A federated analytics platform for clinical research networks that need to collaborate on sensitive health data without centralizing patient-level records. More information is available on the MIP Website.
- About MIP
- 9.2 Release
- What MIP Includes
- Deployment
- Federated Analysis Algorithms
- Jupyter Notebooks and Support Agent
- Data Management
- Architecture
- Onboarding
The Medical Informatics Platform enables privacy-preserving analysis across distributed clinical datasets. Hospitals and data providers keep patient-level data within their local governance boundaries, while researchers can run federated statistical and machine-learning analyses across participating sites.
This repository collects the technical, deployment, data-management, and architecture documentation needed to understand, deploy, and operate MIP.
The redesigned Data Handling step makes advanced preparation part of the Experiment Studio workflow. Researchers can configure cohort filters, missing-value handling, outlier treatment, and transformations, then review summaries and charts of the prepared data before choosing an analysis.
Researchers can create new categorical columns from filter-based rules, defining categories from conditions on existing variables. K-means is also available as a preprocessing option: a fitted clustering result can be reused to create a categorical cluster column for downstream analyses. These options extend the existing missing-value, outlier, and longitudinal transformations.
The Experiment Studio guides researchers through data exploration, preparation, algorithm configuration, execution, and results. Algorithm visualizations and result tables have been redesigned, with PDF reports and CSV exports. The updated dashboard supports experiment folders, search, and side-by-side result comparison.
Jupyter notebooks are a major new feature in MIP. Researchers can open a
JupyterLab workspace from the portal and build reproducible analyses in Python.
The pre-installed mip client provides data discovery, filters, preprocessing,
and algorithm execution through the platform backend. A welcome notebook,
example analyses, and client documentation help researchers get started while
patient-level records remain at the contributing sites.
The new mip-jupyter repository provides the notebook workspace, Python client, and Jupyter images. Example notebooks live in the mip-examples repository.
The notebook workspace includes Cohort Scout, a support agent that helps researchers create and edit notebooks, explain analysis code and results, and navigate the MIP Python client. It is available through Jupyter AI when the assistant service is configured for the deployment.
Exaflow adds histogram-based quartile estimates, a binned Mann-Whitney U test, Standardized Mean Difference, and richer K-means reporting and preprocessing. The backend introduces analysis and specification endpoints and experiment folder management to support the revised workflow. The component releases are listed in MIP building blocks.
MIP combines a web interface, backend services, federated analysis engine, Jupyter notebooks, deployment tooling, and supporting data-management tools. The main MIP building blocks are listed with the repositories that host them.
MIP supports both local development deployments and Kubernetes-based deployments for production-like or federated installations.
The algorithm documentation describes the available federated analyses and links to the underlying Exaflow analytic engine documentation.
Use the notebook workspace to discover available data, compose preprocessing pipelines, run federated algorithms, and document results alongside Python code. Cohort Scout can assist with notebook creation and explanation.
Data providers prepare and validate source datasets and metadata before upload. Researchers configure filters and transformations per analysis in the Experiment Studio or notebook pipelines; these operations do not rewrite uploaded files.
A detailed user guide for the Data Quality Control tool can be found here:
Data Catalog is a component of MIP for EBRAINS. It enables management, visualization, and access to data models and medical conditions.
- High-level view of the architecture, the main building blocks and data flows.
- Onboarding to the Medical Informatics Platform MIP on EBRAINS Collaboratory
This project/research received funding from the European Union’s Horizon 2020 Framework Programme for Research and Innovation under the Framework Partnership Agreement No. 650003 (HBP FPA).