1 of 40

Biolink Model

Workshop

​

Biocuration 2020

​

https://bit.ly/biolink-model-workshop-biocuration-2020

2 of 40

Overview

  • Introduction to Biolink Model
  • Q&A and Discussion
  • Modeling a test dataset using Biolink Model
  • KG-COVID-19 and Biolink Model
  • Addressing community use-cases
  • Aligning community schemas

3 of 40

Introduction

4 of 40

Ontology

“Ontology is a formal specification of a shared conceptualization”

Tom Gruber

​

An ontology is a formal specification of concepts, from a particular domain of knowledge, that are arranged in a hierarchy, as a directed acyclic graph, where concepts are defined in relation to other concepts in the graph.

​

5 of 40

Knowledge Graph

​

  • A knowledge graph has many definitions
  • A broad and inclusive definition:

A knowledge graph (KG) is a graph that represents knowledge where entities are represented as nodes and relationships between these entities are represented as edges.

​

6 of 40

Knowledge Graph

​

​

  • Nodes in a network represent entities (or concepts) and edges represent relationships between entities
  • Graph formalisms:
    • Property graphs
    • RDF graphs
  • Knowledge graphs (KGs) have been around since 1970s (Semantic networks) or 2012 (Google Knowledge Graph)

​

7 of 40

Knowledge Graph

  • KGs are at the peak
  • Several KGs in the industry
  • And a proliferating number of KGs in life sciences

​

8 of 40

Biomedical KGs

​

  • Semantic MEDLINE Database
  • Hetionet
  • WikiData
  • Monarch Initiative
  • SPOKE
  • COVID-19 Graph
  • KG-COVID-19
  • ...

​

​

9 of 40

Advantages of KGs

  • Ability to flexibly represent heterogeneous data and knowledge
  • Ontological KGs allow for deductive inference through logical rules
  • KGs are useful for graph-based machine learning
    • Edge prediction
    • Node classification
  • Store of information for downstream applications

​

10 of 40

Challenges with KGs

  • The formalism used for representing nodes and edges
  • The vocabulary used for representing nodes and edges
    • In the case of nodes: ’Gene’, ‘gene’, ‘gene locus’, ‘genomic feature’, ‘sequenceFeature’
    • In the case of edges: ‘has_phenotype’, ‘has phenotype’, ‘HAS PHENOTYPE’
  • KGs typically lack schemas; Developed as silos
  • Choice of identifiers
  • Modeling decisions for representing data from various sources

​

​

11 of 40

NCATS Biomedical Data Translator

12 of 40

NCATS Biomedical Data Translator

  • To build a system that is capable of taking existing biological and biomedical datasets and translating them into insights
  • Multi-site, multi-year project
  • 14 teams spread across the US + Netherlands
    • Knowledge Providers
    • Autonomous Relay Agents
    • Autonomous Relay Service

​

​

13 of 40

Overview of the architecture

https://doi.org/10.1111/cts.12591

14 of 40

Overview of the architecture

​

We as a consortium agree on 3 things:

  • Shared data model
  • Shared specification for exchange
  • Shared set of tools

​

15 of 40

Biolink Model

​

16 of 40

Biolink Model

  • A high-level data model for representing biological and biomedical knowledge
  • Bridges multiple vocabularies and ontologies
  • Agnostic to the graph formalism used to represent knowledge
  • The model consists of:
    • Entities
    • Associations
    • Predicates
    • Properties

17 of 40

Entities

  • Nodes in a graph
  • Represents entities found in biological and biomedical knowledge
  • Arranged in a hierarchy
  • Root of all entities is the ‘named thing’ class

​

Examples: Gene, Protein, Disease, Phenotypic Feature

​

18 of 40

Entities

Each entity class has

  • its own unique stable URI
  • mappings to other ontologies
  • list of valid ID prefixes

​

Higher-level terms that can be used to categorize nodes in a KG.

​

For more detailed typing, one can use specific terms from an ontology.

​

​

19 of 40

Associations

  • Edges in a graph
  • Represents assertions or statements
  • A hierarchy of associations
  • The root of all associations is the ‘association’ class.

​

Example: GeneToGeneAssociation, GeneToDiseaseAssociation, DiseaseToPhenotypicFeatureAssociation

​

20 of 40

Associations

​

  • An association connects a subject node and an object node via a predicate
  • The nature of the association is defined based on its properties
  • An association can have properties like provided_by, evidence, publications
  • Certain associations can have additional properties that are unique, as in the case of DiseaseToPhenotypicFeatureAssociation
    • frequency_qualifier

​

21 of 40

Predicates

  • High-level relationships
  • Used as predicate in a statement
  • Has mappings to other ontologies
    • Most commonly to Relations Ontology

22 of 40

Biolink Model

  • Common dialect for representing knowledge
  • Bridge across Neo4j and RDF graphs
  • Mapping existing models to the Biolink Model
  • Shared space for discussion on,
    • Evidence
    • Provenance
    • Context
    • Confidence

23 of 40

Curation process

Within Translator,

  • Weekly data modeling calls
  • Following a set of well defined guidelines for governance
  • Set up weekly Help Desk for users

​

Broadly, we want to support use cases from the wider community.

​

​

24 of 40

Tools to work with Biolink Model

25 of 40

Biolinkml

  • Biolink Modeling Language: https://github.com/biolink/biolinkml
  • A modeling framework
  • YAML as the source of truth
  • Generate,
    • JSON Schema: validation for JSON
    • Python Dataclasses: building Python APIs and writing ETL
    • Java classes: building Java APIs and writing ETL
    • GraphQL: building APIs on top of data stores
    • JSON-LD context: RDF to JSON serialization
    • RDF Turtle: RDF graphs
    • OWL: reasoning
    • Shape Expressions (ShEx): validation of RDF graphs

​

26 of 40

Biolinkml

  • Polymorphism
    • (mixins/traits + strict inheritance)
    • Class-specific overrides of slots
  • Rich annotation
  • Imports (borrowed from OWL)
  • Formal semantics

definition

Class definition

Slot definition

element

has 0..*

is_a 0..1

range 0..1

schema

imports 0..*

Core metamodel

(simplified subset)

27 of 40

Biolink Model Toolkit

  • A Python API for working with the Biolink Model
  • Provides convenience methods for querying the model
  • https://github.com/biolink/biolink-model-toolkit

​

28 of 40

Knowledge Graph Exchange

  • https://github.com/biolink/kgx
  • A Python library and set of command line utilities for exchanging Knowledge Graphs (KGs) that conform to or are aligned to the Biolink Model.
  • KGX allows you to work with,
    • SPARQL endpoints or RDF serializations
    • Neo4j endpoints or Neo4j dumps
    • CSV/TSV
    • JSON
    • OWL
    • OBOGraph JSON

​

29 of 40

Knowledge Graph Exchange

  • Transform KGs from one graph formalism to another
  • Create KGs or subgraphs
  • Merge two or more KGs
  • Validate KG against the Biolink Model
  • Apply graph operations
    • Graph Merge
    • Clique Merge
    • ID remapping
    • Graph summary

​

​

​

30 of 40

Biolink Model in the real world

31 of 40

SRI Reference KG

  • A Biolink Model compliant version of the Monarch Knowledge Graph
  • Built for the NCATS Biomedical Data Translator
  • Contains ontologies like GO, HP, MONDO, UBERON, ChEBI
  • Contains datasets like BioGRID, Reactome, ClinVar, Orphanet, OMIM, STRING
  • Available at scigraph.ncats.io

​

32 of 40

KG-COVID-19

  • A framework for building a KG with datasets relevant for COVID-19
  • Part of an even larger project called the National Virtual Biotechnology Laboratory
  • First instance of the Knowledge Graph Hub
  • https://github.com/Knowledge-Graph-Hub/kg-covid-19

33 of 40

Illuminating the Druggable Genome

  • Build a KG as a substrate for machine learning with the goal of link prediction
  • Uses Biolink Model for representing various curated datasets and databases like DrugCentral

34 of 40

Bringing it all together

35 of 40

Knowledge Graph Hub

  • The goal of KG Hub is to serve as a collective resource to simplify the process of generating biological and biomedical KGs and thus reducing the barrier for entry to new participants
  • In a KG Hub, each independent effort for building a KG is an instance of the KG Hub
  • Each KG follows a set of design principles, one of which is the use of Biolink Model as a schema

​

36 of 40

Knowledge Graph Hub

  • Also provides recommendations on set of design patterns to use for building and exchanging KGs
  • More information about KG Hub can be found at knowledge-graph-hub.github.io
  • Instances of KG Hub can be found at https://github.com/knowledge-Graph-Hub

​

37 of 40

Biolink Model - Future directions

  • The model itself is constantly being refined
  • Future additions include,
    • Variant and genotypes (GENO)
    • Evidence, Provenance, Context, and Confidence (SEPIO)
    • Genomic locations (FALDO)
  • Align to other controlled vocabularies, schemas, and ontologies

​

​

38 of 40

Contributing to the Biolink Model

  • GitHub: https://github.com/biolink/biolink-model/
    • Issues for requesting improvements to the model
    • Pull Requests for contributing to the model
  • Documentation: https://biolink.github.io/biolink-model/
  • Biolink Model on Gitter

​

39 of 40

Acknowledgement

  • Members from NCATS Biomedical Data Translator Consortium
  • Members from Monarch Initiative
  • Members of BBOP

​

​

Funding

  • NCATS Biomedical Data Translator grant 1OT2TR003449-01

​

  • Chris Mungall (LBNL)
  • Harold Solbrig (John Hopkins University)
  • Anne Thessen (Oregon State University)
  • Matt Brush (Oregon Health State University)
  • Richard Bruskiewich (STAR Informatics)
  • Justin Reese (LBNL)
  • Jim Balhoff (RENCI)
  • Melissa Haendel (Oregon State University)
  • Kent Shefchek (Oregon State University)
  • Michel Dumontier (Maastricht University)
  • Vincent Emonet (Maastricht University)
  • Stephen Ramsey (Oregon State University)
  • Andrew Su (Scripps Research Institute)
  • Karamarie Fecho (RENCI)
  • Chris Bizon (RENCI)
  • Steven Cox (RENCI)
  • Vlado Dancik (Broad Institute)

​

​

40 of 40

Q&A