The Neuroscience Information Framework
Establishing a practical semantic framework for neuroscienceMaryann Martone, Ph. D.
University of California, San Diego
NIF TeamAmarnath Gupta, UCSD, Co Investigator
Jeff Grethe, UCSD, Co Investigator
Gordon Shepherd, Yale University
Perry Miller
Luis Marenco
David Van Essen, Washington University
Erin Reid
Paul Sternberg, Cal Tech
Arun Rangarajan
Hans Michael Muller
Giorgio Ascoli, George Mason University
Sridevi Polavarum
Anita Bandrowski, NIF Curator
Fahim Imam, NIF Ontology Engineer
Karen Skinner, NIH, Program Officer
Lee Hornbrook
Kara Lu
Vadim Astakhov
Xufei Qian
Chris Condit
Stephen Larson
Sarah Maynard
Bill Bug
Karen Skinner, NIH
What does this mean?
• 3D Volumes• 2D Images• Surface meshes• Tree structure• Ball and stick models• Little squiggly lines
Data People
Information systems
The Neuroscience Information Framework: Discovery and utilization of web-based resources for neuroscience
http://neuinfo.org
UCSD, Yale, Cal Tech, George Mason, Washington Univ
Supported by NIH Blueprint
A portal for finding and using neuroscience resources
A consistent framework for describing resources
Provides simultaneous search of multiple types of information, organized by category
Supported by an expansive ontology for neuroscience
Utilizes advanced technologies to search the “hidden web”
Where do I find…
• Data• Software tools• Materials• Services• Training• Jobs• Funding opportunities
• Websites• Databases• Catalogs• Literature• Supplementary
material• Information portals
...And how many are there?
NIF in action
Query expansion: Synonyms and related concepts
Boolean queriesData sources
categorized by “data type” and level of nervous
system
Simplified views of complex data
sources
Tutorials for using full resource when getting there from
NIF
Hippocampus OR “Cornu Ammonis” OR “Ammon’s horn”
NIF searches across multiple sources of information
• NIF data federation• Independent
databases registered with NIF or through web services
• NIF registry: catalog• NIF Web: custom
web index (work in progress)
• NIF literature: neuroscience-centered literature corpus (Textpresso)
Guiding principles of NIF• Builds heavily on existing technologies (open source tools)• Information resources come in many sizes and flavors• Framework has to work with resources as they are, not as we
wish them to be– Federated system; resources will be independently maintained– Very few use standard terminology or map to ontologies
• No single strategy will work for the current diversity of neuroscience resources
• Trying to design the framework so it will be as broadly applicable as possible to those who are trying to develop technologies
• Interface neuroscience to the broader life science community• Take advantage of emerging conventions in search and in
building web communities
Registering a Resource to NIF
Level 1NIF Registry: high level descriptions from NIF vocabularies supplied by human curators
Level 2Access to deeper content; mechanisms for query and discovery; DISCO protocol
Level 3Direct query of web accessible databaseAutomated registrationMapping of database content to NIF vocabulary by human
The NIF Registry
• Very, very simple model
• Annotated with NIF vocabularies• Resource type• Organism
• Reviewed by NIF curators
Level 2: Updates and deeper integration• DISCO involves a collection of files that reside on each participating
resource. These files store information describing: - attributes of the resource, e.g., description, contact person, content of the resource, etc. -> updates NIF registry - how to implement DISCO capabilities for the resource
• These files are maintained locally by the resource developers and are “harvested” by the central DISCO server.
• In this way, central NIF capabilities can be updated automatically as resources evolve over time.
• The developers of each resource choose which DISCO capabilities their resource will utilize
Luis Marenco, MD, Rixin Wang, PhD, Perry L. Miller, MD, PhD, Gordon Shepherd, MD, DPhilYale University School of Medicine
DISCO Level 2 Interoperation
• Level 2 interoperation is designed for resources that have only Web interfaces (no database API).
• Different resources require different approaches to achieve Level 2 interoperation. Examples are:
1. CRCNS - requires metadata tagging of Web pages
2. DrugBank - requires directed traversal of Web pages to extract data into a NIF data repository
3. GeneNetwork - requires Web-based queries to achieve “relational-like” views using “wrappers”
DrugBank Example
The DrugBank Web interface showing data about a specific drug (Phentoin).
DrugBank Example (continued)
This DISCO Interoperation file specifies how to extract data from the DrugBank Web interface automatically.
DrugBank Example (continued)
A NIF user views data retrieved from DrugBank in response to a query in a transparent, integrated fashion.
Level 3• Deep query of federated databases with
programmatic interface• Register schema with NIF
– Expose views of database– Map vocabulary to NIFSTD
• Currently works with relational and XML databases– RDF capability planned for NIF 2.5 (April 2010)
• Works with NIF registry: databases also annotated according to data type and biological area
Integrated views and gene search
Is GRM1 in cerebral cortex?
• NIF system allows easy search over multiple sources of information• Well known difficulties in search
• Inconsistent and sparse annotation of scientific data• Many different names for the same thing• No standards for data exchange or annotation at the semantic level
– Lack of standards in data annotation require a lot of human investment in reconciling information from different sources
Allen Brain Atlas
MGD
Gensat
Cerebral CortexAtlas Children Parent
Genepaint Neocortex, Olfactory cortex (Olfactory bulb; piriform cortex), hippocampus
Telencephalon
ABA Cortical plate, Olfactory areas, Hippocampal Formation
Cerebrum
MBAT (cortex) Hippocampus, Olfactory, Frontal, Perirhinal cortex, entorhinal cortex
Forebrain
MBL Doesn’t appear
GENSAT Not defined Telencephalon
BrainInfo frontal lobe, insula, temporal lobe, limbic lobe, occipital lobe
Telencephalon
BrainmapsEntorhinal, insular, 6, 8, 4, A SII 17, Prp, SI
Telencephalon
Modular ontologies for neuroscience
NIF covers multiple structural scales and domains of relevance to neuroscience Incorporated existing ontologies where possible; extending them for neuroscience where necessary Normalized under the Basic Formal Ontology: an upper ontology used by the OBO Foundry Based on BIRNLex: Neuroscientists didn’t like too many choices Cross-domain relationships are being built in separate files Encoded in OWL-DL, but also maintained in a Wiki form, a relational database form and any other way it is needed
NIFSTD
NS FunctionMolecule InvestigationSubcellular Anatomy
Macromolecule Gene
Molecule Descriptors
Techniques
Reagent Protocols
Cell
Instruments
Bill Bug
NS Dysfunction QualityMacroscopicAnatomyOrganism
Resource
How are ontologies used?
• Search: query expansion– Synonyms– Related classes– “concept based queries”
• Annotation:– Resource categorization– Entity mapping
• Ranking of results– NIF Registry; NIF Web
Concept-based search: Entity mapping
Brodmann area 3 Brodmann.3
Synonyms and explicit mapping of database content help smooth over terminology differences and custom terminologies
Concept-based query: GABAergic neuron
• Simple search will not return examples of GABAergic neurons unless the data are explicitly tagged as such
• Too many possible classifications to get them all by keywords
• Classes are logically defined in NIF ontology• i.e., a GABAergic neuron is any
neuron that uses GABA as a neurotransmitter
Reclassification based on logical definitions: GABA neuron is any member of class neuron that has neurotransmitter GABA
NIF Bridge File• NIF Cell• NIF Molecule
• Define a set of properties that relate neuron classes to molecule classes, e.g.,
• Neuron• Has
neurotransmitter
• Purkinje cell is a Neuron
• Purkinje cell has neurotransmitter GABA
NIF Bridge Files: Defining classes to enhance neuroscience search
• Neuron by neurotransmitter• Neuron by brain region• Neuron by circuit role• Neuron by morphology
• Molecule by role• Neurotransmitter• Drug• Drug of abuse
NIF 2.5: Linking NIF entities
NIF Cards: Linked data
NIF 3.0+: NIF Annotation Standards
• Tying keyword search to quantitative definitions
• NIF is developing mappings between “data types” (e.g., age) and NIF annotation standards, e.g., adult
• Query string: adult mouse– Translation: Adult mouse = >/= 40 days postnatal
Building NIF ontologies: Balancing act• Different schools of thought as to how to build vocabularies and
ontologies• NIF is trying to navigate these waters, keeping in mind:
– NIF is for both humans and machines– Our primary concern is data– We have to meet the needs of the community– We have a budget and deadlines
• Building ontologies is difficult even for limited domains, never mind all of neuroscience, but we’ve learned a few things– Reuse what’s there: trying to re-use URI’s rather than map when possible– Make what you do reusable: adopt best practices where feasible
• Numerical identifiers, unique labels, single asserted simple hierarchies– Engage the community– Avoid “religious” wars: separate the science from the informatics– Start simple and add more complexity
• Create modular building blocks from which other things can be built
Ontologies, etc
Mike Bergman
What we’ve learned• Strategy: Create modular building blocks that can be knit
into many things– Step 1: Build core lexicon (NeuroLex)
• Classes and their definitions• Simple single inheritance and non-controversial hierarchies• Each module covers only a single domain• Understandable by an average human
– Step 2: NIFSTD: standardize modules under same upper ontology• OBO compliant in OWL
– Step 3: Create intra-domain and more useful hierarchies using properties and restrictions
– Brain partonomy
– Step 4: Bridge two or more domains using a standard set of relations– Neuron to brain region– Neuron to molecule, e.g., GABAergic neuron
Neurolex• More human centric• More stable class structure
– Ontologies take time; many versions are retired-not good for information systems that are using identifiers
• Synonyms and abbreviations were essential for users• Can’t annotate if they can’t find it• Can’t use it for search if they can’t find it• Facilitates semi-automated mapping
• Contains subsets of ontologies that are useful to neuroscientists– e.g., only classes in Chebi that neuroscientists use
• Wanted the community to be able to see it and use it– Simple understandable hierarchies
• Removed the independent continuants, entity etc• Used labels that humans could understand
– Even if they were plural (meninges vs meninx)
Maintaining multiple versions
• NIF maintains the NIF vocabularies in different forms for different purposes– Neurolex Wiki: Lexicon for community review and
comment– NIFSTD: set of modular OWL files normalized
under BFO and available for download– NCBO Bioportal for visibility and mapping services– Ontoquest: NIF’s ontology server
• Relational store customized for OWL ontologies• Materialized inferred hierarchies for more efficient
queries
NeuroLex Wiki
http://neurolex.org Stephen Larson
“The human interface”Semantic media wiki
NIF Architecture
Gupta et al., Neuroinformatics, 2008 Sep;6(3):205-17
Summary
• NIF has tried to adopt a flexible, practical approach to assembling, extending and using community ontologies – We use a combination of search strategies: string based, lexicon-based,
ontology-based– We believe in modularity– We believe in starting simple and adding complexity– We believe in single asserted hierarchies and multiple inferred hierarchies– We believe in balancing practicality and rigor
• NIF is working through the International Neuroinformatics Coordinating Facility (INCF) to engage the community to help build out the Neurolex and start adding additional relations– Neuron registry task force– Structural lexicon
Where do we go from here?• NIF 2.5
– Automatic expansion of logically defined terms
• More services– NIF vocabulary
services are available now
• More content• More, more, more NIF Blog
NIF is learning lessons in practical data integration
Musings from the NIF…• No single approach, technology, philosophy, tool,
platform will solve everything• The value of making resources discoverable is not
appreciated• Developing resources (tools, databases, data) that are
interoperable at this point is an act of will• Decisions can be made at the outset that will make it easier or harder to
integrate• We build resources for ourselves and our constituents, not
automated agents• We get mad when commercial providers don’t make their products
interoperable• Many times the choice of terminology is based on expediency or who
taught you biology rather than deep philosophical differences
• Ontologies, terminologies, lexicons, thesauri• Developing a semantic framework on top of unstable resources is difficult• Need to adopt a flexible approach
NIF Evolution
V1.0: NIFNIF 1.5+++++
Then Now Later