Changelog¶
0.0.4 (2026-04-15)¶
Current release.
Preserved parallel edges in the NetworkX
MultiDiGraphengine so distinct evidence edges with the same source, predicate, and target are no longer overwrittenCorrected KEGG fixed-width record parsing for multi-line
GENEandPATHWAYsections used by the seeding workflowTightened
metabolism_chain()traversal so carcinogen-specific chains do not absorb unrelated unlabeled branches through shared enzymesfilter_knowledge_graph()now returns detached model copies rather than aliasing the source graph objectsRestored a clean strict
mypypass for the shipped source tree
0.0.3 (2026-03-21)¶
Release metadata synchronized for the current PyPI/GitHub publication.
Updated package and app version identifiers to
0.0.3
0.0.2 (2026-03-19)¶
Current development release.
Added graph ingestion modes:
exploratoryandstrictAdded canonical grounding metadata, record origin tracking, and custom predicates
Added validated/exploratory graph visibility filtering for preview, data, export, and persistence
Added visibility-aware JSON, HTML, JS, and GEXF export helpers
Added revision visibility tracking and SQLite schema migration support
Added clean repository shutdown via context-manager support and fixed prior SQLite resource warnings
Added Ollama/OpenAI backend abstraction and mode-aware seeded graph preparation
0.0.1 (2026-03-17)¶
Initial release.
Pydantic v2 data models for 7 node types and 10 edge types
NetworkX MultiDiGraph engine with load/merge/validate
LLM-powered extraction via OpenAI structured outputs
Streamlit app with manual entry, LLM extraction, and gene panel loading
D3.js force-directed graph viewer with dark theme
Export to JSON,
graph-data.js(D3 viewer), and GEXF (Gephi)Curated Tier 1 (13 genes) and Tier 2 (23 genes) reference panels
Referenced activity-score tables for 18 genes, including evidence metadata
Test coverage across models, engine, exporter, storage, and reference data
CI/CD: ruff linting, pytest matrix (3.10–3.12), PyPI publish workflow