Skip to content

Derived Schemas

This section describes rules that can be applied to a schema to obtain a derived schema.

The derived schema can be materialized as a static schema object, or it may be a dynamic view onto a schema object.

Derived schemas are also known as induced schemas or inferred schemas.

flowchart TD M[Asserted Schema] -->|input| Derivation{Derivation Procedure} Derivation -->|output| Mstar[Derived Schema] R[Rules] -->|input| Derivation

Derivations happen via derivation rules, using a collection of defined functions.

Conventions

We use m to denote the input or asserted schema (model), and mD to denote the derived schema

Informative Example

We provide a non-normative example of a schema in canonical YAML. This will be referred to in the remainder of the document.

The schema is organized in a local folder as follows:

person.yaml:

id: https://w3id.org/linkml/examples/person
name: person
prefixes:
  person: https://w3id.org/linkml/examples/person
  linkml: https://w3id.org/linkml/
default_prefix: person
imports:
  - linkml:types
  - core

classes:
  Person:
    is_a: NameThing
    description: >-
      A person, living or dead
    slots:
      - age_in_years
      - vital_status

slots:
    age_in_years:
      description: >-
        The age of a person in years
      range: integer
      multivalued: false
    vital_status:
      description: >-
        The vital status of a person
      range: VitalStatusEnum

enums:
  VitalStatusEnum:
    description: >-
      The vital status of a person
    permissible_values:
      ALIVE:
      DECEASED:

core.yaml:

id: https://w3id.org/linkml/examples/core
name: person-core
prefixes:
  person: https://w3id.org/linkml/examples/person
  linkml: https://w3id.org/linkml/
default_prefix: person
imports:
  - linkml:types

classes:
  NamedThing:
    attributes:
      id:
        range: string
        identifier: true
      name:
        range: string
...

Functions

This section defines functions that can be used in schema derivation rules.

Normalize as List Function

The function L(v) normalizes v to a list/collection:

v L(v)
None []
[v1, v2, ...] [v1, v2, ...]
v if no preceding matches apply [v]

i.e.

L(v) = 
  [] if v == None
  v if v == [...]
  [v] otherwise

Identifier Value Function

The function K(v) returns the identifier value of an InstanceOfClassDefinition, if present, otherwise None

In the metamodel, the identifier SlotDefinitionName is always name, so this can be calculated:

K(v) = v.name

URI and CURIE functions

  • a CURIE is a string that conforms to the W3 CURIE specification
  • a URI is a string that conforms to rfc3987
  • The function URI(s, v) takes a string as input and normalizes it to a URI
    • if the input is already a URI, then return it
    • if the input is a CURIE, then expand it using the prefix map in s; this is known as expansion
  • The function CURIE(s, v) takes a string as input and normalizes it to a CURIE
    • if the input is already a CURIE, then return it
    • if the input is a URI, then shorten it using the prefix map in s; this is known as contraction

When a CURIE is expanded, it is first decomposed according to the grammar defined in W3 CURIE specification:

curie := [ [ prefix ] ':' ] reference

prefix := NCName

reference := irelative-ref (as defined in IRI)

The following concatenation is applied

URI(<prefix>: <reference>) = m.prefixes[prefix].prefix_reference + <reference>

For the reverse operation, any URI that starts with a prefix reference can be used to generate a valid CURIE. The one with the shortest reference is chosen as the canonical.

Resolve Functions

The function Resolve takes as input a InstanceOfReference and returns an InstanceOfClass that is the referenced object.

i Resolve(i)
SlotDefinition&<V> m.slots[<V>]
ClassDefinition&<V> m.classes[<V>]
EnumDefinition&<V> m.enums[<V>]
TypeDefinition&<V> m.types[<V>]
SchemaDefinition&<V> see below

The rules for schema resolution are as follows:

  1. Determine the location of V, which may be on the local file system or on a network
    • if V is in a user-supplied import map, then look up this value
    • if V is a URL, then retrieve from this location
    • if V is a CURIE, then expand using the rules below, and retrieve from this location
    • otherwise retrieve from local file system, relative to the current schema, assuming a .yaml suffix
  2. Load the YAML file at this location following the rules for parsing YAML in Part 6, yielding a SchemaDefinition instance.

Examples:

i Resolve(i) in mD
ClassDefinition&Person m.classes["Person"] = ClassDefinition(name="Person", ...)
SlotDefinition&vital_status m.slots["vital_status"] = SlotDefinition(name="vital_status", ...)

Lookup Functions

The notation i.<slot> denotes a function that looks up the slot assignment slot in instance i. If i is a InstanceOfReference it is first Resolved, as per the table above. If i is collection of references, then each element is resolved.

Closure Function

The function C(e, f) takes as input an element e and a function f and returns the mathematical closure of f

C(e, f) = { f(e) } U { f(e') | e' in C(e, f) }

The ReflexiveClosure C* includes e

C*(e, f) = C(e, f) U { e }

Function: Parents

Parents itself is the union of is_a and mixins, plus the builtin ClassDefinition Any:

P(e) = L(e.is_a) ∪ L(e.mixins) } ∪ { Any }

Function: Ancestors

The ancestors function A returns the Closure of the Parents function

A(e) = C(e, P)

The function ReflexiveAncestors A uses the ReflexiveClosure*.

A(e) = C(e, P**)

Function: Imports Closure

The imports closure function I returns the reflexive ReferenceClosure of the direct imports.

I(s) = C(s, s.imports*)

Function: Element CURIEs and URIs

For each element in the schema, we can derive a model URI and an element URI. These may be identical.

The following table is used:

Type Element URI Model URI
ClassDefinition e.class_uri <m.default_prefix>:<SafeCamel(e.name)>
SlotDefinition e.slot_uri <m.default_prefix>:<SafeSnake(e.name)>
TypeDefinition e.uri <m.default_prefix>:<SafeCamel(e.name)>
EnumDefinition e.enum_uri <m.default_prefix>:<SafeCamel(e.name)>
PermissibleValue e.meaning ModelURI(Enum) + "." + Safe(e.text)

If the element URI slot is not set, then the model URI is used as the element URI.

URIs can be contracted to CURIEs using the prefixes map in the model, see next function.

Function: Applicable Slots

The applicable slots for a class are all valid slots to use for an instance of that class. It is the set of all direct slots for that class and its ancestors.

ApplicableSlots(c) = c'A(c) DirectSlots(c'*)

DirectSlots(c) = L(c.slots) ∪ L(c.attributes)

Algorithm: Combine Slots

The combine slot algorithm takes two slots s1 and s2. s1 is assumed to have precedence over s2.

CombineSlots(s1, s2) = 
  s = SlotDefinition()
  for ms in MetaModel.slots:
    s.ms = CombineSlotsMetaslots(ms, s1.ms, s2.ms)

The CombineSlotsMetaslots function is evaluated by looking up the table below:

metaslot v1 v2 expression
v v v
None v2 v2
v1 None v1
.name == maximum_value min(v1,v2)
.name == minimum_value max(v1,v2)
.name == pattern CombinePattern(v1,v2)
.name == range A(r1) ∩ A(r2)
.multivalued == True L(v1) U L(v2)
.range == boolean v1 OR v2
v1

The preconditions are matched in order of precedence, thus if no preconditions save the last one match, then the value is taken from the first, higher precedence slot.

Algorithm: Combine Schemas

Two schemas combined by combining their individual elements

CombineSchemas(m1, m2):
  m = SchemaDefinition()
  m.classes = CombineElements(m1.classes, m2.classes)
  m.types = CombineElements(m1.types, m2.types)
  m.slots = CombineElements(m1.slots, m2.slots)
  m.enums = CombineElements(m1.enums, m2.enums)
  m.subsets = CombineElements(m1.subsets, m2.subsets)  
CombineElements(E1, E2):
  E1ids = { e.id for e in E1 }
  E2ids = { e.id for e in E2 }
  for e in E1:
    if e.id in E2ids:
      raise Error
    else:
      yield e
  for e in E2:
    if e.id in E1ids:
      raise Error
    else:
      yield e    

Algorithm: Calculate Derived Slot

DerivedSlot(s, c) takes a slot definition name s and a class definition name c and returns a derived slot that is pre-populated with the appropriate inferred values.

The procedure first applies slot usage and attributes from the class and its ancestors, then it applies top level slot definitions for that slot and the slot ancestors. In the latter case, only metaslots marked inheritable are propagated.

DerivedSlot(m,s,c):
  d = SlotDefinition(name=s, alias=s)
  ApplySlotUsage(d,s,c)
  if s in m.slots:
    d = CombineSlots(d, m.slots[s])
    for s' in A(s):
      for ms in MetaModel.slots:
        if ms.inheritable:
          d.ms = CombineSlotsMetaslots(ms, d.ms, m.slots[s'].ms)   
  AddMissingValues(d, c) 
  return d

The ApplySlotUsage function iteratively combines the value of the slot_usage and attributes metaslots for a class, and then for class ancestors, with mixins having priority.

ApplySlotUsage(d, s, c):
  for s' in { c.slot_usage, c.attributes}
    s' = CombineSlots(d, s')
  for c' in **L**(c.`mixins`) ∪ **L**(c.`is_a`)
    ApplySlotUsage(d, s, c') 

The function AddMissingValues(s, c) is calculated for a SlotDefinition according to the following table

Preconditions Postconditions
s.inlined_as_dict=True s.inlined=True
PK(c)=None s.inlined=True

Algorithm: Calculate Permissible Values

The set of permissible values for an Enum may be specified as a static fixes list OR as a dynamic query OR both

PVs(e):
  pvs = Texts(e.permissible_values)
  for e' in e.inherits:
    pvs = pvs U PVs(e')
  for e' in e.minus:
    pvs = pvs - PVs(e')
  for pv in e.concepts:
    pvs = pvs U pv
  for q in e.reachable_from:
    pvs = pvs U ResolveQuery(q)

The ResolveQuery applied to q operates over an external ontology of vocabulary resource indicated by r.source_ontology. Each such ontology has a graph presentation which the query is resolved against. Each resource may choose to present itself as a graph in an application specific way. For OWL ontologies, it is recommended that the OWL TBox is presented as a graph following the Relation Graph pattern, i.e. axioms such as A SubClassOf R some B are presented as edges from A to B with label R.

ResolveQuery(q: EnumExpression):
  G = as_graph(q.source_ontology)
  G = FilterEdges(p in g.relationship_types)
  pvs = {}
  for n in q.source_nodes
    pvs = pvs U Closure(G, parent, n)
    if q.reflexive:
      pvs = pvs U {n}
  return pvs

Derivation Rules

The following rules are applied to deduce a derived schema mD.

Rule: Ensure Metamodel ids are unique

Each m' in I(m) must have a unique identifier, m'.id.

not exists m' in I(m) and m' in I(m) and m'.id == m'.id

Rule: Populate Schema Metadata

For each m' in I(m), populate the following metamodel slots:

  • for each element e in m', set e.from_schema to m'
  • if m'.default_range is not set, set it to string.

Rule: Combine Import Closure

A derived model mD is the result of combining the import closure of m:

mD = Aggregate(CombineSchemas, I(m))

When copying an element x from an import into mD, the name x.name must be unique - if the same name has been used in another model, the derivation procedure fails, and an error is thrown.

Rule: Derived Attributes

Derived attributes are populated by applying the DerivedSlot algorithm to each applicable slot for any class

for c in m.classes:
  for s in ApplicableSlots(c):
    d = DerivedSlot(m,s,c)
    c.attributes[s] = d

Rule: Derived Class and Slot URIs

For each class or slot, if a class_uri or slot_uri is not specified, then this is derived using the rules specified for Element URIs above.

For example, given:

prefixes:
  foo: http://example.org/foo/
  bar: http://example.org/bar/
default_prefix: foo
classes:
  A:
    class_uri: bar:A
    ...
  B:
    ...

the following values would be set:

  • A.class_uri = http://example.org/bar/A
  • B.class_uri = http://example.org/foo/B

Rule: Derived Permissible Values

TODO

Rule: Generation of patterns from structured patterns

For any slot s, if s.structured_pattern = p and p is not None then s.pattern is assigned a value based on the following procedure:

If p.interpolated is True, then the value of s.syntax is interpolated, by replacing all occurrences of braced text {VAR} with the value of VAR. The value of VAR is obtained using m.settings[VAR], where m is the schema in which p is introduced.

If p.interpolated is not True, then the value of s.syntax is used directly.

If p.partial_match is not True, then s.pattern has a '^' character inserted at the beginning and a '$' character inserted as the end.

Structural Conformance Rules

This section specifies conformance rules which the derived schema must satisfy.

Note that some of these rules may be derivable from treating the schema as an instance of the metamodel and applying validation rules in part 5, but we still list them below for completeness.

Rule: Each referenced entity must be present

Every ClassDefinition, ClassDefinitionReference, SlotDefinitionReference, EnumDefinitionReference, and TypeDefinitionReference must be resolvable within mD

However, not every element needs to be referenced. For example, it is valid to have a list of SlotDefinitions that are never used in mD.

ClassDefinition Structural Conformance Rules

Each c in mD.classes must conform to the rules below:

  • c must be an instance of a ClassDefinition
  • c must have a unique name c.name, and this name must not be shared by any other class or element in mD
  • c lists permissible slots in c.slots, the range of this is a reference to a SlotDefinition in mD.slots
  • c defines how slots are used in the context of c via a collection of SlotDefinitions specified in c.slot_usage
  • c may define local slots using c.attributes, the value of this is a. collection of SlotDefinitions
  • c may have certain boolean properties defined such as c.mixin and c.abstract
  • c must have exactly one value for c.class_uri in mD, and the value must be an instance of the builtin type UriOrCurie
  • c may have parent ClassDefinitions defined via c.is_a and c.mixins
    • the value of c.is_a must be a ClassDefinitionReference
    • the value of c.mixins must be a collection of ClassDefinitonReferences
  • For any parent p of c, if p.mixin is True, then c.mixin SHOULD be True
  • c includes additional rules in c.rules and c.classificiation_rules
  • c may have any number of additional slot-value assignments consistent with the validation rules provided here with the metamodel MM

SlotDefinition Structural Conformance Rules

Each s in mD.slots must conform to the rules below:

  • s must be an instance of a SlotDefinition
  • s must have a unique name s.name, and this name must not be shared by any other type or element
  • s must have a range specified via s.range in mD
  • s may have an assignment s.identifier which is True if s plays the role of a unique identifier
  • s may have certain boolean properties defined such as s.mixin and s.abstract
  • s must have exactly one value for s.slot_uri in mD, and the value must be an instance of the builtin type UriOrCurie
  • s may have parent SlotDefinitions defined via s.is_a and s.mixins
    • the value of s.is_a must be a SlotDefinitionReference
    • the value of s.mixins must be a collection of SlotDefinitionReferences
    • For any parent p of s, if p.mixin is True, then s.mixin SHOULD be True
  • s may have any number of additional slot-value assignments consistent with the validation rules provided here with the metamodel MM

TypeDefinition Structural Conformance Rules

Each s in mD.types must conform to the rules below:

  • t must be an instance of a TypeDefinition
  • t must have a unique name t.name, and this name must not be shared by any other type or element
  • t must have a mapping to an xsd type provided via t.uri in mD
  • t may have a parent type declared via t.typeof
  • t may have any number of additional slot-value assignments consistent with the validation rules provided here with the metamodel MM

EnumDefinition Structural Conformance Rules

Each e in mD.enums must conform to the rules below:

  • e must be an instance of a EnumDefinition
  • e must have a unique name e.name, and this name must not be shared by any other enum or element
  • e lists all static permissible values via e.permissible_values, the value of which is a list of instances of the MM class PermissibleValue
  • e may have any number of additional slot-value assignments consistent with the validation rules provided here with the metamodel MM

ClassDefinitionReference Structural Conformance Rules

Each r in mD.class_references must conform to the rules below:

  • r must be an instance of a ClassDefinitionReference
  • r must have a unique name r.name, and this name must not be shared by any other type or element

Metamodel Conformance Rules

Both the asserted and derived schema should be valid instances of the LinkML metamodel MM using the instance validation rules described in the next section