Content uploaded by Juha Muilu
Author content
All content in this area was uploaded by Juha Muilu on Mar 18, 2014
Content may be subject to copyright.
Genotype-Phenotype Databases:
Challenges and Solutions For The Post-Genomic Era
Gudmundur A Thorisson(1), Juha Muilu(2), Anthony J Brookes(1)*
(1) Department of Genetics, University of Leicester, University Road, Leicester,
LE1 7RH, UK.
(2) Institute for Molecular Medicine Finland, University of Helsinki,
Haartmaninkatu 8, FIN-00290, Helsinki, Finland.
* Corresponding Author:
Tel: +44 (0)116 2523401
Fax: +44 (0)116 2523378
Email: ajb97@leicester.ac.uk
Abstract
Today’s flow of research data concerning the genetic basis of health and disease is
rapidly increasing in speed and complexity. In response, many projects are seeking to
ensure that there are appropriate informatics tools, systems, and databases available to
manage and exploit this flood of information. Previous solutions such as central
databases, journal based publication, and manually intensive data curation are now
being enhanced with new systems for federated databases, database publication, and
more automated management of data flows and quality control. Along with emerging
technologies that enhance connectivity and data retrieval, these advances should help
to create a powerful knowledge environment for genotype-phenotype information.
The World Wide Web has become an indispensable tool for biomedical researchers
striving to understand how genes cause disease. Websites such as the PubMed
literature search service1, the Ensembl2, UCSC3 and NCBI1 genome browsers, and the
BLAST1 sequence search service, are examples of the internet resources that many
biologists use on an almost daily basis. Behind the scenes these resources are based
upon very similar technologies and design principles (i.e., they have standard
‘architectures’), but their user-interfaces differ widely in terms of style, functionality and
content. This diversity complements the diverse needs of the field, but to investigate a
given biological question a user may need to browse many websites and still never feel
sure they’ve tracked down all the information they might need.
While the proliferation of data resources can be frustrating for traditional biologists it
presents an even bigger challenge for ‘omics’ researchers who need to automate large-
scale data aggregation across many different sites. Historically, such users were forced
to write software to automatically surf websites to extract information originally designed
for human consumption. As noted by Stein4, this ‘screen scraping’ approach has
numerous disadvantages. Instead, there are better ways to inter-connect large sets of
related information so that they can be searched and downloaded from a single portal.
In this review, we consider how this has been tackled in the past for genotype-to-
phenotype (G2P) data, and look at how the relevant technologies are currently being
improved. We discuss some of the technical issues surrounding database development,
and the recent trend towards an increased emphasis upon federated database
solutions, which can link independent databases through a central portal and be married
with the proven benefits of traditional central databases in which related data is stored
all in one place. Looking further into the future, we consider even more revolutionary
approaches to data integration and utilization, and discuss potential challenges that
need to be addressed.
Lessons from the past
To understand the obstacles and the opportunities surrounding modern G2P databases,
it is helpful to consider how the field has grown and evolved into its current state. Until
recently, online stores of genetic data tended to be built as ‘central databases’ (Figure
1), and this model has served the field very well given its previous requirements.
Sequence Databases. The earliest databases of prominence in genetics were designed
to hold DNA sequence data. In the early 1980s, as soon as the use of commercial
technologies for DNA sequencing became widespread, such depositories were needed
to facilitate exchange and comparison of DNA sequences. Three major central
databases were constructed for this purpose in Japan, the USA, and Europe; the
DDBJ5, GenBank6 and EMBL7 respectively. In the mid 1980s the International
Nucleotide Sequence Database Collaboration (INSDC, http://www.insdc.org) was
established to promote full content exchange between these databases
The INSDC constitutes a large-scale central database project, and it provides an
excellent example of how effective the central databasing strategy can be. Arguably,
however, this project succeeded as well as it did because: DNA sequence information is
simple to represent as a directly annotated string of letters and sequence regions (i.e. it
is ‘1-dimensional’), and despite a massive growth in data volume, the scale of the
problem did not exceed the capabilities provided by concomitant advances in computer
technology.
Model Organism Databases. Model organism databases (MODs) provide a second
example of how the central database model can be used effectively. These databases
specialize in capturing genomic, phenotypic, and other information for a model
organism. Examples include Wormbase8, the Rat Genome Database9 and the Mouse
Genome Informatics Database10. A single or a few groups working closely together,
armed with expert knowledge on their model organism of interest, were the typical
creators of these early central G2P databases. The resulting websites provide a focal
point for information gathering and access, plus centralised services such as a genome
browser interface and tools for comparative genomics.
A simplistic assessment of MODs would put their success down to the very limited
volume of data they have to manage, compared to the amounts flowing into a global
nucleotide sequence database. But this view fails to allow for the fact that the data
contained in an MOD are far more diverse and complex than mere 1-dimensional
sequence strings (i.e., comprising genetic and phenotype related ‘2-dimensional’
information). This complexity makes it far more difficult to organize the data within a
single depository. The MODs overcame this hurdle by the virtue of good leadership and
the relatively small community sizes that made it possible for agreements to be reached
on matters such as data model standards, laboratory protocols, terminologies, and
curation practices. Having these standards in place ensured that effective data
management, interpretation and exchange could occur between the central database
and many different laboratories. The MOD experience thus emphasizes the absolute
need for robust and universal standards in order to aggregate and integrate G2P data.
Extending this principle, MODs for many different species are now working together as
part of the Generic Model Organism Database project (GMOD, http://www.gmod.org) to
further harmonise and standardize their activities.
Central Databases for Human ‘Mendelian Mutations’. The databasing of human G2P
relationships has lagged behind what has been achieved in other species. There are
many reasons for this, one of which is the far larger size and dispersed and diverse
nature of the underlying research community that necessarily includes biologists,
clinicians, epidemiologists, statisticians, etc. This has made it difficult to agree and
deploy a full series of universal standards. Furthermore, the standards themselves are
difficult to devise for human G2P relationships due to the complexity the data. These
complexities include: the full spectrum of medical diagnoses and clinical test results that
are often open to subjective interpretation; a wide range of normal traits that may vary
with age and between populations; and a myriad of sequencing, genotyping and other
laboratory procedures that are employed for the generation of primary data, which itself
may be analysed and utilized in a plethora of different ways.
However, attempts have been made to capture a broad picture of human Mendelian
G2P knowledge via the central database model. Good progress has been made by the
Online Mendelian Inheritance in Man11 (OMIM:
http://www.ncbi.nlm.nih.gov/sites/entrez?db=omim) which provides a genotype-
phenotype catalogue of human genetic disorders, and which first appeared in book form
over 40 years ago12. The project went online in 1990 and is now maintained by a
substantial team of curators who manually extract experimental findings from the
literature. This has resulted in a compilation of high-quality records on ~11,000 genes
and diseases (as of April 1st, 2008). But this is a long way from being a fully
comprehensive summary of all knowledge on human G2P relationships, even for
Mendelian disorders. The task is simply too large for one team to manage in a
centralized fashion, and given the complexity of the source information, OMIM is forced
to package its content in narrative form. This makes it unsuitable for automated mining
or deep integration with other database resources. Similar data collection issues are
being faced by the Human Gene Mutation Database13 (HGMD: http://www.biobase-
international.com/pages/index.php?id=hgmddatabase), which uses manual curation to
extract from the literature a list of mutations that underlie Mendelian disorders, to then
place these in a structured and readily searchable format. Another project with data
collection challenges is the Pharmacogenetics and Pharmacogenomics Knowledge
Base14 (PharmGKB: http://www.pharmgkb.org) which collates extensive knowledge
about the relationships amongst drugs, diseases and genes, to assist
pharmacogenomics research. These examples illustrate some fundamental limitations,
primarily relating to data complexity and quantity, to the central database concept,
especially in the context of human G2P information.
Locus-Specific Databases for Human ‘Mendelian Mutations’. Taking an alternative
approach, many groups have collated primarily Mendelian G2P information for just one
or a few genes of relevance to one or a few diseases of interest. The first of these
‘Locus-Specific Databases’ (LSDBs) was published in 1976, in the form of a catalog of
human globin mutations15. Following a slow but steady increase in the number of online
LSDBs, ~700 are now listed at the Human Genome Variation Society website16
(http://www.hgvs.org/dblist/glsdb.html). LSDB entries tend to be rich in information
content and enhanced by domain-specific expert curation. As well as published
information they typically include unpublished DNA variation along with evidence
concerning pathogenicity. Unfortunately, these databases are created independently of
one another, with little coordination or harmonisation, and with little or no dedicated
funding.
LSDBs range from a simple, non-networked spreadsheets through to fully-fledged
online databases. Consequently they represent a fragmented network of silos full of rich
information (Figure 2) across which it is not possible to efficiently exchange or integrate
G2P information. In their current disjointed form, LSDBs are essentially at the opposite
end of the spectrum from central G2P databases that provide shallower but genome-
wide perspective. Neither is ideal, and so another approach - or combination of
approaches - would seem to be needed.
Challenges for Modern G2P Databases
MODs, LSDBs and related databases have taught us that it will not be straightforward to
computationally orchestrate the totality of human G2P information. Furthermore, the
challenge is growing: high-throughput genomic data generation is now within the reach
of many laboratory budgets. In addition it is now, or soon will be, possible to explore
phenomena such as structural variation, DNA methylation, rare/unique alleles, and even
somatic genome changes in a comprehensive manner.
Currently, genetic association studies, especially genome-wide association studies
(GWAS) represent a particularly prolific source of G2P data. The principal databases
set up to store and organize GWAS data include the dbGaP archive in the US17
(http://view.ncbi.nlm.nih.gov/dbgap), the EGA in Europe (http://www.ebi.ac.uk/ega), and
the GWAS Database of Japan (https://gwas.lifesciencedb.jp/cgi-
bin/gwasdb/gwas_top.cgi). Other related projects include HGVbaseG2P
(http://www.hgvbaseg2p.org), GAD18 (http://geneticassociationdb.nih.gov), the Type 1
Diabetes Genetics Consortium (http://www.t1dgc.org), and disease-specific efforts
AlzGene19 (http://www.alzforum.org/res/com/gen/alzgene), PDGene
(http://www.pdgene.org), and SZGene20
(http://www.schizophreniaforum.org/res/sczgene/). At present, all of these employ the
central database model, although it is unclear whether this approach will suffice in the
longer term. As the field advances these databases will have to grapple with complex
issues such as: increasingly convoluted data governance issues pertaining to different
countries and legislatures; rapid changes in the scale, depth and format of the primary
and processed data; and the likely addition of other forms of variation and levels of
etiologic complexity. The over-riding need is to achieve a sufficient degree of global
comprehensive coverage to make ‘2-dimensional’ G2P databases as successful as ‘1-
dimensional’ DNA sequence databases.
Database creators, their patrons, and their funding bodies are acutely aware of the need
for more and better human G2P databases. Based on past experiences as elaborated
above, and in light of recent technology developments, we suggest six key areas that
need attention if improved G2P databases are to be built:
Data quantity: The scale of current and future G2P research means that datasets will
keep getting larger and more numerous. This acceleration in the rate of data production
might even start to outstrip the ability of database technologies to handle the
information. For example, results may be repeatedly split and merged, and re-examined
– as in the case of GWAS studies which share control materials and cross-compare
their primary datasets, and which are being extended by further empirical work and by
statistical re-analysis. The SNP data underlying GWAS results are also now being used
to extract information on structural variation; and all of the above can be repeated ad-
infinitum for every phenotype and sub-phenotype characterized. The data volume will
be further increased due to the addition of other forms of variation, other areas of
bioscience, and the emergence of routine whole genome sequencing21. Therefore, data
quantity must be a key consideration in database design.
Data quality: Even though database records will never be completely error free, efforts
must be made to avoid inaccuracies wherever possible. Such activities can be split into
manual data curation efforts that involve reading and redrafting of data that involves
knowledge and concepts (e.g., from the literature into databases such as OMIM),
versus automated validation of generally larger datasets with more straightforward
content (such as markers, genotypes, and sequences). Quality control should be
optimized right from the stage of data generation, but databases can only become
involved from the stage of guiding researchers in the preparation of accurate and
appropriate data submissions. Once data is received, databases should then deploy
their own quality assurance measures to check for internal consistency and
completeness of the submission. In scenarios where this requires manual curation,
domain experts are invaluable, and they will often interact with the data submitters in
performing their task. Future federated and community curation efforts (e.g., Wiki
systems) will need to be carefully managed if they are to match the high standards
achieved by current manual curation activities. The responsibility for that will lie, at least
partly, with each database in the federated system, though oversight may be applied by
stakeholders such as funders, international consortia, and feedback systems from the
community as a whole. Today, databases obviously do try to ensure high data quality,
and they know that this is a challenging business22, but perhaps in the future their
obligations should go even further. For example, consistency with other data sources
could be assessed; such as by comparing SNP allele frequencies with previous
datasets to identify fundamental laboratory or data management errors, or cases where
the wrong DNA strand has been referenced. Across the full breadth of G2P data there
are many features that could be checked to ensure accuracy, and standards and
guidance need to be developed to underpin data curation throughout the path of
generating data through to placing it in public databases. Ideally, software support for
this will increasingly be provided.
Data complexity: Although data quantity is a matter for concern, it will hopefully be
overcome in time by improvements in data processing algorithms and innovations in
computer science. More indomitable, however will be the matter of data complexity.
Biological data, especially G2P data of the future, differs from that of most other ‘big
science’ (such as astronomy) by its high-level of complexity23. Consider phenotypes for
example: studies such as the UK Biobank (http://www.ukbiobank.ac.uk) and the
Framingham Heart Study (http://www.framinghamheartstudy.org) collect thousands of
phenotypic variables in a prospective manner, with each item supported by extensive
metadata (information that describes the primary data), for tens or hundreds of
thousands of subjects. The utilized phenotype definitions may change as knowledge
advances, and patient phenotype categorizations may change with age and treatment.
Given this data complexity, standardization of the ‘phenotype’ parameter is needed, and
this is one of the goals of several projects, such as P3G24, HuGENet25, and the PhenX
project (http://www.PhenX.org). The information that describes how genotypes connect
to phenotypes – i.e., the ‘2’ in G2P - is even more complex. A plethora of constantly
evolving methods, strategies, and analyses that offer varying levels of precision, may be
used to work out how DNA sequences control phenotypes. The results provide mere
clues (sometimes of a contradictory nature) as to the underlying etiologic processes.
Environmental effects (as being considered by the Genes, Environment and Health
Initiative: http://www.gei.nih.gov), a person’s genetic background, and chance, also feed
into this. Thus, the complexity of the analytical methods and the experimental results
make it difficult to store G2P information in a structured way, or to fully optimize the
integration and presentation systems.
Knowledge representation: As more and more analyses are performed on ever more
extensive and cross-domain datasets, it will become increasingly difficult to
comprehensively gather and present all the resulting hypotheses and conclusions
(‘knowledge representation’). The issue of how to present conclusions is distinct from
the question of what tools and systems are developed to generate those ideas, and how
the systems interface with databases. Traditionally, scientific journals have been the
principal vehicle for distributing the interpretation of data, but it not clear whether their
current modus operandi will enable them to keep pace as the rate of new discoveries
continues to grow exponentially. The human-readable narrative format of journals does
not easily lend itself to the storage of ‘interpretations’ and ‘concepts’ in databases.
However, some databases, such as OMIM, whose scope extends into knowledge
representation, do employ free text. This resource illustrates the value of handling this
kind of information beyond journals, but it also shows that it is limiting to store this data
without any rigid structure. G2P knowledge representation therefore needs to become
better structured and anchored upon appropriate ontologies if databases are to be more
than just high-tech lists of primary experimental data.
Data access: As G2P datasets become larger and more diverse it will become
increasingly difficult to locate any particular data item. To tackle this issue, more
powerful tools for database searching will obviously be needed, but it is equally
important that those improved search engines are connected to all the relevant data that
needs to be searched. Although this could be seen as an argument for widespread
adoption of the central database model, as detailed above, data size and complexity
make it impossible to gather all the information in one central depository. Instead,
complementary ways to consolidate the tasks of data access and data presentation
across many different databases (e.g., LSDBs) are needed, such that the interrogated
information itself never needs to leave its remote source. Such single point of access –
or federated database - solutions that tap into large volumes of diverse data are
technically feasible. An example from outside the G2P domain is the ENCODEdb
portal26, which offers a simple query interface that searches across all the ENCODE
experimental data deposited in several public databases.
Data sensitivity: Once the likely phenotypic consequences of carrying a particular
sequence variation are established, knowing one’s genotype at a given locus becomes
meaningful. It also becomes meaningful for one’s relatives who share various fractions
of your genome, and it probably is something that health providers, employers, insurers,
and even governments and the criminal justice system might want to (rightly or wrongly)
know about. This raises complex ethical dilemas27-30. Even if genome data are
anonymised (as they usually are in epidemiological studies), there is a risk of re-
identification of persons based on their genome variation profile, and/or their phenotype
and environmental profiles31. Exemplifying this, a recent paper has shown that an
individual’s involvement in a genetic study can be reliably established from just
summary level allele frequency data (i.e., no individual genotypes) if this is available for
two matched sample groups (such as GWAS cases and controls), and if the genome
profile of the subject of interest is known32. A full discussion of the myriad questions
surrounding ‘data sensitivity’ are beyond the scope of this review (see ref. 33), but a few
points are worthy of mention. Currently, most databases try to avoid showing sufficient
genotype or phenotype information to enable re-identification. This may not, however,
be as easy as it seems - for example, in LSDBs where rare mutations/diseases may be
reported along with geographical data. When G2P data do raise the possibility of re-
identification, the current default position is to not make it publicly available, and pass
access requests to the original custodians of the information. This stance was
immediately adopted for summary level allele frequency datasets once it became clear
that they could be used for individual identification34. But when even summary level data
cannot be shared for unfettered research access, maybe it is time to start questioning
when and where the protection of an individual’s privacy becomes overly paranoid or
too onerous to implement, given that it detracts from the wider research benefits of
making data freely available? If, in reality, it will not be possible to completely ensure the
anonymity of all research participants, then perhaps the optimal way forward may be to
accept this, to make data more freely available, and concentrate instead on preventing
and punishing abuse of the data. Given the existence of such perplexing privacy issues,
an ethics advisory voice should arguably be an integral part of every G2P database.
The Future: The Untapped Power of Federation
Given the above considerations, federated databases can be expected to play an
increasingly large role in the future of G2P information management. Before we discuss
that role in more depth, it is useful to reflect on extreme versions of the federated and
centralized models (Figure 1). The fully centralized model would involve all generated
information being automatically piped into one large data center, from where all search
and presentation activities are managed. A completely federated solution would involve
all the domains’ information being organized into geographically discrete packets
(databases), with no regular data flows between them. Global searches would be
mediated by portals that scan all available contents, and data presentation would be
powered by each database for each item of its own content. Neither of these extremes
is a realistic option for the G2P domain, due to the limitations of each model as
described below, and a hybrid model would seem to offer the best way forward. Until
now, however, most successful databases have been based on the centralized model.
This likely reflects the newness of a field that only came into being with the emergence
of the internet, and the fact that the current pressing need for more advanced solutions
is relatively recent. Now, as internet and database technologies rapidly advance,
federated systems are emerging alongside and intermingled with the existing
established central databases.
Both central and federated systems have advantages and disadvantages. The main
advantages of central databases include cost efficiency due to economies of scale,
ease of management, and reliable archiving of the community’s data. In contrast,
federated databases represent a more complicated solution in terms of the required
technologies, but they bring certain advantages that cannot be endowed by a central
database (as detailed below). In large part this relates to ‘ownership’ and accreditation
for the database teams, with the potential result that more and higher quality data can
be gathered in a federated system, due to the reward gained by the workers involved.
Federated and central database systems both provide centralized search capabilities,
although federated alternatives can also offer more sophisticated search options via
direct interrogation of the source databases.
Taking into account the challenges facing the G2P field as outlined above, and the pros
and cons of each model, it seems evident that a purely centralized model cannot fulfill
all the requirements of an optimal G2P databasing solution, and that some degree of
federation is critical to the success of this enterprise. So what would a group of
databases need to do to become usefully federated to an optimum degree? The first
decision would involve the level of federation to be achieved. Essentially, this equates to
deciding what portion of their content each database would wish to make available for
other computers to read over the internet. They might choose to provide none, and
instead transfer to one or more common search centers some pre-agreed ‘core’ data
elements for each record they hold, along with links back to those entries in their
database. The search system would then use that assemblage of minimal data items to
enable multi-database searches, and report search results as a series of annotated
links pointing back to the source databases. This partly centralized and partly federated
solution is being piloted by several closely collaborating initiatives as a way to begin
federating LSDBs35. Alternatively, and with more effort, the search platform could itself
go and get the full details for each record of interest from the different sources (perhaps
even by ‘screen-scraping’ if necessary) and compile this into a uniform dataset for
presentation.
A more elegant way of federating would involve making some or all of the record details
from each remote database directly searchable by other computers. This approach
removes the need for sending in and regularly updating core datasets, thereby ensuring
that searches through the central portal always query the very latest datasets. It also
addresses the scalability problems outlined in the previous section, since any new
LSDB merely needs to register its existence with the central portal to become part of the
multi-database search catalogue. Another advantage is that it minimizes the workload of
the central search system, as it no longer has to chase up and manage ever-changing
core datasets every 24 hours or so. Finally, it alleviates many of the data complexity
issues faced by central databases, in that each nodal database can provide and
customize (at the final display stage) whatever additional record details it deems
appropriate above and beyond the common data items made available as part of the
federated search. Achieving this ‘complete’ federation, however, requires all
participating databases to accept certain rules of the engagement. For example, the
level of autonomy that each team can enjoy, in terms of database design, system
execution, and the degree of association with the rest of the federation, must not be so
high as to make the whole federation ineffective. Furthermore, all nodal databases must
either adhere to certain standards so that their records can be easily integrated with
those of others, or place advanced ‘translation’ software on top of their database so that
search requests and result datasets can be freely communicated between remote and
local computers.
Finally, certain other advantages of federation are also worthy of specific comment. The
first relates to empowering and rewarding database creators (Figure 3). It takes effort to
design, build, fund and continuously manage and curate a database – and it is all too
often a thankless task. The federated model, however, places a lot more control and
recognition in the hands of those running the individual databases. Federated
databases have complete control over what records, and what details per record, are
made available to different users at any point in time. This may be very important in the
case of commercial databases, and it is highly important in the context of data
sensitivity as mentioned above. Second, the federated structure distributes data
management and curation work among many individuals, making the most of the expert
knowledge of these individuals. A third advantage is that the federated structure
enables new search portals to be set up quickly and easily, potentially offering unique
new perspectives – for example, a gene-centric view for researchers specializing in a
single gene, a disease-centric view for clinicians, a genome browser-based view for
genomics researchers. Fourth, federated networks by default operate as democracies,
so unilateral changes cannot be imposed on common aspects of the federated system
(e.g., data models). This does not mean that innovation becomes stifled, but rather that
new ideas will be widely debated, piloted and validated before they are implemented.
The G2P Database Network
Today, the components that are needed to create a powerful and highly integrated
system, based upon a partially federated and partially centralized model, are either
already available or in advanced stages of development. The key missing ingredients
needed to bring the G2P network to life relate to expanding technology awareness,
establishing recognition and reward systems, and targeting the appropriate allocation of
sufficient funding. In Europe, many of these issues are being tackled head-on by
initiatives such as the GEN2PHEN project(http://www.gen2phen.org), whilst consortia
such as BBMRI (http://www.biobanks.eu) and ELIXIR (http://www.elixir-europe.org) are
actively planning for the investment of up to several billion Euro into bioscience
database and biobanking infrastructures (see also Table 1).
Various technologies, such as web services and ontologies, that will underpin future
G2P databasing have been discussed elsewhere36,23 (see Box 1 for a summary of the
main components). Most importantly, the field will have to become increasingly
standardized in order for a global network of G2P databases to inter-operate effectively.
This standardization concern matters of both syntax (how data is organized) and
semantics (the meaning of each data item).
A core syntax challenge involves designing and validating robust data models for
different biomedical domains such that the models are inter-compatible. Typically, these
models are also accompanied by standard specifications for data exchange formats,
providing a basis for data exchange between systems. Various existing data models are
currently being cross-compared and harmonised to enable more widespread data
integration within a research domain, and even across different domains (see Box 2).
Semantic challenges involve ensuring that data items are represented in a way that
conveys the same meaning to each and every person (or computer) that reads them.
For example, a field named ‘sample’ may mean ‘blood sample’ in one database, but be
taken to mean ‘an individual sampled from a population’ in another database. The goal
is to structure and specify all of this in ‘ontologies’, and to build software and support
systems that ensure the terms are used correctly37. Semantic standardization is difficult
to achieve38, and tackling this issue across all bioscience sub-fields involves precisely
defining a complete domain lexicon. To break this mammoth task down, researchers
have set about working on ontologies for clearly demarcated subjects (such as ‘Gene’,
‘DNA sequence’ or ‘anatomy’). A large and highly collaborative network of ontology
groups has now grown into the Open Biomedical Ontologies (OBO) consortium39, which
is currently tackling what may be the biggest ontology challenge of all: ‘phenotypes’.
Encouragingly, much progress has already been made on this undertaking, especially
by the model organism database communities.
As syntactic and semantic standards start to fall into place, more groups are likely to
start to consider building federated databases. Technologies will then be required that
can broadcast, deliver, receive, or interrogate (locally or remotely) the available
datasets. An early example of one such technology is the Distributed Annotation
System (DAS)40 – a simple protocol for exchanging annotations on genomic sequences.
Many databases already make their records available via their own DAS server to DAS
clients such as the Ensembl browser. Those third-party datasets are then overlaid on
other DAS-supplied information or locally available annotations, such as reference sets
of genes - demonstrating the power of a federated system.
It is hoped that as databases, search platforms, visualization interfaces, and analysis
tools start to become fully federated, and seamlessly joined together, the system will be
transformed from a mere collection of cleverly connected data, into a universal G2P
‘knowledge environment’ – a place where new questions can be asked, new types of
experiments can be performed in silico, and where new knowledge can be created.
That, at least, is the vision of the ‘semantic web’41, the proponents of which claim will
comprise a highly-sophisticated and powerfully connected series of servers and client
computers (the ‘Grid’) that together provide highly-automated data retrieval and analysis
‘web-services’ across the Internet. When this Grid-enabled G2P knowledge
environment becomes a reality, we may find ourselves in a situation where the
distinction between database entries and research manuscripts has become blurred,
and where new paradigms like web publishing and real-time community markup of
databased information are common place. Initial forays into this world of merged
‘database-journal’ publication vehicles are already taking place, one example being a
partnership between the Human Genomics and Proteomics journal and the FINDbase
database (http://www.sage-hindawi.com/journals/hgp/). Indeed, traditional journals may
become a thing of the past, as they may well have evolved to become an integral part of
this unified Grid of biomedical knowledge.
Future challenges and opportunities
Despite the optimistic future of G2P databases, a number of basic issues (mostly non-
technical ones) remain to be solved.
One of these involves getting enough people sufficiently well trained in the relevant
technologies to build and connect all of the G2P databases. To achieve this, educators
must decide to fund and organize such training. In parallel, software engineers can
drastically reduce the level of competence required of the users by devising ‘off the
shelf’ solutions. This philosophy lies at the heart of the GEN2PHEN project, which is
producing empty ‘database-in-a-box’ installation packages along with training, and
open-source complete genetic association database systems for download – all of
which are based upon the PaGE-OM data model (http://www.pageom.org), giving them
the option of federation.
Another issue is the question of tracking who is building G2P databases and populating
them with useful data. Several initiatives have recently been launched to look into this.
For example, the idea of ‘microattribution’ has been proposed whereby database
systems would track the interest in each database record42, record this in relation to the
original submitter of the record, and thereby steadily assemble a metric for the value of
each person’s database contributions. Similarly, database creators would be able to
extrapolate from this kind of information something akin to a journal impact factor. This
will require cooperation of journals, database creators and funders, all of whom would
need to use an agreed tracking system. Additionally, success would depend upon there
being a way to assign globally-unique identifiers (GUIDs) to individual data packets
(e.g., database records) on the internet. Since the semantic web will also need such
identifiers (not only for data, but also for services, concepts, metadata, etc) several
solutions to this problem are now being evaluated. An alternative ‘Bio-Resource Impact
Factor’ (BRIF) approach to accreditation has been proposed43, the scope of which would
include G2P databases. BRIF is more directly akin to the journal Impact Factor. As
journals are steadily overtaken by databases as the preferred means for getting data
into the public domain, BRIF, or something like it, will be needed to demonstrate
researchers’ productivity and the importance of their work. It will also help in making it
evident to funding bodies that database creation efforts are worthy of support – a
message that presently needs some re-enforcement44.
Considerable sums of money are being spent on G2P research to improve our
understanding of health and disease so that medical care will advance. It follows
therefore that the G2P databases should remain tightly focused on the needs of the
medical community. The problem, however, is that whereas imprecise knowledge and
uncertain data are an essential part of research, the clinical world requires more
straightforward, reproducible information upon which to base its decision making.
Perhaps then, one of the real unmet challenges for G2P databases is that of distilling
from basic research data the kinds of watertight conclusions and predictions that would
help physicians diagnose a patient. Close attention needs to be paid to how this
information is presented, as researchers and clinicians typically have very different
expectations when searching for information. These important issues are well known to
those who work at the interface of bioinformatics and medical informatics. Examples of
key efforts designed to close the gap between the two include the development of
electronic healthcare records (EHR), and the work of the Health Level Seven (HL7)
organization towards genomics standards in medical information.
In summary, the field of G2P databasing is at a significant stage in its development,
taking into account lessons from the past, and being challenged by the exponentially
growing and rapidly changing datasets of the present and the future. In this review, we
have tried to cover the technical solutions and the logical ways forward for the field, all
of which are buttressed by extensive open-source collaboration and contribute to the
emerging ‘cyberinfrastructure for the biological sciences’45. We can therefore expect
G2P databasing to advance significantly in the near future, enabling research and
clinical practitioners to make the very best use of the wealth of G2P data now being
generated.
Figure Legends
Figure 1: Extreme Models for Database Integration
Radical forms of centralized and federated database networks are illustrated on either
side of this image. In the centralized model on the left, outstations or ‘nodes’ (small
circles) merely gather and prepare data for transfer to a massive central ‘hub’ (large
circle), where it is stored, integrated, and made available for searching and
presentation. In the federated model on the right, the outstations are replaced by fully-
functional databases (the eight medium sized circles) that gather and expertly curate
data, provide various means for human and machine based searching/accessing of this
information, and offer a range of data presentation options. In this latter model, the hub
receives no data from the nodes, but undertakes the important job of coordinating the
nodes and brokering searches across the other databases. The genotype-to-phenotype
database network of the future will probably be based upon a hybrid of these two
extreme models.
Figure 2: Databases and Database Networks
Early genotype-to-phenotype (G2P) databases were based upon many different designs
(indicated by differentially colored circles) with very few connections between them
(indicated by lines joining the circles). As the field develops, databases will instead be
built upon more standardized design and operation principles, enabling extensive inter-
connectivity between projects. Each resource in the resulting network may have an
emphasis upon being a data storage ‘node’ (smaller circles) or a data searching ‘hub’
(larger circles), or a combination of the two. It is hoped that emerging semantic web
technologies will develop the network further into a powerful and seamless G2P
knowledge environment.
Figure 3: Success Depends Upon Recognition and Reward
The utility of any future genotype-to-phenotype database network and its supporting
infrastructure will depend upon how effectively information gets into that system.
Individuals responsible for establishing and operating this ‘data flow’ - from the wet
bench scientist that produces the raw data, right through to the people that make the
integrated datasets available for searching and access - will all need to be recognized,
rewarded, and thereby motivated to play their part. Mechanisms for achieving this in the
context of databases (as opposed to data publication via journals) are yet to be put in
place.
Box 1: Technologies in Genotype-to-Phenotype Databasing
Databases and database networks will be key to organizing, storing, and providing
access to the wealth of biomedical data already produced and yet to be generated. The
task of building the necessary databases is primarily a technological construction effort,
since the required technology solutions are already well developed or at least identified
in principle. Some of the core concepts behind these technologies are listed below:
Object (Data) model - A formalized conceptualization of how data elements, or objects,
are structured and organized, and how they are connected to other data elements. This
may include semantic information on those objects and connections, by way of
references to ontologies (see below). An example is the Microarray Gene Expression
Object model46 (MAGE-OM) which standardizes the representation of micro-array
information, spanning experiment design and data.
Exchange format – The specifications of the syntax, or physical representation, of data
complying with the model. This is essential for unambiguous transmission of data
between computers. Examples range from the simple FASTA format used to exchange
DNA and protein sequence data, through to the elaborate XML-based MAGE-ML for
MAGE-OM compliant microarray data.
Ontology - A controlled vocabulary of terms for concepts, including their meaning and
well-defined relationships between them. Ontologies enable the representation of
domain-specific knowledge and, when used properly, make database searches far more
powerful. Examples include Gene Ontology47 (GO) for annotating gene products from
many species, and FuGO48 for functional genomics investigations.
Globally-unique identifier (GUID) - A digital object identifier, which is guaranteed to be
unique and persistent across the intended usage domain. GUIDs solve data integration
problems that result from ambiguity in names, or identity, of biological concepts and
objects, such as genes and proteins. GUIDs are key component of Semantic Web
technologies such as the Resource Description Framework (RDF:
http://www.w3.org/RDF). Examples currently being evaluated include Persistent URLs
(http://www.purl.org) and Life Science Identifiers49 (LSIDs).
Web services - A series of standard protocols for facilitating machine-to-machine
interaction over the Internet. Web services simplify the task of ‘plumbing together’
distributed data retrieval or analysis services over the network, forming the basis of the
service-oriented architecture (SOA: http://www.w3.org/TR/2003/WD-ws-arch-
20030514/). An example of a service-oriented ‘Grid’ is provided by the National Cancer
Institute’s caGrid50.
Molecular
_sample
Individual
Variation
_assay
Assayed_genomic_
genotype
Genomic_
variation
Genomic
_allele
Individual
Genotype
Marker
Allele
a) MIQAS
b) PaGE-OM
Box 2: Harmonising Data Models
Data models can often be aligned or ‘mapped’ to each other to identify similarities and
differences. Once this is done, and equivalent concepts are relationships thereby
identified, it is then possible to specify a consensus model and/or derive a data
exchange format with which both models will be compatible.
This is illustrated in the figure for sub-portions of two related data models, a) the
Minimum Information for QTLs and Association Studies specification (MIQAS:
http://miqas.sourceforge.net), and b) the Phenotype and Genotype Experiment Object
Model (PaGE-OM: http://www.pageom.org). Equivalent entities in these models are in
some cases named differently (e.g., ‘Marker’ and ‘Genomic_variation’), and so to
highlight the corresponding item pairs they have been placed in the same relative
positions in the diagrams and are shown in the same colour. Naming discrepancies are
problematic, especially when the same name is used to mean different things between
models (e.g., ‘Sample’ for a person or a reagent). Such confusion is eliminated when
models include semantic information on their components, by references to ontologies.
Solid lines in the above illustrations indicate relationships (e.g., a ‘Marker’ has ‘Alleles’),
and the dotted lines in PaGE-OM indicate that relationships may be optional to allow for
data elements that might not be known (e.g., the Variation_assay used in a genotyping
experiment). Based upon such mapping diagrams, data exchange format can be
specified that support only the common components from the models, or they may be
extended to include some non-common items whereupon those data elements would be
declared optional.
Similar inter-model mapping can be done between sub-domains where the underlying
models may be quite different, so long as the models have at least some common
concepts or attributes. Datasets produced according to those models may then be
connected together, rather than fully merged, by linking through those common fields on
data rows where the values are identical (e.g., a SNP marker identifier), as explained
elsewhere51.
Table 1: Genotype-to-Phenotype Database Infrastructure and Coordination
Projects
BBMRI
(http://www.biobanks.eu/)
ESFRI program for preparing to construct a pan-European
Biobanking and Biomolecular Resources Research
Infrastructure. An ESFRI project
caBIG
(https://cabig.nci.nih.gov/)
The Cancer Biomedical Informatics Grid. Data integration
network and application infrastructure developed for the
cancer research community.
CASIMIR
(http://www.casimir.org.uk/)
Coordination and Sustainability of International Mouse
Informatics Resources. EU funded project on co- ordination
and integration of mouse model organism databases.
EATRIS
(http://www.eatris.eu)
The European Advanced Translational Research
Infrastructure in Medicine. An ESFRI project aimed at
translating research findings into improved diagnosis,
disease prevention, and treatment.
ECRIN
(http://www.ecrin.org/)
European Clinical Research Infrastructures Network. An
ESFRI program aimed at integrating national clinical
research facilities into a pan-European infrastructure.
EGEE
(http://www.eu-egee.org/)
Enabling Grids for E-sciencE. Large EU funded
multidisciplinary infrastructure project.
ELIXIR
(http://www.elixir-europe.org)
European Life-science Infrastructure for Biological
Information. EU funded bioinformatics infrastructure
program for life science research. An ESFRI project.
EMBRACE
(http://www.embracegrid.info)
European Model for Bioinformatics Research and
Community Education. Collaboration network in area of
Grid computing and databasing in biomolecular research.
ESFRI
(http://cordis.europa.eu/esfri/)
The European Strategy Forum on Research Infrastructures.
EUROGENETEST
(http://www.eurogentest.org/)
EU-funded Network of Excellence fostering standardization
and harmonization of genetic testing across Europe
including informatics, ethics, new technologies, education
and quality management.
GEN2PHEN
(http://www.gen2phen.org)
Genotype to Phenotype databases. An EU funded project
aiming to unify human and model organism G2P databases
towards increasingly holistic views into this information.
GMOD
(http://gmod.org)
Generic Model Organism Database project. A collection of
open source software tools for creating and managing
genome-scale biological databases.
HL7
(http://www.hl7.org/)
Standardization organization operating on healthcare arena.
HL7 V3 is messaging standard developed by the HL7 for
health care domain
HuGENet
(http://www.cdc.gov/genomics/hugenet/)
Human Genome Epidemiology Network. International
collaboration of individuals and organization in field of
genetic epidemiology.
HVP
(http://www.humanvariomeproject.org/)
An open organisation aiming to help catalogue all human
Mendelian genetic variation, making that information freely
available to researchers, clinicians and patients worldwide.
MIBBI52 (http://www.mibbi.org)
Minimum Information for Biological and Biomedical
Investigations is a project and web resource promoting the
development and use of minimum information specifications
and checklists.
OBIBA
(http://www.obiba.org/)
Obiba is an open source project whose aim is to build an
open software infrastructure applications and software
components for biobanking. One of the P3G core projects.
OBO
(http://obofoundry.org/)
The Open Biomedical Ontologies. International community
for supporting development and use of ontologies in
biomedical domain.
OpenEHR
(http://www.openehr.org)
Data standards and modeling framework for managing
electronic health care data.
P3G
(http://www.p3gconsortium.org/)
The Public Population Project in Genomics. An international
consortium promoting collaboration between population
genomics researchers. The P3G observatory
(http://www.p3gobservatory.org) provides a central
repository of relevant tools and information.
References
1. Wheeler, D.L. et al. Database resources of the National Center for Biotechnology
Information. Nucleic Acids Res 35, D5―12(2007).
2. Hubbard, T. et al. The Ensembl genome database project. Nucleic Acids Res 30, 38―41(2002).
3. Kent, W.J. et al. The Human Genome Browser at UCSC. Genome Res. 12, 996―1006(2002).
4. Stein, L. Creating a bioinformatics nation. Nature 417, 119―120(2002).
5. Miyazaki, S. et al. DDBJ in the stream of various biological data. Nucleic Acids Research 32,
D31―34(2004).
6. Benson, D.A. et al. GenBank. Nucleic Acids Research 36, D25―30(2008).
7. Kanz, C. et al. The EMBL Nucleotide Sequence Database. Nucleic Acids Research 33,
D29―33(2005).
8. Chen, N. et al. WormBase: a comprehensive data resource for Caenorhabditis biology and
genomics. Nucleic Acids Res 33, D383―389(2005).
9. Twigger, S.N. et al. The Rat Genome Database, update 2007―easing the path from disease
to data and back again. Nucleic Acids Res 35, D658―662(2007).
10. Bult, C.J. et al. The Mouse Genome Database (MGD): mouse biology and model systems.
Nucleic Acids Res 36, D724―728(2008).
11. Hamosh, A. et al. Online Mendelian Inheritance in Man (OMIM), a knowledgebase of human
genes and genetic disorders. Nucleic Acids Res 33, D514―517(2005).
12. McKusick, V.A. Mendelian Inheritance in Man. A Catalog of Human Genes and Genetic Disorders.
(Johns Hopkins University Press: 1966).
13. Ball, E.V. et al. Microdeletions and microinsertions causing human genetic disease:
common mechanisms of mutagenesis and the role of local DNA sequence complexity. Hum
Mutat 26, 205―213(2005).
14. Altman, R.B. PharmGKB: a logical home for knowledge relating genotype to drug response
phenotype. Nat Genet 39, 426―426(2007).
15. Lehmann, H. & Kynoch, P.A.M. Human haemoglobin variants and their characteristics. (North-
Holland Publishing: Amsterdam, 1976).
16. Horaitis, O. et al. A database of locus-specific databases. Nat Genet 39, 425(2007).
17. Mailman, M.D. et al. The NCBI dbGaP database of genotypes and phenotypes. Nat Genet 39,
1181―1186(2007).
18. Becker, K.G. et al. The Genetic Association Database. Nat Genet 36, 431―432(2004).
19. Bertram, L. et al. Systematic meta-analyses of Alzheimer disease genetic association
studies: the AlzGene database. Nat Genet 39, 17―23(2007).
20. Allen, N.C. et al. Systematic meta-analyses and field synopsis of genetic association studies
in schizophrenia: the SzGene database. Nature Genetics 40, 827―834(2008).
21. Mardis, E.R. The impact of next-generation sequencing technology on genetics. Trends
Genet 24, 133―141(2008).
22. Howe, D. et al. Big data: The future of biocuration. Nature 455, 47―50(2008).
23. Goble, C. & Stevens, R. State of the nation in data integration for bioinformatics. J Biomed
Inform. 41, 687-93(2008)
24. Knoppers, B. et al. Population Genomics: The Public Population Project in Genomics
(P(3)G): a proof of concept? Eur J Hum Genet 16, 664-665(2008).doi:10.1038/ejhg.2008.55
25. Ioannidis, J.P.A. et al. A road map for efficient and reliable human genome epidemiology.
Nat Genet 38, 3―5(2006).
26. Elnitski, L.L. et al. The ENCODEdb portal: simplified access to ENCODE Consortium data.
Genome Res 17, 954―959(2007).
27. Hoyweghen, I.V. & Horstman, K. European practices of genetic information and insurance:
lessons for the Genetic Information Nondiscrimination Act. JAMA 300, 326―7(2008).
28. Diergaarde, B. et al. Genetic information: Special or not? Responses from focus groups with
members of a health maintenance organization. Am J Med Genet A 143, 564―9(2007).
29. Gilbar, R. Patient autonomy and relatives' right to know genetic information. Medicine and
law 26, 677―97(2007).
30. Knoppers, B.M. et al. The emergence of an ethical duty to disclose genetic research results:
international perspectives. Eur J Hum Genet 14, 1170―8(2006).
31. Godard, B. et al. Data storage and DNA banking for biomedical research: informed consent,
confidentiality, quality issues, ownership, return of benefits. A professional perspective.
Eur J Hum Genet 11 Suppl 2, S88―122(2003).
32. Homer, N. et al. Resolving individuals contributing trace amounts of DNA to highly
complex mixtures using high-density SNP genotyping microarrays. PLoS Genet 4,
e1000167(2008).
33. Cambon-Thomsen, A., Rial-Sebbag, E. & Knoppers, B.M. Trends in ethical and legal
frameworks for the use of human biobanks. Eur Respir J 30, 373―382(2007).
34. Zerhouni, E.A. & Nabel, E.G. Protecting aggregate genomic data. Science 322, 44(2008).
35. Giardine, B. et al. PhenCode: connecting ENCODE data with mutations and phenotype. Hum
Mutat 28, 554―562(2007).
36. Stein, L.D. Integrating biological databases. Nat Rev Genet 4, 337―345(2003).
37. Stevens, R., Goble, C.A. & Bechhofer, S. Ontology-based knowledge representation for
bioinformatics. Brief Bioinform 1, 398―414(2000).
38. Quackenbush, J. Standardizing the standards. Mol Syst Biol 2, 2006.0010(2006).
39. Smith, B. et al. The OBO Foundry: coordinated evolution of ontologies to support
biomedical data integration. Nat Biotechnol 25, 1251―1255(2007).
40. Dowell, R.D. et al. The Distributed Annotation System. BMC Bioinformatics 2, 7(2001).
41. Berners-Lee, T., Hendler, J. & Lassila, O. The Semantic Web - A new form of Web content
that is meaningful to computers will unleash a revolution of new possibilities. Scientific
American 284, 34―43(2001).
42. Editors Compete, collaborate, compel. Nature Genetics 39, 931(2007).
43. Kauffmann, F. & Cambon-Thomsen, A. Tracing biological collections: between books and
clinical trials. JAMA 299, 2316―2318(2008).
44. Merali, Z. & Giles, J. Databases in peril. Nature 435, 1010―1011(2005).
45. Stein, L.D. Towards a cyberinfrastructure for the biological sciences: progress, visions and
challenges. Nat Rev Genet 9, 678―88(2008).
46. Spellman, P.T. et al. Design and implementation of microarray gene expression markup
language (MAGE-ML). Genome Biol 3, research0046.1-0046.9(2002).
47. The Gene Ontology Consortium. Gene Ontology: tool for the unification of biology. Nat
Genet 25, 25―29(2000).
48. Jones, A.R. et al. The Functional Genomics Experiment model (FuGE): an extensible
framework for standards in functional genomics. Nat Biotechnol 25, 1127―1133(2007).
49. Clark, T., Martin, S. & Liefeld, T. Globally distributed object identification for biological
knowledgebases. Brief Bioinform 5, 59―70(2004).
50. Saltz, J. et al. caGrid: design and implementation of the core architecture of the cancer
biomedical informatics grid. Bioinformatics 22, 1910―1916(2006).
51. Wang, X., Gorlitsky, R. & Almeida, J.S. From XML to RDF: how semantic web technologies
will change the design of 'omic' standards. Nat Biotechnol 23, 1099―1103(2005).
52. Taylor, C.F. et al. Promoting coherent minimum reporting guidelines for biological and
biomedical investigations: the MIBBI project. Nat Biotechnol 26, 889―896(2008).
Acknowledgements
The authors acknowledge the valuable ideas, advice, and funding, provided by the
GEN2PHEN project as part of the European Community's Seventh Framework
Programme (FP7/2007-2013) under grant agreement number 200754, that enabled the
preparation of this review.
Current Databases;
disparate silos
Future databases;
an inter-connected network
Figure 1
Central database model Federated database model
Figure 2
Data
production
& analysis
Data
packaging
& release
Data
deposition
into the
WWW
Data
aggregation
or federation
Data search
&
presentation
Data flow
Accreditation and Reward
Figure 3
