Architecture Overview
The ColabBio Base Platform is built around a modern, cloud-native microservices architecture designed to run on top of Kubernetes. It provides the essential building blocks for secure clinical data processing, open-source integration, and federated learning.
Design Philosophy
Section titled “Design Philosophy”ColabBio enforces a strict architectural boundary between Generic Platform Components and Tenant-Specific Projects:
- Generic Platform Components (
cb-*): Foundational microservices, UIs, and devops tools that apply across the entire ColabBio ecosystem (e.g.,cb-web,cb-vfs). These are completely isolated from tenant data. - Tenant Projects: Specific institutional deployments or use-cases (e.g.,
WSI_MIL). These contain their own isolated data pipelines, models, and tenant-specific configurations.
Open Source Service Stack
Section titled “Open Source Service Stack”ColabBio integrates best-in-class open-source projects into a cohesive medical intelligence platform.
1. FHIR Layer
Section titled “1. FHIR Layer”- HAPI FHIR Server: A complete FHIR R4 server with an integrated web UI.
- HAPI PostgreSQL: Persistent backend database for the FHIR server.
2. OMOP CDM Layer (OHDSI Stack)
Section titled “2. OMOP CDM Layer (OHDSI Stack)”- OMOP PostgreSQL: The standard CDM v5.4 schema and database.
- OHDSI WebAPI: RESTful API over the OMOP database for cohorts and analysis.
- OHDSI Atlas: Web UI for cohort exploration and design.
- ACHILLES & DataQualityDashboard: Batch jobs for data characterization and ensuring thousands of automated Data Quality checks.
3. Data Loading & ETL
Section titled “3. Data Loading & ETL”- Synthea & ETL-Synthea: Generators and loaders for synthetic patient data (used heavily in dev/testing).
- WhiteRabbit & USAGI: Production tools for analyzing real sources and mapping local clinical vocabularies to standard OMOP concepts.
- Vocab Loader: Automated system to load OMOP vocabularies (SNOMED, LOINC, etc.).
4. Future Expansion Layers
Section titled “4. Future Expansion Layers”- Genomics (v3): Incorporating Hail (PySpark) on GCS/S3, CanDIG for API access, and HTSlib for VCF/CRAM access.
- DICOM / WSI (v4): Integration with Orthanc (PACS), OpenSlide, and PixelMed.
Network & Security Architecture
Section titled “Network & Security Architecture”The platform uses environment-based namespaces (colabbio-dev, colabbio-stage, colabbio-prod) to guarantee strict isolation.
All traffic flows through an Ingress Controller that performs TLS termination and routes requests to internal services (like HAPI or OHDSI). To prevent CORS and CSRF issues—especially when embedding viewers like OMERO inside our Flutter apps—ColabBio prioritizes path-based routing under a single Ingress domain.