The Library of Integrative Network-based Cellular Signatures (LINCS) is an NIH Common Fund program. The idea is to perturb different types of human cells with many different types of perturbations such as: drugs and other small molecules; genetic manipulations such as knockdown or overexpression of single genes; manipulation of the extracellular microenvironment conditions, for example, growing cells on different surfaces, and more. These perturbations are applied to various types of human cells including induced pluripotent stem cells from patients, differentiated into various lineages such as neurons or cardiomyocytes. Then, to better understand the molecular networks that are affected by these perturbations, changes in level of many different variables are measured including: mRNAs, proteins, and metabolites, as well as cellular phenotypic changes such as changes in cell morphology. The BD2K-LINCS Data Coordination and Integration Center (DCIC) is commissioned to organize, analyze, visualize and integrate this data with other publicly available relevant resources. In this course we briefly introduce the DCIC and the various Centers that collect data for LINCS. We then cover metadata and how metadata is linked to ontologies. We then present data processing and normalization methods to clean and harmonize LINCS data. This follow discussions about how data is served as RESTful APIs. Most importantly, the course covers computational methods including: data clustering, gene-set enrichment analysis, interactive data visualization, and supervised learning. Finally, we introduce crowdsourcing/citizen-science projects where students can work together in teams to extract expression signatures from public databases and then query such collections of signatures against LINCS data for predicting small molecules as potential therapeutics.
Introduction to Metadata and Ontologies | Part 2
Loading...
來自 Icahn School of Medicine at Mount Sinai 的課程
Big Data Science with the BD2K-LINCS Data Coordination and Integration Center
12 個評分
Icahn School of Medicine at Mount Sinai
12 個評分
從本節課中
Metadata and Ontologies
This module includes a broad high level description of the concepts behind metadata and ontologies and how these are applied to LINCS datasets.
與講師見面
Avi Ma’ayan, PhDDirector, Mount Sinai Center for Bioinformatics
Professor, Department of Pharmacological Sciences
[SOUND] Hi, my name
is Stephan Schurer
from the University
of Miami Medical school.
This lecture is an introduction to data standards, metadata, and ontologies.
It follows a brief overview given in the previous lecture.
In the Biomedical Data Sciences,
we can categorize data standards into three general categories.
The first one is reporting guidelines or
checklists, these are also often called minimum information standards.
The second one are controlled vocabularies terminologies and
the third one are data exchange formats.
So, the reporting guidelines are minimum information checklists to specify what
kind of information need to be captured about an experiment or
a study for a particular purpose.
So, for example, if you have a survey about people,
then you may want to capture their age, gender or other information.
Controlled Vocabularies, they
are terminological resources that provide identification and definition of entities.
So we have a unique format to refer to something.
For example for gender, people choose male or female.
Data exchange formats are specifications how data are encoded so
that they are computer readable and computer processable.
This is of course very important for different software systems or
different computational agents can talk to each other,
so they understand each other's information.
So typical data exchange formats for example, would be JSON, or Tab Delimited,
or Comma Delimited CSV.
An important consideration for data is how they are organized.
So data structures refer to the organization of data.
So for example, the entity relations diagram in the relational database schema.
One ontology that organizes information in a certain way or
how data is structured in a schemalist database or
the relationship of different objects and object oriented database.
Many biomedical data standards have been developed over the years for
many different applications.
So it can be very hard to find those on the Internet.
One solution to this problem is the BioSharing Standards site,
which is maintained at Oxford University E-Research Center.
The BioSharing information resources have been curated.
So that means they have been reviewed and organized by experts.
That adds a lot of value, so
you can be sure that those resources are maintained and of high quality.
BioSharing site includes many terminological resources, data format and
model resources and reporting guidelines.
These are the three data standards that we covered earlier.
In addition it also has progressively policies, databases, and other resources.
But why are metadata so important?
And specifically, why are standardized, formalized metadata so important?
What justifies the significant effort of developing reporting guidelines,
terminologies and data exchange formats.
It should be quite clear that we really can do anything with data if you don't
have metadata.
So really, metadata are always on the data but
here we're talking specifically about standardized metadata.
And that is very important if you want to exchange information,
reuse information, share information, and build information systems.
So specifically, metadata are very important to
facilitate data replicability, reproducibility, and data reuse.
They're also critical to enable us to interpret results,
perform data analysis, and develop hypotheses.
Metadata enable the repurposing of data for other projects.
So data that I developed for one purpose in one project can be
used in another project because they fit together based on their specific metadata.
And metadata are important for
information systems so we can query and search data sets,
we can integrate data and you can exchange data between different systems.
So, for example, any term or any information that you search on data or
by which you search data are typically captured by metadata and
to do this in an organized way, we need metadata standards.
Talking about Data Replicatability, Data Reproducibility and Reuse of Data.
It is very important to precisely define what those terms actually mean.
In many cases, in the videos talk about reproducibility, but
in fact they mean replicability or in some cases they even mean just repeatability.
So here is very useful definition of those different terms.
Repeat means the same experiment done in the same laboratory,
often by the same individual, exactly the same way.
So repeat simply means doing exactly the same thing in exactly the same way
multiple times, and getting the same results.
Replicate means do the same experiment but in a different laboratory.
So it refers to the fact that somebody else can do the same exact experiment.
Reproduce refers to the same experiment but in a different setup.
So that would be different instrumentation, different laboratory,
different individuals, different reagents, or at least different reagent batches.
So Reproducibility means that they can get
the same kind of biological result but not in exactly the same way
by repeating exactly the same experiment in the same way.
So I want to re-emphasize the point here Replicability is not Reproducibility.
Reuse of data is, of course, a different experiment.
But they have some connection between the two experiments or the two different
datasets and this connection is made explicit by standard estimated data.
So I can use data generated in one experiment to compliment data generated
in another experiment.
This data integration or
data linking is only possible with explicit standardized metadata.
So to combine different types of datasets generated in different experiments
that have some connection of course requires that there are standardized and
unified metadata across those different experiments.
And of course to reuse data and to add value requires a data of high quality.
So, that would typically also implies that the data are reproducible.
What information, which specific metadata should be captured for
particular types of experiment if you find the minimal information reporting
guidelines as mentioned earlier.
A great resource for minimal information reporting guidelines is the MIBBI,
Minimal Information for Biological and Biomedical Investigations.
This MIBBI foundry is now integrated in the BioSharing cycle I mentioned earlier.
If followed, many more information reporting guidelines assure that the data
can be easily verified, analyzed, and
clearly interpreted by the wider scientific community.
These recommendations also facilitate a foundation of structure databases,
public repositories, and the development of data analysis tools.
To build information systems and
data analysis tools requires not only that the minimum metadata are captured, but
also that the metadata are captured in a standardized and unified way.
This is a controlled vocabularies command.
The BioSharing site makes an effort to actually link
reporting guidelines to terminological resources.
Controlled vocabularies or
thesauri describe what things mean by linking terms to human descriptions.
So they link entities into identity criteria, so
that they can refer to the same thing, using a language term.
By doing so, they enable us to share knowledge in a common language.
Thesauri also often includes natural language synonyms, and
those are important for search and text mining applications.
Controlled vocabularies therefore facilitate the linking and
searching of information in software systems.
For example, in the system they can search or analyze different types of
information that are linked by those controlled vocabularies.
And, of course, they enable humans but not computational agents to reference and
exchange knowledge.
If you want to build a computational system that can operate on knowledge,
one first has to formalize knowledge in a way that it can be understood and
processed by the computer.
One way of doing this is using ontologies.
The original use of the term ontology
goes back to the Greek philosophers Parmeniedes and Plato.
The philosophical definition of ontology is the study of existence and
the nature of being.
So in philosophy, ontology would ask which things exist,
what is the nature of things, and what is reality?
The computer science definition of ontology, which we are talking about here
is a formal description of knowledge of a subject domain of interest.
So an ontology would be a formal,
logic based definition of the types, their properties and
their interrelationships within real objects in a particular subject domain.
For example, to formalize the domain family, we can define the relationships
husband wife, mother daughter, brother sister, and so on.
If done properly, the knowledge in the subject domain,
in this case family, can be formalized using logical actions.
So for example, in the case of family we would understand that a cousin
means a child of a father's or mother's, brother or sister, for example.
To be computer processable, of course all of this also needs to be
formalized in a machine processable specification.
One such data format for ontologies, for example, is Web Ontology Language,
in using a particular version of logic or description logic.
So given all this, a very brief definition of ontology can be given as,
an ontology is a specification of a conceptualization.
In addition to controlled vocabulary in which entities are defined using
human definitions, An ontology contains entities, which are called classes and
their relationships which are called object properties.
By doing so
an ontology allows to capture an abstract knowledge using logical axioms.
And this is done using explicit specification based on logic.
And using a language.
For example, BEP, Ontology Description Language Description Logic, OWL-DL.
So an ontology allows us to build a formal knowledge model, and
allows to compute with that knowledge, using so-called reasoning engines.
Ontologies are a foundation of semantic web information systems.
Semantic Web is also sometimes referred to as Web 3.0.
So ontologies are formalized representations of knowledge to
enable computing said knowledge.
That is in contrast to controlled vocabularies,
which allow humans to exchange knowledge but not computational agents.
For those of you who know relational databases and
are interested in Semantic Web technologies, here is a brief contrast of
relational database management systems for those ontologies.
Relational databases operate under the closed world assumption.
That means there is a pre-defined schema and
every entity has to fit somewhere in that schema.
Ontologies, in contrast, operate under the open world assumption.
That means that a class is defined by its relationships to other classes and that
the reasoning engine decides where it is placed in the framework of the ontology.
Relational database management systems do not support reasoning.
Everything we want to query from the system needs to be put in explicitly to
the database.
In contrast, semantic web technologies support inference reasoning.
For example, it would be relatively simple to infer somebody's cousin
if you know his or her parents, the parents' siblings and their children.
So in contrast, the relational database system,
we do not explicitly have to put in who is somebody's cousin.
But you can infer that information.
To ask specialized query to a relational database, you need to know the schema and
all of the combined information from different tables.
In contrast, semantic web technologies,
via formal semantics, provide a restriction free framework to ask queries.
In a way, those queries work by traversing a graph that is defined by the ontology.
As a consequence, in relational database systems,
data sharing is often not easy, because there are no formal semantics.
And if you want to exchange any information,
the schemas would have to be somehow compatible.
In contrast, semantic web technologies provide easier data sharing and
knowledge sharing.
Data and knowledge sharing using semantic web technologies requires that
the ontologies talk to each other, and this is facilitated by formal semantics,
and this is relatively easy for common domains.
For example, family, where everybody agrees what the relationships and
the classes are.
But it is much more difficult and not yet achieved for very complex knowledge.
For example, in the biomedical domains where there are many different ontologies
for overlapping subdomains.
Information systems based on semantic web technologies use so-called triple stores.
In a triple store,
all relationships between all individuals are explicitly stored.
Relational databases are very powerful for
very large datasets that all have the same structure.
Where triple store can be more flexible, and
it's better applicable to very complex domain knowledge.
Relational database systems are an established technology, and
there are industry standards, for example, Oracle.
Open source examples of relational database systems include MySQL and
Postgres.
In contrast, semantic web technologies such as Jena or Virtuoso are still
relatively early stage, and standards in this domain are still emerging.
Hundreds of biomedical ontologies have been developed over the years.
One of the most comprehensive repositories of biomedical ontologies is
the NCBO Bioportal, which has already been introduced in the previous lecture.
Another important resource of biological and
biomedical ontologies is the OBO Foundry.
In contrast to the NCBO Bioportal,
which has a very open policy in allowing developers to deposit their ontologies,
the OBO Foundry takes a much more selective approach.
OBO Foundry ontologies have to comply with Foundry principles.
Including design decisions, naming conventions and
the use of certain upper level ontologies.
New ontologies are typically admitted as OBO Foundry candidate anthologies and only
after a thorough review process can they be promoted to OBO Foundry ontologies.
One of the goal of the OBO Foundry is to promote the collection of compatible, and
non-overlapping biological and biomedical ontologies.
The OBO Foundry is a good starting point to research existing biological and
biomedical ontologies.
Another important resource of biological and biomedical ontologies
is the EBI Ontology Lookup Service at the European Bioinformatics Institute.
The EBI Ontology Lookup Service provides a centralized query interface for
almost 100 ontologies.
During the last section of this lecture,
I want to briefly introduce the LINCS metadata standards.
Recall that LINCS signatures have three primary dimensions.
The first dimension is the biological model system.
LINCS by logical model systems include proliferating immortalised cell lines,
primary cells and use pure reportant stem cells and differentiated cells.
Another dimension is the perturbation of the model system.
LINCS perturbagens include small molecules, RNAis, proteins and
other reagents.
The third dimension of LINC signatures are the molecular entities and
cellular features that are detected unquantified in an assay.
These, for example, can include genes which are quantified in transcription and
profiling assays, or proteins which are detected and quantified via antibodies and
proteomics assays.
They also include phosphoproteins that can be quantified via mass spectrometry for
example via P100 Probe and P100 phosphoprotein or mass profiling assays.
These also include cellular features which are detected in various links
phenotypic cell profiling assays.
We have been developing metadata standards to uniquely identify molecular entities,
model systems,
and other concepts that are important to describe link signatures and assays.
To metadata standards specifications assure that all metadata are kept to
sufficient detail that enable us to integrate the data And to create a common
view across all the different LINCS data generated by different LINCS assets.
During identifying metadata entities, we also capture annotations that
are important to link LINCS data to third party resources, and
to analyze and query link data.
For example, cell lines annotated by tissues and
diseases, often molecules will be annotated by known targets.
LINCS metadata standards specifications for many categories have been released and
are published on the LINCSproject.org website under Data Standards.
The standardized LINCS metadata entities including small molecules,
cells, proteins, transcribed genes, and their annotations, Are registered and
stored in the dedicated information system, the LINCS Metadata Registry.
The LINCS Metadata Registry also captures the assays,
organization, projects, roles, and data sets, and
all the relationships between those and the LINCS metadata categories.
It is accessible here, the given URL.
Clicking on any of the titles of the user interface it will bring up a table with
all the entries and the annotations of that particular category.
For example, here, link sell lines with values annotations including species,
organ and many different other annotations, IDs, provider and so on.
This interface also allows filtering and
text based clearing of all the records of this given category.
It also allows downloading of all the entries and the annotations.
One can also download a template which can then be used
to upload additional records of that given category.
The links may not add a registry It's integrated with other software systems.
Together, those systems form the integrated knowledge environment that we
are developing for the Links project.
The design of the MDR system as a core component of the Links integrated
knowledge environment emphasizes the importance of structured and
standardized metadata.
Finally, a selected list of useful ontology resources.
The NCBO Bioportal It's the largest repository of biomedical ontologies.
The OBO Foundry includes only carefully selected and
thoroughly reviewed ontologies.
The EBI Ontology Lookup Service provides a web service interface to carry multiple
ontologies from a single location with a unified output format.
OntoBee is a link data server for
ontologies It de-references an ontology term and
allow us to use that term via its URL in an HTML webpage or audio source code.
OntoFox is a web based ontology tool that allow us to extract ontology terms and
actions.
It is useful to extract modules from existing ontologies and
reuse those modules in new ontologies.
Protege is probably the most widely used ontology editor.
Which you will need to develop or modify an ontology.
It is a free open source tool, developed at Stanford University.
An introductory guide to develop your own first ontology,
Ontologies Development 101 is available at the Protege website.
Formal ontologies are to be developed in the Web Ontology Language.
The direct model theoretic semantics
of Web Ontology Language 3 is available at the W3C site.
This document requires a basic understanding of description logics.
An introduction to description logics is available on Wikipedia.
[MUSIC]
