ExplorerChain

ExplorerChain enables decentralized, privacy-preserving predictive modeling in healthcare and genomics by exchanging model parameters via a permissioned blockchain to build more generalizable models without sharing raw patient data.

Key Features:

• Blockchain Infrastructure: Operates on a permissioned blockchain network that restricts participation to authorized entities.
• Privacy-Preserving Online Learning: Implements Expectation Propagation Logistic Regression (EPLR) for decentralized training without sharing raw patient data.
• Consensus Algorithm — PoINT: Uses the Proof-of-Information-Timed (PoINT) consensus algorithm to validate transactions and model updates within the network.
• Online Machine Learning: Facilitates continuous learning from distributed institutional data sources by exchanging and updating model parameters.
• Metadata of Transactions: Records transaction metadata to provide transparency and traceability of operations and model updates.
• Core Algorithms: Includes core, new network, and new site/data algorithms for integrating and updating participating sites and datasets.

Scientific Applications:

• Myocardial infarction prediction: Enables cross-institutional model training to predict myocardial infarction outcomes without sharing patient-level data.
• Cancer biomarker identification: Supports distributed analyses for identifying cancer biomarkers across institutional cohorts.
• Predicting length of hospitalization post-surgery: Enables collaborative prediction of postoperative length of stay using decentralized models.

Methodology:

Computational elements explicitly stated include Expectation Propagation Logistic Regression (EPLR) for decentralized online learning, the Proof-of-Information-Timed (PoINT) consensus algorithm for validating transactions and model updates, a semi-trust assumption for participant behavior, data format normalization across institutions, consideration of non-determinism in distributed systems, and exchange of model parameters rather than raw data.