kraken

Build system for Phenome Health's unified general-purpose knowledge graph, KRAKEN (Knowledge Research & Analysis Kit for Evidence Networks).

Overview

KRAKEN integrates biomedical knowledge from many sources into a single Biolink Model-compliant knowledge graph, tailored for multiomic and wellness research. Release 2.3.0 spans roughly 26.5M nodes and 120.7M edges drawn from 121 primary knowledge sources. The build system harmonizes each source to a common schema, resolves entities across sources using provided equivalency mappings, and exports flat files ready for downstream tools.

Knowledge sources

KRAKEN integrates:

• Aggregator knowledge graphs: RTX-KG2, ROBOKOP, Translator KG Open
• Domain-specific knowledge graphs: Multiomics KG, Microbiome KG, Clinical Trials KG, Drug Approvals KG
• Ontologies & vocabularies: UMLS, LOINC, LIPID MAPS, RefMet, NCBI Gene
• Identifier equivalences: Babel, the build behind the Translator Node Normalizer
• Specialized resources: the NIH CDE Repository, PGS Catalog, and published biological-age (Earls et al. 2019), biological-BMI (Watanabe et al. 2023), and long COVID (Institute for Systems Biology) datasets

Each aggregator contributes many upstream primary knowledge sources; see artifacts/metagraphs/ for the full per-source breakdown.

Usage

Configure sources and build steps in config/build_config.yaml, then run:

uv run python build.py

To build KRAKEN yourself, including getting the input data and the hardware it needs, see CONTRIBUTING.md.

Build process

1. Harmonize - convert each source to a common Biolink format/schema
2. Integrate - merge sources, performing entity resolution using each source's equivalency mappings
3. Post-process - generate a coherent, small test version of the graph

Validation and metagraph creation are done both on each harmonized artifact and on the final integrated graph.

Output

• Unified KG: artifacts/integrated/kraken_*.jsonl
• Metagraphs: artifacts/metagraphs/

Requirements

Python 3.10+ with dependencies managed by uv:

uv sync

Citation

KRAKEN is described in a forthcoming manuscript; please cite that publication. Each release is also archived on Zenodo - concept DOI 10.5281/zenodo.21940866, which always resolves to the latest version.

License

Released under the MIT License.

Acknowledgments

Development of this codebase was assisted by Claude Code (Anthropic).