Draft implementation of compositional grounding

mira/modeling/__init__.py

@@ -1,9 +1,10 @@
 __all__ = ["Model", "Transition", "Variable", "ModelParameter"]
 
 import logging
-from typing import Mapping, Optional
+from typing import Dict, Hashable, Mapping, Optional
 
 from mira.metamodel import (
+    Concept,
     ControlledConversion,
     ControlledProduction,
     GroupedControlledConversion,
@@ -53,13 +54,13 @@ def get_parameter_key(transition_key, action):
 class Model:
     def __init__(self, template_model):
         self.template_model = template_model
-        self.variables = {}
+        self.variables: Dict[Hashable, Variable] = {}
         self.parameters = {}
         self.transitions = {}
         self.make_model()
 
     def assemble_variable(
-        self, concept, initials: Optional[Mapping[str, Initial]] = None,
+        self, concept: Concept, initials: Optional[Mapping[str, Initial]] = None,
     ):
         """Assemble a variable from a concept and optional
         dictionary of initial values.
@@ -80,8 +81,8 @@ def assemble_variable(
             ("identity", f"{k}:{v}")
             for k, v in concept.get_included_identifiers().items()
         )
-        context_key = sorted(concept.context.items())
-        key = [concept.name] + grounding_key + context_key
+        context_key = concept.get_properties_key()
+        key = [concept.name, *grounding_key, *context_key]
         key = tuple(key) if len(key) > 1 else key[0]
         if key in self.variables:
             return self.variables[key]

mira/modeling/viz.py

@@ -14,6 +14,11 @@
 ]
 
 
+def _process_context_keys(contexts):
+    # see Concept.get_properties_key for the structure
+    for context in contexts:
+        yield context[1], context[3] or context[4]
+
 class GraphicalModel:
     """Create a graphical representation of a transition model."""
 
@@ -30,7 +35,10 @@ def __init__(self, model: Model):
                 label = name
                 shape = "oval"
             else:
-                cc = " | ".join(f"{{{k} | {v}}}" for k, v in itt.chain(identifiers, contexts))
+                cc = " | ".join(
+                    f"{{{k} | {v}}}"
+                    for k, v in itt.chain(identifiers, _process_context_keys(contexts))
+                )
                 label = f"{{{name} | {cc}}}"
                 shape = "record"
             self.graph.add_node(

mira/sources/petri.py

@@ -95,11 +95,13 @@ def state_to_concept(state):
         identifiers = {}
     # Example: 'mira_context': "[('city', 'geonames:5128581')]"
     context = dict(ast.literal_eval(state['mira_context']))
-    return Concept(name=state['sname'],
-                   identifiers=identifiers,
-                   context=context,
-                   initial_value=state.get('mira_initial_value'))
-
+    concept = Concept(
+        name=state['sname'],
+        identifiers=identifiers,
+        initial_value=state.get('mira_initial_value'),
+    )
+    concept = concept.with_context(**context)
+    return concept
 
 def transition_to_templates(transition, input_concepts, output_concepts,
                             controller_concepts):

mira/sources/sbml.py

@@ -91,6 +91,7 @@ def __init__(self):
 
     def parse_uri(self, uri):
         if self.converter is None:
+            # self.converter = bioregistry.get_default_converter()
             self.converter = curies.Converter.from_reverse_prefix_map(
                 bioregistry.manager.get_reverse_prefix_map(include_prefixes=True)
             )
@@ -451,7 +452,7 @@ def _extract_concepts(sbml_model, *, model_id: Optional[str] = None) -> Mapping[
         annotation_string = species.getAnnotationString()
         if not annotation_string:
             logger.debug(f"[{model_id} species:{species_id}] had no annotations")
-            concepts[species_id] = Concept(name=species_name, identifiers={}, context={})
+            concepts[species_id] = Concept(name=species_name)
             continue
 
         annotation_tree = etree.fromstring(annotation_string)
@@ -494,23 +495,32 @@ def _extract_concepts(sbml_model, *, model_id: Optional[str] = None) -> Mapping[
             name=species_name or species_id,
             identifiers=identifiers,
             # TODO how to handle multiple properties? can we extend context to allow lists?
-            context=context,
-        )
+        ).with_context(**context)
         concept = grounding_normalize(concept)
         concepts[species_id] = concept
 
     return concepts
 
 
-def grounding_normalize(concept):
+def grounding_normalize(concept: Concept) -> Concept:
     # Has property acquired immunity == immune population
-    if not concept.get_curie()[0] and \
-            concept.context == {'property': 'ido:0000621'}:
+    concept_curie = concept.get_curie()
+
+    # Rewrite property to main identifier - this is specific to BioModels
+    if (
+        not concept_curie[0]  # i.e., not grounded
+        and concept.has_object_property("property", "ido:0000621")
+    ):
         concept.identifiers['ido'] = '0000592'
-        concept.context = {}
-    elif concept.get_curie() == ('ido', '0000514') and \
-            concept.context == {'property': 'ido:0000468'}:
-        concept.context = {}
+        concept.properties = []
+
+    # Delete annotation of "susceptibility to infectious agent" -
+    # this is specific to BioModels.
+    elif (
+        concept_curie == ('ido', '0000514')
+        and concept.has_object_property("property", "ido:0000468")
+    ):
+        concept.properties = []
     return concept
 
 

notebooks/compositional_grounding_demo.ipynb

@@ -0,0 +1,158 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "3196f03c",
+   "metadata": {},
+   "source": [
+    "# Compositional Grounding\n",
+    "\n",
+    "**Who cares?**\n",
+    "\n",
+    "Anyone using DKG to endow semantics on to their data strutures (TA1, TA2, TA3) but\n",
+    "want to make new terms based on existing terms\n",
+    "\n",
+    "## First Steps - Make a Concept"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "285b19b3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from mira.metamodel.templates import Concept\n",
+    "\n",
+    "susceptible = Concept(\n",
+    "    name=\"susceptible_population\",\n",
+    "    # Correspond to nodes in the DKG\n",
+    "    identifiers={\"ido\": \"0000514\"},\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c759e283",
+   "metadata": {},
+   "source": [
+    "## Old Way to add \"Context\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "49266bce",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "susceptible_vaccinated = Concept(\n",
+    "    name=\"susceptible_population\",\n",
+    "    # Correspond to nodes in the DKG\n",
+    "    identifiers={\"ido\": \"0000514\"},\n",
+    "    context={\n",
+    "        \"vaccination_status\": \"vo:0001376\",\n",
+    "    }\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "3678a60a",
+   "metadata": {},
+   "source": [
+    "## Problems with the Old Way\n",
+    "\n",
+    "1. Not much structure\n",
+    "2. Doesn't enable ascribing semantics to what each entry in this \"context\" dictionary means\n",
+    "3. Can't differentiate between references to ontology terms or primitive data types (int, float, bool, str)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5ba61bff",
+   "metadata": {},
+   "source": [
+    "## New Way\n",
+    "\n",
+    "Flavor 1: use ontology terms:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6c157048",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from mira.metamodel.template import ObjectProperty, Term\n",
+    "\n",
+    "susceptible_vaccinated = Concept(\n",
+    "    name=\"susceptible_population\",\n",
+    "    identifiers={\"ido\": \"0000514\"},\n",
+    "    properties=[\n",
+    "        ObjectProperty(\n",
+    "            predicate=Term(name=\"vaccination_status\"),\n",
+    "            value=Term(name=\"vaccinated\", identifiers={\"vo\", \"0001376\"})\n",
+    "        )\n",
+    "    ]\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "842515df",
+   "metadata": {},
+   "source": [
+    "Flavor 2: use booleans (e.g., to reduce complexity)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b46d78a8",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from mira.metamodel.template import DataProperty\n",
+    "\n",
+    "Concept.parse_raw({...})\n",
+    "\n",
+    "susceptible_vaccinated = Concept(\n",
+    "    name=\"susceptible_population\",\n",
+    "    identifiers={\"ido\": \"0000514\"},\n",
+    "    properties=[\n",
+    "        DataProperty(\n",
+    "            predicate=Term(\n",
+    "                name=\"vaccination_status\", \n",
+    "                identifiers={\"vo\", \"0001376\"},\n",
+    "            )\n",
+    "            value=True\n",
+    "        )\n",
+    "    ]\n",
+    ")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.0"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

tests/test_petri_source.py

@@ -1,5 +1,6 @@
 from mira.examples.sir import sir, susceptible, infected, recovered
 from mira.metamodel import *
+from mira.metamodel.templates import ObjectProperty, Term, _index_properties
 from mira.modeling import Model
 from mira.modeling.petri import PetriNetModel
 from mira.sources.petri import state_to_concept, template_model_from_petri_json
@@ -12,7 +13,14 @@ def test_state_to_concept():
     concept = state_to_concept(state)
     assert concept.name == 'susceptible_population'
     assert concept.identifiers == {'ido': '0000514'}
-    assert concept.context == {'city': 'geonames:5128581'}
+
+    properties_index = _index_properties(concept.properties)
+    assert properties_index == {
+        ("", "city"): ObjectProperty(
+            value=Term(name="", identifiers={"geonames": "5128581" }),
+            predicate=Term(name="city", identifiers={})
+        )
+    }
 
 
 def test_petri_reverse():

tests/test_template.py

@@ -3,7 +3,7 @@
 from mira.metamodel import ControlledConversion, Concept, NaturalConversion, \
     NaturalDegradation, Template, GroupedControlledConversion, \
     GroupedControlledProduction
-from mira.metamodel.templates import Config
+from mira.metamodel.templates import Config, DataProperty, ObjectProperty, Term
 from mira.dkg.web_client import is_ontological_child_web
 
 # Provide to tests that are not meant to test ontological refinements;
@@ -94,14 +94,35 @@ def test_concepts_equal():
     c1_w_ctx = c1.with_context(location="Berlin")
     c2 = Concept(name="infected population", identifiers={"ido": "0000511"})
     c2_w_ctx = c2.with_context(location="Stockholm")
-    c3 = Concept(name="infected population", context={"location": "Stockholm"})
-    c4 = Concept(name="infected population", context={"location": "Berlin"})
+    c3 = Concept(name="infected population", properties=[
+        ObjectProperty(
+            predicate=Term(name="location"),
+            value=Term(name="Stockholm"),
+        )
+    ])
+    c4 = Concept(name="infected population", properties=[
+        ObjectProperty(
+            predicate=Term(name="location"),
+            value=Term(name="Berlin"),
+        )
+    ])
+    c5 = Concept(name="infected population", properties=[
+        ObjectProperty(
+            predicate=Term(name="location"),
+            value=Term(name="Berlin"),
+        ),
+        DataProperty(
+            predicate=Term(name="vaccination status"),
+            value=True,
+        ),
+    ])
 
     assert c1.is_equal_to(c2)
     assert not c1_w_ctx.is_equal_to(c2_w_ctx, with_context=True)
     assert c1_w_ctx.is_equal_to(c2_w_ctx, with_context=False)
     assert c3.is_equal_to(c4, with_context=False)
     assert not c3.is_equal_to(c4, with_context=True)
+    assert not c4.is_equal_to(c5, with_context=True)
 
 
 def test_template_type_inequality_is_equal():
@@ -206,7 +227,8 @@ def test_concept_refinement_grounding():
 def test_concept_refinement_simple_context():
     spatial_region_gnd = Concept(name="spatial region", identifiers={"bfo": "0000006"})
     spatial_region_ctx = spatial_region_gnd.with_context(location="Stockholm")
-    assert len(spatial_region_ctx.context)
+    spatial_region_ctx_2 = spatial_region_ctx.with_context(status="Yup")
+    assert 1 == len(spatial_region_ctx.properties)
     kw = {"refinement_func": is_ontological_child_web, "with_context": True}
 
     # Test both empty
@@ -218,6 +240,10 @@ def test_concept_refinement_simple_context():
     # Test other has context, refined does not
     assert not spatial_region_gnd.refinement_of(spatial_region_ctx, **kw)
 
+    # Test 2-deep refinement
+    assert spatial_region_ctx_2.refinement_of(spatial_region_ctx, **kw)
+    assert not spatial_region_ctx.refinement_of(spatial_region_ctx_2, **kw)
+
 
 def test_concept_refinement_context():
     spatial_region_gnd = Concept(name="spatial region", identifiers={"bfo": "0000006"})