zk_crs

Federated Theta Agent (FedML Clinical Simulation)

This project demonstrates how to build a federated learning agent for decentralized clinical research using FedML. It simulates a small synthetic clinical dataset, distributes it across clients, and runs federated training using FedML’s client/server orchestration. The design mimics how this would run on a decentralized infrastructure like Theta Cloud + Edge.

✨ Key Features

Synthetic Clinical Dataset
- Features: age, systolic/diastolic blood pressure, glucose, comorbidity, lab marker.
- Binary classification task (risk prediction).
Non-IID Client Data Split
- Data is distributed across clients to simulate real-world hospital settings.
Federated Training via FedML
- Uses FedML’s FedMLRunner to coordinate server + clients.
- Clients train locally with a simple MLP classifier.
- Server aggregates updates using FedAvg.
Local Test Launcher
- Spawns one server + multiple client processes on the same machine for quick testing.

DEZI NETWORK

Decentralised Clinical Research Framework

Key Components:

Blockchain Infrastructure: Solana Blockchain with advanced advanced privacy features which use zero-knowledge proofs (ZKPs) and homomorphic encryption to shield sensitive data (patient information) while maintaining regulatory compliance and auditability.
Federated Learning Integration: Clinical research frameworks can leverage federated learning, allowing institutions to collaboratively train AI models without sharing raw patient data. Only encrypted model updates are posted to the blockchain, preserving privacy and adhering to regulations like GDPR.
Zero-Knowledge Proofs (ZKPs): zk-SNARKs enable verification of computations or model updates on-chain without revealing underlying sensitive data, ensuring both privacy and data integrity.
Decentralized Access Control: Solana’s privacy tools and protocols like Arcium and Lit enable fine-grained, programmable access controls and encrypted computation, so only authorized parties can access or process clinical data.

Summary Table

Feature	Solana Implementation	Benefit for Clinical Research
Confidential Balances	ZKPs + Homomorphic Encryption	Private, compliant data sharing
Federated Learning	Decentralized model training, blockchain logs	No raw data leaves institution
zk-SNARKs	On-chain proof of computation	Verifiable, privacy-preserving
Decentralized Access Control	Protocols like Lit, Arcium	Secure, granular data permissions

This combination allows for privacy-preserving, decentralized clinical research on Solana, enabling secure collaboration between healthcare institutions without compromising patient confidentiality or regulatory compliance

🎯 Benefits

Benefit	Explanation
Privacy-preserving	Raw data never exposed; only proofs are shared
Verifiable science	Prevents p-hacking or data tampering
Scalable AI	Agents automate workflows while ensuring compliance
Decentralized trust	Blockchain ensures that no single party controls the process
Regulatory compliance	Designed with GDPR, HIPAA in mind (e.g., consent tracking, pseudonymity)

Existing Frameworks

There are privacy-preserving decentralized clinical research frameworks, primarily leveraging blockchain technology. Most existing solutions use private or consortium blockchains (like Hyperledger Fabric or Ethereum) to manage clinical trial data, automate consent, and control access via smart contracts[1][5][6]. These frameworks focus on data security, transparency, and auditability, often storing sensitive data off-chain or encrypted, and using decentralized storage systems like IPFS[1][6]. While many are still proof-of-concept or pilot projects, they demonstrate the feasibility of privacy-preserving, decentralized clinical research management[1][6][7].

Citations: [1] https://www.medrxiv.org/content/10.1101/2024.06.12.24308813v1.full-text [2] https://www.sciencedirect.com/science/article/pii/S1551714424002672 [3] https://pmc.ncbi.nlm.nih.gov/articles/PMC7153067/ [4] https://www.tandfonline.com/doi/full/10.1080/17576180.2025.2452774?src=exp-la [5] https://ijtech.eng.ui.ac.id/article/view/6703 [6] https://pmc.ncbi.nlm.nih.gov/articles/PMC8075489/ [7] https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/cmu2.12488 [8] https://rymedi.com/transforming-clinical-trials-with-blockchain-technology-innovations-and-challenges-in-2024/

alt text

+———————————————————————————–+ | Privacy-Preserving Decentralized Clinical Research | | (Solana-based) | +———————————————————————————–+ | | | +——————-+ +——————-+ +—————-+ | | | Hospital/Clinic | | Research Center | | Data Partner | | | +——————-+ +——————-+ +—————-+ | | | | | | | | (Local Data Storage) | (Local Data Storage) | | | +———–+—————+————–+—————+ | | | | | | v v | | +———————————————————+ | | | Federated Learning Clients (Local Model Training) | | | +———————————————————+ | | | | | | v v | | +———————————————————+ | | | Model Updates (Encrypted, ZK-Proof Attached) | | | +———————————————————+ | | | | | | v v | | +—————————————————————————+ | | | Solana Blockchain (Smart Contracts Layer) | | | | - Stores encrypted model updates | | | | - Verifies ZK-proofs for privacy/compliance | | | | - Implements access control (Lit Protocol, Arcium, etc.) | | | +—————————————————————————+ | | | | | | v v | | +—————————————————————————+ | | | Authorized Researchers/Regulators (View/Access Encrypted Data) | | | +—————————————————————————+ | | | +———————————————————————————–+

+—————————————————————————+ [ Hospital/Clinic ] [ Research Center ] [ Data Partner ] | | | | (Local Data Storage) | (Local Data Storage) | +———–+————-+————-+————+ | | v v [ Federated Learning Clients (Local Model Training) ] | | v v [ Model Updates (Encrypted, ZK-Proof Attached) ] | v [ Solana Blockchain (Smart Contracts, Access Control, ZK Verification) ] | v [ Authorized Researchers/Regulators (Access Encrypted/Aggregated Results) ] +—————————————————————————+

Smart contract methods

Patient Smart Contract

addPatient: Registers a patient for the study and sets access permissions

editPermissions: Changes a patient’s access permissions

getPeople: Gets the set of patients registered for the study

Researcher Smart Contract

addQuery: Submits a new query to the blockchain

addQueryResult: Places a hash of a query result on the blockchain

getQueries: Retrieves waiting queries from the blockchain

getUnsolvedCount: Retrieves the number of waiting queries from the blockchain

Data Set sources: https://ncri1.partners.org/ProACT/

Key repos which are good start for development of agents and blockchain:

https://github.com/LeoYML/clinical-agent

https://github.com/FederatedAI/FATE-LLM

License

This project is licensed under the MIT License - see the LICENSE file for details.

https://github.com/ictashik/BlockChain_ClinicalTrial

This site is open source. Improve this page.