CPP-CQA Relationship Modeler

Models CPP to CQA relationships for core MSAT process development and technology transfer. Quantifies relationship strength, direction, mechanism, and evidence level per ICH Q8(R2) Design Space methodology. Generates risk-weighted Tech Transfer Risk Scores to prioritise DoE studies and control strategy documentation before site transfer.

---

The Problem

In pharmaceutical MSAT (Manufacturing Science and Technology), understanding which Critical Process Parameters (CPPs) drive which Critical Quality Attributes (CQAs) is the foundation of:

• ICH Q8(R2) Design Space — multivariate CPP-CQA characterisation to define proven acceptable ranges
• Control Strategy (CTD 3.2.P.2.3 / 3.2.S.2.6) — per-unit-operation control requirements linked to CQA risk
• Technology Transfer — identifying which CPP-CQA links are under-characterised and pose receiving-site risk
• Process Development — prioritising DoE studies based on strength and risk of each CPP-CQA relationship

CPP-CQA Relationship Modeler maps all CPP → CQA linkages across programs, quantifies each relationship by strength (0–1), direction (positive/negative/optimal-range/complex), mechanism (cell physiology, protein chemistry, separation science, etc.), and evidence level (Level 1 DoE through Level 4 mechanistic model) — then computes a weighted Tech Transfer Risk Score to drive action.

---

What It Does

1. CPP-CQA Relationship Mapping — Per program, per unit operation:

Relationship Field	Values	Implication
Strength	0.0 – 1.0	Probability of CQA failure on CPP excursion
Direction	Positive / Negative / Optimal Range / Complex	Process behaviour shape
Mechanism	Cell Physiology / Protein Chemistry / Glycosylation / Separation Science / Chemical Reaction / Physical Processing	Mechanistic basis
Evidence Level	Level 1 (DoE) → Level 4 (Mechanistic)	Regulatory defensibility

2. Relationship Strength Classification — ICH Q8(R2)-aligned:

Strength	Category	Action
≥ 0.75	Strong	Mandatory Level 1 DoE; NOR/PAR must be established
0.55 – 0.74	Moderate	Level 2 OFAT studies; enhanced in-process control
< 0.55	Weak	Platform data may suffice; routine CPV monitoring

3. Tech Transfer Risk Score — Weighted composite (0–100%):

• High-risk links score 1.0 × strength, Medium-risk 0.5 × strength, Low-risk 0.0 × strength
• Normalised to total relationship strength across the program

Score	Verdict	Action
≥ 70%	High Risk	Halt transfer — upgrade evidence for all Strong links
40–69%	Medium Risk	Transfer with enhanced monitoring
< 40%	Low Risk	Standard CPV per ICH Q10

4. CQA Sensitivity Profiles — Per CQA across all unit operations:

• Number of CPPs affecting the attribute
• Count of strong links and high-risk links
• Average relationship strength → High / Medium / Low sensitivity verdict
• Control strategy recommendation per ICH Q8(R2)/Q10

5. CPP Reach Analysis — Per parameter:

• How many CQAs the CPP influences (reach breadth)
• Maximum relationship strength to any CQA
• Top risk level — drives prioritisation of NOR tightening at receiving site

6. Critical Links — Strong (≥0.75) + High-risk links listed with:

• Specific required action (Level 1 DoE, comparability runs, enhanced monitoring)
• Mechanistic rationale from DoE/OFAT study notes
• ICH Q8(R2) / Q10 / Q11 regulatory reference

7. REST API — Full JSON API for integration with LIMS, ELN, or QMS

---

Architecture

CPPCQAModeler/
│
├── cppcqa/
│   ├── app.py        # Flask factory — web UI + REST API (port 5091)
│   ├── modeler.py    # Relationship analysis engine — strength, risk, profiles
│   ├── model.py      # SQLite persistence (WAL mode, 5-table normalised schema)
│   ├── seed.py       # Auto-seeds on first launch
│   └── templates/
│       ├── base.html          # Dark theme, navigation, badge/alert CSS
│       ├── dashboard.html     # Program summary, risk KPIs, reference tables
│       ├── program.html       # CQA sensitivity + CPP reach + critical links
│       ├── relationships.html # Full per-unit-operation relationship tables
│       └── tech_transfer.html # Tech transfer risk score, CQA-level breakdown
│
├── data/
│   └── mock_data.py   # 3 programs × 12 ops × 37 CPPs × 18 CQAs × 56 relationships
│
├── tests/
│   └── test_cppcqa.py  # 166 tests across 5 classes
│
├── run.py              # Entrypoint (port 5091)
└── requirements.txt    # Flask + pytest

---

Program Corpus (Mock Data)

3 GxP manufacturing programs across all product types and phases:

Program	Type	Pathway	Phase	CPPs	CQAs	Links	Verdict
Emafilimab (Anti-PD-1 mAb)	Biologic	BLA	Clinical Phase 3	16	6	23	Medium Risk
Nexarelin HCl Tablets 25 mg	Small Molecule	NDA	Commercial	12	6	18	Medium Risk
GeneCure-X (Lentiviral Gene Therapy)	Cell/Gene Therapy	BLA	Clinical Phase 1	9	6	15	High Risk

Total: 12 unit operations · 37 CPPs/KPPs · 18 CQAs · 56 CPP-CQA relationships

Representative Relationships

Program	Unit Operation	CPP	CQA	Strength	Direction	Evidence	TT Risk
Emafilimab	Production Bioreactor	Bioreactor pH	Potency	0.85	Optimal Range	Level 1	High
Emafilimab	CEX Chromatography	Elution Salt Concentration	Purity	0.88	Optimal Range	Level 1	High
Nexarelin	API Synthesis	Reagent Equivalents	Assay	0.88	Positive	Level 1	High
Nexarelin	Tablet Compression	Compaction Force	Dissolution	0.80	Negative	Level 1	Medium
GeneCure-X	Vector Production	MOI	Transduction Efficiency	0.90	Positive	Level 2	High
GeneCure-X	Downstream Purification	Gradient Salt Concentration	Empty Capsid Ratio	0.85	Negative	Level 2	High

---

Quickstart

git clone https://github.com/timjm25/CPPCQAModeler.git
cd CPPCQAModeler

pip install -r requirements.txt

python run.py
# → http://127.0.0.1:5091

Auto-seeds 3 MSAT programs, 12 unit operations, 37 CPPs, 18 CQAs, and 56 relationships on first launch. No configuration required.

---

Pages

Route	Description
/	Dashboard — KPI tiles, programs table with risk scores and verdicts, strength/evidence/risk reference tables
/program/<id>	Program detail — CQA sensitivity table, CPP reach table, critical links with mechanistic notes
/program/<id>/relationships	Full CPP-CQA relationship tables grouped by unit operation — strength bar, direction, mechanism, evidence badges
/program/<id>/tech-transfer	Tech transfer risk assessment — risk score card, CQA-level breakdown, critical links with required actions

---

REST API

Method	Endpoint	Description
GET	/api/v1/programs	All programs with risk scores, verdicts, CPP/CQA/link counts
GET	/api/v1/program/<id>	Program summary — risk score, verdict, all counts
GET	/api/v1/program/<id>/relationships	All CPP-CQA relationships with strength, direction, mechanism, evidence, risk
GET	/api/v1/program/<id>/cqas	CQA list with specifications and types
GET	/api/v1/program/<id>/cpps	CPP/KPP list with NOR bounds and unit operations
GET	/api/v1/stats	Summary statistics
GET	/healthz	Health check

Examples

# All MSAT programs with tech transfer risk scores
curl http://127.0.0.1:5091/api/v1/programs

# GeneCure-X CPP-CQA relationship model (High Risk CGT)
curl http://127.0.0.1:5091/api/v1/program/PRG-003/relationships

# Emafilimab CQA list with specifications
curl http://127.0.0.1:5091/api/v1/program/PRG-001/cqas

Relationships Response (excerpt)

{
  "relationships": [
    {
      "rel_id": "REL-003-001",
      "cpp_name": "Multiplicity of Infection",
      "cqa_name": "Transduction Efficiency",
      "op_name": "Vector Production",
      "strength": 0.90,
      "strength_category": "Strong",
      "direction": "positive",
      "mechanism": "cell_physiology",
      "evidence_level": "Level 2",
      "tech_transfer_risk": "High"
    }
  ]
}

---

Testing

python3 -m pytest tests/ -v
# 166 tests — 5 classes — passes in < 2s

Class	Tests	Scope
TestMockData	27	3 programs, 12 ops, 37 CPPs, 18 CQAs, 56 relationships; required fields; referential integrity; valid enumerations; per-program counts
TestModeler	42	strength_category, direction_label, evidence_label, risk_score_from_relationships, program_risk_verdict, cqa_profile, cpp_profile, critical_links, analyze_program — all boundary conditions
TestModel	40	Seed, idempotent re-seed, all get_* methods, analyze_program, verdict checks, dashboard_summary, stats
TestFlaskRoutes	27	All 4 pages × 3 programs, 404 handling, alert banners, CQA/CPP content, risk verdict rendering
TestRestAPI	30	All 7 endpoints, response structure, field completeness, per-program counts, strength types, 404 handling

---

License

MIT License — see LICENSE.

Copyright (c) 2026 Tim Maguire.