Technical Platform

MATSim-EpiSim Framework

The centrepiece of EPISERVE is the MATSim-EpiSim framework — a highly detailed Agent-Based Model (ABM) developed within the MODUS-Covid project.

Core features:

• Simulation of individual mobility and activities
• Detailed disease states
• Integration of pharmaceutical and non-pharmaceutical interventions
• Modular architecture for flexible extensions

Data Hub

Internal Data Lake infrastructure based on:

• lakeFS for data version control
• Ceph storage platform (ZIB) for long-term data storage
• Apache Airflow for workflow management
• S3-compatible interface for parallel uploads/downloads

Data sources:

• Infection rates (RKI)
• Genomic surveillance data
• Mobility patterns
• Weather data
• Social media data

Open Data Hub:

• InvenioRDM for FAIR data publication
• Persistent identifiers (DOIs)
• Open access for the research community

AI Integration

Machine learning for:

• Data validation and consistency checking
• Fast surrogate models (Graph Neural Networks)
• Automated calibration
• Sentiment analysis (social media)

Models:

• Random Forests
• Graph Neural Networks (GNNs)
• Bayesian Optimisation
• Large Language Models (LLMs) for text analysis

High-Performance Computing

ZIB infrastructure:

• Over 120,000 CPU cores
• More than 150 GPU units
• Member of the National High Performance Computing Network (NHR)
• Container virtualisation via RISE (Research Infrastructure Support Environment)

Simulation Architecture

Multi-scale approach:

• Agent-Based Models (ABM) for detailed regions
• ODE models for faster computations
• PDE models for continuous spatial analysis
• Hybrid couplings (ABM/ODE/PDE)

Graph Neural Networks:

• Spatially resolved predictions
• Near-instant simulation results
• Principal Component Analysis (PCA) for dimensionality reduction

Software Stack

Development:

• Version control: Git/GitLab
• CI/CD: GitLab Pipelines
• Testing: Automated test infrastructure
• Code quality: Regular audits

Visualisation:

• Apache Superset for interactive dashboards
• JupyterHub for data exploration
• Web-based front-ends for user interaction

Interfaces

API access:

• REST API for model execution
• Programmatic access to the Data Hub
• Integration with external systems

Compatibility:

• Designed for integration with TI BBD (DZIF)
• National and international data interoperability
• FAIR principles (Findable, Accessible, Interoperable, Reusable)

---

Open Source & Open Science

All developed models, frameworks, and data will be made available as open source and open access to ensure maximum transparency, reproducibility, and community-wide use.