gem-mapping-studio 0.2.2__tar.gz → 0.2.4__tar.gz

gem_mapping_studio-0.2.4/PKG-INFO

@@ -0,0 +1,291 @@
+Metadata-Version: 2.4
+Name: gem-mapping-studio
+Version: 0.2.4
+Summary: Forome Genetic Evidence Model: a reference data model, UMLS/OMOP crosswalk tooling, and SHACL validation for basic-science genetic evidence
+Author-email: Michael Bouzinier <michael.bouzinier@forome.org>
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://w3id.org/genetic-evidence-model/
+Project-URL: Repository, https://github.com/ForomePlatform/genetic-evidence-model
+Keywords: genetic evidence,semantic model,UMLS,crosswalk,SHACL,biomedical informatics,variant interpretation,curation
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE-code.txt
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: ruamel.yaml>=0.18
+Requires-Dist: rdflib>=7.0
+Requires-Dist: pyshacl>=0.27
+Requires-Dist: requests>=2.28
+Requires-Dist: PyMuPDF>=1.23
+Requires-Dist: pypdf>=4.0
+Provides-Extra: dev
+Requires-Dist: pytest>=7; extra == "dev"
+Provides-Extra: local
+Requires-Dist: psycopg[binary]>=3.1; extra == "local"
+Dynamic: license-file
+
+# A Semantic Model of Genetic Evidence
+
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.22260686.svg)](https://doi.org/10.5281/zenodo.22260686)
+[![PyPI](https://img.shields.io/pypi/v/gem-mapping-studio.svg)](https://pypi.org/project/gem-mapping-studio/)
+
+A conceptual framework for representing scientific and genetic evidence from
+the biomedical literature in a form suitable for variant interpretation,
+automated reasoning, and AI-ready clinical infrastructure.
+
+This repository accompanies a manuscript in preparation: an extended
+version to be deposited on arXiv and a condensed version intended for
+journal submission.
+
+## What this repository contains
+
+```
+.
+├── paper/          LaTeX source for the manuscript (main.tex, references.bib)
+├── schema/         SHACL shapes + supporting definitions
+├── annotations/    One YAML file per annotated publication + raw PDF extractions
+├── case-reports/   Per-paper case reports (one for each of the six annotations)
+├── protocols/      Annotation protocol documents (canonical rules + per-mode workflows)
+├── skills/         Two Claude skills (annotation, review) that operationalise the protocols
+├── data/umls/      UMLS crosswalk of the dimensional vocabulary + decision log (DECISIONS.md)
+├── src/python/     The `gem-mapping-studio` package: Mapping Studio, crosswalk harness,
+│                   validators, PDF highlight/callout extraction (`forome.gem.*`)
+├── figures/        Source files for figures used in the paper
+├── docs/           Usage guides (STUDIO.md: the Mapping Studio)
+├── .github/        CI configuration validating annotations against the schema
+├── CHANGELOG.md    What changed in each tagged release
+└── KNOWN_LIMITATIONS.md  What the model, schema, corpus, and tooling do not yet do
+```
+
+## The annotation corpus
+
+Six publications, chosen to span the major epistemic shapes of genetic
+evidence encountered in literature-based variant interpretation:
+
+| Annotation                      | Role in the paper                    | Source                       |
+| ------------------------------- | ------------------------------------ | ---------------------------- |
+| `jossin2017.yaml`    (Llgl1)    | molecular mechanism                  | manual (ground truth)        |
+| `davis2011.yaml`     (TTC21B)   | breadth exemplar                     | manual (ground truth)        |
+| `nelson1992.yaml`    (CD18)     | classical molecular genetics         | manual (ground truth)        |
+| `gupta2015.yaml`     (ATP6AP2)  | low-credibility edge case            | manual (ground truth)        |
+| `v0/duerr2006.yaml`  (IL23R)    | clean GWAS exemplar                  | AI-drafted, expert-reviewed  |
+| `v1/inouye2018.yaml` (metaGRS)  | polygenic-score model-extension case | AI-drafted, expert-reviewed  |
+
+The four manual annotations live at the top of `annotations/`. The two
+AI-drafted annotations are versioned: `annotations/v0/` holds the
+original protocol-v0 drafts (curator-reviewed), and `annotations/v1/`
+holds their re-annotation under the current protocol. The paper
+documents **Duerr `v0`** and **Inouye `v1`** (the protocol matured on
+the Duerr review and was then applied to Inouye); the other version of
+each is retained for comparison.
+
+For manually annotated papers, PDF highlights and sticky-note callouts are the
+authoritative ground truth. The YAML is a structured transformation of those
+artifacts, with every assertion carrying a `source_span` pointing back to the
+specific page and quoted passage. Disagreements between the annotator and the
+AI reviewer are not silently resolved: they are captured as `reviewer_query`,
+`reviewer_suggestion`, or `reviewer_disagreement` fields so they remain
+auditable.
+
+For AI-drafted annotations, the same `source_span` anchoring is used, and
+curator review is the evaluation signal.
+
+Per-paper case reports are in `case-reports/`. Each report summarizes the
+paper's role in the corpus, the decomposition into `GeneticEvidence` items,
+the candidate extensions surfaced, the reviewer flags, and notes for
+downstream consumers.
+
+## Annotation protocols
+
+The `protocols/` directory documents how annotations are produced, separately
+from what they describe (the schema) and what they contain (the corpus). The
+intent is that an annotation under this schema is reproducible and citable
+under a versioned protocol, not an artifact of a particular annotator's
+unwritten conventions.
+
+- `protocols/PROTOCOL.md`: the canonical, mode-agnostic rules. Covers
+  decomposition principles (the lumper default), dimension assignment,
+  source-anchoring requirements, flag taxonomy, candidate-extension
+  promotion, normalization handling. Current version: 1.0.
+- `protocols/PROTOCOL_AUTONOMOUS.md`: the operational workflow for autonomous
+  AI annotation (single pass, no curator in the loop). Specifies input
+  quality gating, self-consistency checks, mandatory confidence-summary
+  emission, and failure handling.
+- `protocols/REVIEW_PROTOCOL.md` and
+  `protocols/REVIEW_PROTOCOL_INTERACTIVE.md`: the assertion-by-assertion
+  review protocol and its interactive, curator-in-the-loop variant, used
+  to adjudicate the AI-drafted annotations. Current version: 1.0.
+- `protocols/LABELING_EXAMPLES.md`: worked cases for the recurring
+  judgment calls referenced by the protocols above.
+
+A staged interactive *annotation* protocol (curator review of the
+decomposition before dimension filling) is planned but not yet
+specified.
+
+The two AI-drafted annotations in the corpus (Duerr 2006, Inouye 2018) were
+produced under what became `PROTOCOL_AUTONOMOUS.md` v1.0; their `provenance`
+blocks record the protocol version retroactively.
+
+## Skills
+
+Two Claude skills in `skills/` operationalise the protocols:
+
+- **`genetic-evidence-annotation/`** — autonomous drafting of a YAML annotation
+  for a paper (the autonomous protocol). Triggers on requests like "annotate
+  this paper under the genetic-evidence model" or "produce a GEM YAML annotation
+  for paper X".
+- **`genetic-evidence-review/`** — interactive, curator-in-the-loop review of an
+  existing annotation (the review protocol). Triggers on "review this GEM
+  annotation", "audit this annotation against the paper", and similar; it emits
+  a review log, a review report, and the updated annotation.
+
+Both follow the Agent Skills open standard and are portable across agents that
+support it (Claude Code, Cursor, Copilot, and others), not Claude-specific.
+Each skill references shared material **outside** its own folder, the protocols
+in `protocols/`, the schema in `schema/`, and exemplar annotations in
+`annotations/`, so it has to be installed together with that material.
+
+### Installing in Claude Code (or another in-repo agent)
+
+Run the agent inside a checkout of this repository and copy or symlink the
+skill folder into your skills directory (project-local `.claude/skills/` or
+user-global `~/.claude/skills/`), for example:
+
+```bash
+ln -s "$PWD/skills/genetic-evidence-annotation" .claude/skills/
+ln -s "$PWD/skills/genetic-evidence-review"     .claude/skills/
+```
+
+The skills' repo-relative references (`protocols/...`, `schema/...`,
+`annotations/...`) resolve because the agent runs at the repository root.
+
+### Installing on Claude.ai (web / mobile / desktop)
+
+Do **not** zip the `skills/<name>/` folder directly: the skill depends on files
+outside that folder (protocols, schema, exemplars) that a bare zip would miss.
+Instead build a self-contained bundle with the provided script:
+
+```bash
+./build_skill_bundle.sh --skill annotation   # -> genetic-evidence-annotation-skill.zip
+./build_skill_bundle.sh --skill review        # -> genetic-evidence-review-skill.zip
+# add --check for an input-validation dry run
+```
+
+The script gathers `SKILL.md`, the relevant protocols, the schema, and (for the
+annotation skill) the four curator-led exemplar annotations into one zip,
+rewriting the paths for the flat bundle layout. Upload the resulting zip via
+Customize > Skills > + Create skill (the bundle's top-level folder must be the
+root of the zip, which the script ensures).
+
+### Using the skills
+
+Once installed, ask in natural language, for example:
+
+> Annotate `papers/smith2024.pdf` under the genetic-evidence model.
+
+> Review the GEM annotation in `annotations/smith2024.yaml` against the paper.
+
+The annotation skill confirms the input paper, output path, and schema location,
+then produces a single YAML file matching the existing annotations. The review
+skill walks the annotation item by item with the curator and emits its three
+artifacts (review log, review report, and the updated annotation).
+
+## Extraction pipeline
+
+The `forome.gem.extraction` modules read a PDF with highlights and callouts
+and emit a structured JSON record of every annotation, including the text
+covered by each highlight (extracted via coordinate lookup) and the free-text
+notes attached to callouts.
+
+```bash
+python3 -m forome.gem.extraction.extract_annotations paper.pdf out.json
+```
+
+Two extractors are provided. `extract_annotations.py` uses PyMuPDF and is the
+recommended one: it reliably recovers the text under each highlight.
+`extract_annotations_pypdf.py` uses pypdf and is provided as a fallback.
+
+Dependencies:
+
+```
+pip install pymupdf pypdf pyyaml
+```
+
+## Schema
+
+Two complementary representations:
+
+- `schema/genetic_evidence.shacl.ttl`: SHACL shapes encoding the class
+  hierarchy, dimension types, cardinalities, and conditional activation
+  rules. This is the machine-checkable validation layer.
+- `schema/dimensions.md`: the human-readable enumeration reference for all
+  categorical value types (knowledge domain, method, target type, etc.).
+- `schema/EXTENSIONS.md`: the authoritative log of all candidate extensions
+  surfaced during corpus annotation, including their promotion or
+  retraction status.
+
+Annotations in `annotations/` are validated by the CI workflow in
+`.github/workflows/validate.yml`, which runs on every push and pull request
+and blocks merges on failure:
+
+- **`parse-yaml`** — every annotation YAML parses cleanly.
+- **`shacl-validate`** — `scripts/validate_annotations.py` converts each
+  annotation to RDF (`extraction/yaml_to_rdf.py`) and runs `pyshacl` against
+  `schema/genetic_evidence.shacl.ttl`. The shapes enforce the always-required
+  dimensions, the value enumerations, the implemented conditional-activation
+  rules (variant ascertainment, mode of inheritance, organism), and a
+  mandatory `source_span` on every assertion; an annotation that violates any
+  of these is rejected. Remaining conditional-presence and reviewer-flag
+  shapes are open work (see `schema/examples.md`).
+- **`coverage`** — `scripts/compute_coverage.py` regenerates the
+  dimension-coverage table from the YAML annotations (the source of the
+  paper's Supplementary Note SN7 and `annotations/coverage.md`).
+
+Run the same checks locally with `python3 scripts/validate_annotations.py`
+and `python3 scripts/compute_coverage.py` (requires `pyshacl rdflib pyyaml`).
+
+## UMLS crosswalk and the GEM Mapping Studio
+
+`data/umls/` holds the term-level crosswalk of the dimensional vocabulary to
+UMLS (`umls_crosswalk.yaml`), the curator adjudications behind it, and the
+public decision log `DECISIONS.md`. The crosswalk was produced with the
+**GEM Mapping Studio**, a standalone, model-agnostic curation tool for
+defining mapping axes and adjudicating value-level mappings against UMLS:
+
+```bash
+pip install gem-mapping-studio      # PyPI; console scripts gem-mapping-studio, gem-validate, gem-coverage, ...
+```
+
+Usage guide: **[`docs/STUDIO.md`](docs/STUDIO.md)** — how to build a new
+mapping from scratch in any domain, and how to work on the GEM mapping in
+this repository (from a checkout with `pip install -e .`, or with the
+released tool pointed at `data/umls`). The package also ships `gem-validate`
+and `gem-coverage`, which reproduce the corpus validation and the coverage
+table reported in the paper from a clean checkout.
+
+## Citing this work
+
+See `CITATION.cff`. Concept DOI (all versions):
+[10.5281/zenodo.22260686](https://doi.org/10.5281/zenodo.22260686);
+the release described in the manuscript is v0.2.2,
+[10.5281/zenodo.22260773](https://doi.org/10.5281/zenodo.22260773).
+A new version DOI is minted via the Zenodo–GitHub integration at each
+tagged release (record metadata in `.zenodo.json`).
+
+## License
+
+Content (annotations, documentation, the paper sources) is licensed
+CC-BY-4.0 (`LICENSE.txt`). Code — `src/`, `scripts/`, and the extraction
+tooling — is licensed Apache-2.0 (`LICENSE-code.txt`).
+
+## Contact
+
+See the corresponding-author block on the paper's title page.

{gem_mapping_studio-0.2.2 → gem_mapping_studio-0.2.4}/README.md

@@ -1,5 +1,8 @@
 # A Semantic Model of Genetic Evidence
 
+[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.22260686.svg)](https://doi.org/10.5281/zenodo.22260686)
+[![PyPI](https://img.shields.io/pypi/v/gem-mapping-studio.svg)](https://pypi.org/project/gem-mapping-studio/)
+
 A conceptual framework for representing scientific and genetic evidence from
 the biomedical literature in a form suitable for variant interpretation,
 automated reasoning, and AI-ready clinical infrastructure.
@@ -18,9 +21,14 @@ journal submission.
 ├── case-reports/   Per-paper case reports (one for each of the six annotations)
 ├── protocols/      Annotation protocol documents (canonical rules + per-mode workflows)
 ├── skills/         Two Claude skills (annotation, review) that operationalise the protocols
-├── extraction/     Scripts that extract highlights and callouts from annotated PDFs
+├── data/umls/      UMLS crosswalk of the dimensional vocabulary + decision log (DECISIONS.md)
+├── src/python/     The `gem-mapping-studio` package: Mapping Studio, crosswalk harness,
+│                   validators, PDF highlight/callout extraction (`forome.gem.*`)
 ├── figures/        Source files for figures used in the paper
-└── .github/        CI configuration validating annotations against the schema
+├── docs/           Usage guides (STUDIO.md: the Mapping Studio)
+├── .github/        CI configuration validating annotations against the schema
+├── CHANGELOG.md    What changed in each tagged release
+└── KNOWN_LIMITATIONS.md  What the model, schema, corpus, and tooling do not yet do
 ```
 
 ## The annotation corpus
@@ -158,13 +166,13 @@ artifacts (review log, review report, and the updated annotation).
 
 ## Extraction pipeline
 
-The scripts in `extraction/` read a PDF with highlights and callouts and emit
-a structured JSON record of every annotation, including the text covered by
-each highlight (extracted via coordinate lookup) and the free-text notes
-attached to callouts.
+The `forome.gem.extraction` modules read a PDF with highlights and callouts
+and emit a structured JSON record of every annotation, including the text
+covered by each highlight (extracted via coordinate lookup) and the free-text
+notes attached to callouts.
 
 ```bash
-python3 extraction/extract_annotations.py paper.pdf out.json
+python3 -m forome.gem.extraction.extract_annotations paper.pdf out.json
 ```
 
 Two extractors are provided. `extract_annotations.py` uses PyMuPDF and is the
@@ -210,11 +218,33 @@ and blocks merges on failure:
 Run the same checks locally with `python3 scripts/validate_annotations.py`
 and `python3 scripts/compute_coverage.py` (requires `pyshacl rdflib pyyaml`).
 
+## UMLS crosswalk and the GEM Mapping Studio
+
+`data/umls/` holds the term-level crosswalk of the dimensional vocabulary to
+UMLS (`umls_crosswalk.yaml`), the curator adjudications behind it, and the
+public decision log `DECISIONS.md`. The crosswalk was produced with the
+**GEM Mapping Studio**, a standalone, model-agnostic curation tool for
+defining mapping axes and adjudicating value-level mappings against UMLS:
+
+```bash
+pip install gem-mapping-studio      # PyPI; console scripts gem-mapping-studio, gem-validate, gem-coverage, ...
+```
+
+Usage guide: **[`docs/STUDIO.md`](docs/STUDIO.md)** — how to build a new
+mapping from scratch in any domain, and how to work on the GEM mapping in
+this repository (from a checkout with `pip install -e .`, or with the
+released tool pointed at `data/umls`). The package also ships `gem-validate`
+and `gem-coverage`, which reproduce the corpus validation and the coverage
+table reported in the paper from a clean checkout.
+
 ## Citing this work
 
-See `CITATION.cff`. A DOI is minted via the Zenodo–GitHub integration at
-each tagged release (record metadata in `.zenodo.json`); the concept DOI
-will be added to `CITATION.cff` after the first release.
+See `CITATION.cff`. Concept DOI (all versions):
+[10.5281/zenodo.22260686](https://doi.org/10.5281/zenodo.22260686);
+the release described in the manuscript is v0.2.2,
+[10.5281/zenodo.22260773](https://doi.org/10.5281/zenodo.22260773).
+A new version DOI is minted via the Zenodo–GitHub integration at each
+tagged release (record metadata in `.zenodo.json`).
 
 ## License
 

{gem_mapping_studio-0.2.2 → gem_mapping_studio-0.2.4}/pyproject.toml

@@ -1,13 +1,14 @@
 [build-system]
-requires = ["setuptools>=68", "wheel"]
+requires = ["setuptools>=77"]
 build-backend = "setuptools.build_meta"
 
 [project]
 name = "gem-mapping-studio"
-version = "0.2.2"
+version = "0.2.4"
 description = "Forome Genetic Evidence Model: a reference data model, UMLS/OMOP crosswalk tooling, and SHACL validation for basic-science genetic evidence"
 readme = "README.md"
-license = { file = "LICENSE-code.txt" }
+license = "Apache-2.0"
+license-files = ["LICENSE-code.txt"]
 requires-python = ">=3.10"
 authors = [
     { name = "Michael Bouzinier", email = "michael.bouzinier@forome.org" },
@@ -19,7 +20,6 @@ keywords = [
 classifiers = [
     "Development Status :: 4 - Beta",
     "Intended Audience :: Science/Research",
-    "License :: OSI Approved :: Apache Software License",
     "Operating System :: OS Independent",
     "Programming Language :: Python :: 3",
     "Programming Language :: Python :: 3.10",