracle.yaml").read_text(encoding="utf-8")) or {}
+    except Exception:  # noqa: BLE001
+        return {}
+
+
+def real_mhc2_load(name: str) -> dict | None:
+    """The real NetMHCIIpan-4.0 epitope load for a bundled antigen by name, or None when not cached."""
+    rec = (_real_cache().get("mhc2") or {}).get(name)
+    if not rec:
+        return None
+    panel = (_real_cache().get("method") or {}).get("hla2_panel", [])
+    return {"epitope_density": rec["epitope_fraction_strong"], "mhc2_immune_score": rec["immune_score"],
+            "n_cores": rec.get("n_windows"), "n_promiscuous_binders": rec.get("n_strong"),
+            "method": "NetMHCIIpan-4.0 EL %Rank<=2 over a frequent HLA-II panel "
+                      f"({len(panel)} alleles): {', '.join(panel)}",
+            "status": "population-level (frequent-HLA panel; NetMHCIIpan-4.0 eluted-ligand), OOD-gated; NOT a "
+                      "patient-HLA-specific magnitude (known-unknown)", "backend": "netmhciipan_cache"}
+
+
+def mhc2_epitope_load(seq: str, name: str | None = None) -> dict:
+    """MHC-II epitope load. Uses the REAL NetMHCIIpan-4.0 cache when the antigen `name` is cached; otherwise the
+    documented promiscuous-binder PROXY (offline/CI fallback). Score: 1 = least presentable (MHC-I convention)."""
+    if name:
+        real = real_mhc2_load(name)
+        if real:
+            return real
     s = _clean(seq)
     n_cores = max(0, len(s) - 8)
     binders = mhc2_binder_cores(s)
@@ -64,9 +95,10 @@ def mhc2_epitope_load(seq: str) -> dict:
         "epitope_density": round(density, 4),
         "mhc2_immune_score": round(1.0 - min(density, 1.0), 4),       # 1 = least presentable (MHC-I convention)
         "method": "promiscuous MHC-II binder density (P1 hydrophobic anchor + P4/P6/P9 secondary pockets; "
-                  "Stern & Wiley 1994; Southwood 1998)",
+                  "Stern & Wiley 1994; Southwood 1998) — DOCUMENTED PROXY (offline/CI fallback when the "
+                  "NetMHCIIpan-4.0 cache is absent)",
         "status": "population-level sequence-intrinsic proxy (🟡); NOT a trained allele-specific predictor, NOT a "
-                  "patient-HLA-specific magnitude (known-unknown)",
+                  "patient-HLA-specific magnitude (known-unknown)", "backend": "proxy",
         "binder_cores": [c for _, c, _ in binders[:50]],
     }
 

{pen_stack-6.9.0 → pen_stack-6.9.1}/pen_stack/planner/immune_profile.py

@@ -83,8 +83,9 @@ def _writer_antigen_card(design: dict) -> dict | None:
     rec = writer_family_to_sequence(wf) if wf else None
     if not rec or not rec.get("seq"):
         return None
-    el = mhc2_epitope_load(rec["seq"])
-    ad = ada_risk(rec["seq"], rec.get("origin"))
+    nm = rec.get("name")
+    el = mhc2_epitope_load(rec["seq"], nm)            # real NetMHCIIpan-4.0 when cached, else proxy
+    ad = ada_risk(rec["seq"], rec.get("origin"), name=nm)
     return {"writer_family": wf, "representative": rec.get("name"), "accession": rec.get("accession"),
             "origin": rec.get("origin"), "is_foreign": rec.get("origin") == "foreign",
             "mhc2_immune_score": el["mhc2_immune_score"], "epitope_density": el["epitope_density"],

{pen_stack-6.9.0 → pen_stack-6.9.1}/pen_stack.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pen-stack
-Version: 6.9.0
+Version: 6.9.1
 Summary: Open infrastructure for genome writing: the Writable Genome atlas, the Writer Atlas, and the Write Planner.
 Author-email: Anees Ahmed Mahaboob Ali <ahmedaneesm@gmail.com>
 License: MIT
@@ -90,7 +90,7 @@ every design against rule-grounded mechanism, reports calibrated confidence, cit
 [![codecov](https://codecov.io/gh/ahmedanees-m/pen-stack/branch/main/graph/badge.svg)](https://codecov.io/gh/ahmedanees-m/pen-stack)
 [![License: MIT](https://img.shields.io/badge/License-MIT-informational.svg)](LICENSE)
 [![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-blue.svg)](https://www.python.org/)
-[![Version](https://img.shields.io/badge/version-6.9.0-blue.svg)](CHANGELOG.md)
+[![Version](https://img.shields.io/badge/version-6.9.1-blue.svg)](CHANGELOG.md)
 [![Status](https://img.shields.io/badge/status-1.0%20First%20Stable-success.svg)](docs/STABILITY.md)
 [![Tests](https://img.shields.io/badge/tests-378%20passing-success.svg)](tests/)
 [![Lint: ruff](https://img.shields.io/badge/lint-ruff-purple.svg)](https://github.com/astral-sh/ruff)
@@ -155,9 +155,12 @@ driver is **MHC-II / CD4 help → anti-drug antibodies (ADA)**, and the **writer
 (Cas9 elicits MHC-II-presented CD4 responses, Simhadri *Nat Commun* 2021; bridge recombinases / serine integrases
 are bacterial/phage) — yet Stage G scored only the capsid. v6.9 closes that gap, **never collapsing** the axes:
 
-- **A CD4/MHC-II epitope-load axis** (`planner/immune_mhc2.py`) — a grounded, dependency-free **promiscuous MHC-II
-  binder density** over the documented P1 hydrophobic anchor (Stern & Wiley 1994) + secondary pockets, scored over
-  **capsid AND writer** sequences. Population-level proxy (🟡), never a patient-HLA magnitude.
+- **A CD4/MHC-II epitope-load axis** (`planner/immune_mhc2.py`) — **v6.9.1 computes it with the real, licensed
+  NetMHCIIpan-4.0** (EL %Rank≤2 over a frequent HLA-II panel; MHC-I via NetMHCpan-4.1), scored over **capsid AND
+  writer** sequences; only the derived fractions are cached (`configs/mhc_epitope_oracle.yaml`, the binaries are
+  never distributed), with a documented promiscuous-binder proxy as the offline/CI fallback. Population-level proxy
+  (🟡), never a patient-HLA magnitude. *(v6.9.0 used the heuristic proxy; the real tool is more discriminating —
+  human-self albumin 0.066 vs foreign writers 0.11–0.15.)*
 - **An ADA-risk axis** (`planner/ada_risk.py`) — **ADA-risk = MHC-II epitope density × foreignness**, with a
   **self-tolerance filter** (JanusMatrix-style: self epitopes are tolerated; foreign drive ADA). It **recovers
   immunogenic-vs-tolerated**: the foreign writers (real UniProt SpCas9 / ISCro4 / Bxb1) score **above** the human

{pen_stack-6.9.0 → pen_stack-6.9.1}/pen_stack.egg-info/SOURCES.txt

@@ -34,6 +34,7 @@ configs/intent_weights.yaml
 configs/known_unknowns.yaml
 configs/llm.yaml
 configs/metric_guide.yaml
+configs/mhc_epitope_oracle.yaml
 configs/monitor_queries.yaml
 configs/score_axes.yaml
 configs/seroprevalence.yaml

{pen_stack-6.9.0 → pen_stack-6.9.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "pen-stack"
-version = "6.9.0"
+version = "6.9.1"
 description = "Open infrastructure for genome writing: the Writable Genome atlas, the Writer Atlas, and the Write Planner."
 readme = "README.md"
 requires-python = ">=3.11"