Blockchain-Enabled Trust Infrastructure for Verifiable Biomolecular Computational Workflows = 검증 가능한 생체분자 계산 워크플로우를 위한 블록체인 기반 신뢰 인프라

다국어 초록 (Multilingual Abstract)

This thesis investigates how biomolecular computational workflows (BCWs) can be engineered as verifiable systems that integrate tamper-evident provenance, deterministic output verification, and policy-enforced validity for multi-party exchange.Four complementary implementations are designed, deployed on permissioned ledger infrastructure (Purechain for RQ1, RQ3, and RQ4; Hyperledger Fabric for RQ2) under a shared verification pattern, and empirically evaluated across representative biomedical pipelines addressing four research questions. For RQ1 (Evidence and Provenance), a blockchain-enabled microservices framework integrating a permissioned ledger with decentralised storage and role-based access control achieves 485.96 req/s API throughput, 445 TPS on-chain performance, and zero false positives in SHA-256 tampering detection, with end-to-end provenance validated through an HIV-1 protease case study. For RQ2 (Verifiable Computation), a tri-layered
blockchain-integrated verification system extends provenance from tamper-evident history to verification-by-digest, achieving 100% reproducibility across 5 protein targets and 98.2% across 20 targets in six structural categories. For RQ3 (Determinism in Practice), a deterministic blockchain-audited virtual screening pipeline processing 71,853 compounds across 10 targets demonstrates strong predictive performance
(mean R2 = 0.693, AUC-ROC = 0.937) while passing all component-level determinism tests with 100% hash verification across 40 re-executions. For RQ4 (Policy- Enforced Validity), a policy-enforced blockchain-credential architecture achieves F1 = 1.00 on quality-control and status anomalies with verification latency of p50 ∼ 32 ms independent of registry size. Taken together, these four complementary implementations empirically support the three-dimensional validity conjunction Valid(w) = AuditValid(w) ∧ DeterminismValid(w) ∧ PolicyValid(w), providing a composable trust-stack for reproducible, verifiable, and governance-ready biomolecular computational workflows in which each validity dimension is independently demonstrated and their conjunction characterised architecturally rather than through a single integrated execution.

목차 (Table of Contents)

• Chapter 1: Introduction 1
• 1.1 Research Motivation 5
• 1.2 Thesis Contributions 8
• Chapter 2: Background Information 12
• 2.1 Biomolecular Computational Workflows (BCWs) 12
• 2.1.1 BCWs Typical Stages: 14
• 2.1.2 Workflow Artifacts and Outputs: 15
• 2.2 Reproducibility and Determinism in Scientific Pipelines 17
• 2.2.1 Repeatability, Reproducibility, and Replicability 17
• 2.2.2 Sources of Non-Determinism in Computational Pipelines 19
• 2.2.3 Deterministic Execution Controls and Environment Pinning 20
• 2.3 Provenance and Workflow Evidence Models 21
• 2.3.1 Data Provenance and Lineage Concepts 21
• 2.3.2 Provenance Models for Scientific Workflows 23
• 2.4 Blockchain-Based Trust Infrastructure for BCWs 25
• 2.4.1 Trust Gaps in Scientific Workflows 25
• 2.4.2 Blockchain Properties Relevant to Scientific Workflows 26
• 2.4.3 Verifiable Computation and Tamper-Evident Audit Logs 27
• 2.4.4 Practical Constraints: Why Scientific Artifacts Cannot Live On-Chain 28
• 2.4.5 Off-Chain Storage with Cryptographic Integrity Anchoring 28
• 2.5 Policy, Credentials, and Access Control in Distributed Workflows 30
• 2.5.1 Credential Models and Authorization Mechanisms 30
• 2.5.2 Policy Enforcement for Cross-Organizational Exchange 32
• Chapter 3: Related Works 34
• 3.1 Provenance and Reproducibility in BCWs 34
• 3.2 BC-backed Provenance: Auditability, Hybrid Storage, and Integrity Anchoring 37
• 3.2.1 Tamper-evident provenance events and non-repudiation 37
• 3.2.2 Hybrid storage and verifiable provenance at scale 38
• 3.2.3 Permissioned provenance registries and workflow-aware certification 39
• 3.2.4 Off-Chain Artifact Storage and Integrity Anchoring 39
• 3.2.5 Limitation and transition: auditability does not imply policy-validity 41
• 3.3 Credentials and Authorization Mechanisms 41
• 3.3.1 Credential Models: DIDs, Verifiable Credentials, and Presentations 42
• 3.3.2 Authorization Models: RBAC, ABAC, and Cross-Domain Access Control 43
• 3.3.3 Operational Policy Enforcement in Multi-Domain Systems 44
• 3.3.4 Limitation and Transition: From Credential Validity to Policy-Enforced Validity 44
• 3.4 Summary of Related Works and Research Gaps 45
• Chapter 4: Proposed Methodology 48
• 4.1 Overview 48
• 4.2 System Model and Trust Assumptions 50
• 4.2.1 Entities and Roles 50
• 4.2.2 Data Objects and Cryptographic Commitments 51
• 4.2.3 Verification Semantics 53
• 4.2.4 Trust Assumptions and Boundaries 54
• 4.3 Proposed System Architecture 55
• 4.3.1 Three-plane decomposition and trust boundaries 55
• 4.3.2 Canonical artifacts, commitments, and anchoring 56
• 4.3.3 Determinism controls as an architectural primitive 57
• 4.3.4 Policy-enforced exchange as a first-class workflow interface 58
• 4.3.5 End-to-end lifecycle mapping (execution anchoring exchange verification) 58
• 4.3.6 Architectural safety under failures 59
• 4.4 Data Model and Verification Artifacts 59
• 4.4.1 Notation and canonicalization 60
• 4.4.2 Artifact Manifest (canonical JSON) 60
• 4.4.3 Step-chain provenance (tamper-evident step linking) 61
• 4.4.4 Execution-context fingerprint 62
• 4.4.5 Integrity anchoring: hashes, Merkle roots, and proofs 62
• 4.4.6 Verification receipts for policy-enforced exchange 63
• 4.4.7 Validity as a verifiable predicate over artifacts 64
• 4.5 Protocols 65
• 4.5.1 Execute-and-Anchor 65
• 4.5.2 Verify-and-Reproduce 66
• 4.5.3 Policy-Enforced Exchange 67
• 4.5.4 Protocol composition: end-to-end validity 70
• Chapter 5: Performance Evaluation and Result Discussion 71
• 5.1 Overview and Evaluation Strategy 71
• 5.2 Experimental Setup 73
• 5.2.1 Blockchain and Storage Infrastructure 73
• 5.2.2 Datasets and Protein Targets 74
• 5.2.3 Software and Execution Environments 76
• 5.2.4 Evaluation Methodology 76
• 5.3 RQ1 Evidence and Provenance 77
• 5.3.1 API Throughput and Latency 77
• 5.3.2 Blockchain Transaction Performance 79
• 5.3.3 Energy Overhead 80
• 5.3.4 Security Validation 81
• 5.3.5 End-to-End Case Study: HIV-1 Protease 81
• 5.3.6 Discussion: Addressing RQ1 82
• 5.4 RQ2 Verifiable Computation 83
• 5.4.1 Protein Structure Preparation 83
• 5.4.2 Binding-Site Analysis 84
• 5.4.3 Performance Overhead 85
• 5.4.4 Reproducibility and Tamper Detection 86
• 5.4.5 Scalability: 20-Protein Benchmark 87
• 5.4.6 Discussion: Addressing RQ2 88
• 5.5 RQ3 Determinism in Practice 89
• 5.5.1 Consensus AI Performance 90
• 5.5.2 Per-Target Hybrid AIDocking Optimisation 91
• 5.5.3 Enrichment Evaluation 92
• 5.5.4 Scaffold Diversity 92
• 5.5.5 DUD-E External Validation 93
• 5.5.6 Determinism Verification 94
• 5.5.7 Blockchain Provenance Overhead 95
• 5.5.8 Discussion: Addressing RQ3 95
• 5.6 RQ4 Policy-Enforced Validity 97
• 5.6.1 Core Operation Latency 97
• 5.6.2 Validity Enforcement Accuracy 98
• 5.6.3 Ablation Study 100
• 5.6.4 Throughput and Concurrency 100
• 5.6.5 On-Chain Storage Growth 101
• 5.6.6 Discussion: Addressing RQ4 102
• 5.7 Cross-System Synthesis and End-to-End Validity 103
• 5.7.1 Validity Dimension Coverage 103
• 5.7.2 Shared Infrastructure and Composability 104
• 5.7.3 Overhead and Feasibility 104
• 5.7.4 Limitations and Threats to Validity 105
• Chapter 6: Conclusion, Future Works and Open Research Issues 107
• 6.1 Conclusion 107
• 6.2 Future Works 109
• 6.3 Open Research Issues 110
• 6.4 Reproducibility Package and Artifact Availability 111
• 6.4.1 Code Repositories 111
• 6.4.2 Pinned Execution Environments 112
• 6.4.3 Datasets and Dataset Splits 112
• 6.4.4 Smart Contracts and Trained Model Artifacts 112
• 6.4.5 On-Chain Transaction Records and Off-Chain Object References 113
• 6.4.6 Independent Verification Procedure 114
• References 115