An Introduction to Databases and Data Standards for Metabolomics

1
An Introduction to Databases and Data Standards
for Metabolomics

• Dr Helen Jenkins
• Computational Biology Research Group
• haf_at_aber.ac.uk

2
This lecture will cover

• A brief introduction to the -omes and the
  -omics.
• Description of the projects that we are involved
  in and our remit within them.
• Discussion of the requirements for metabolomics
  experiment databases.
• A brief description of our solution.

3
The -omes and -omics (1)

• Genome The complete set of genes in an organism.
• Genomics Identification of genes and sequencing
  their exact order.
• Transcriptome The set of RNA transcripts
  produced by a genome at any one time.
• Transcriptomics Study of which genes are
  switched on when.

4
The -omes and omics (2)

• Proteome The complete set of proteins produced
  by a genome at any one time.
• Proteomics Study of the proteome.
• Metabolome The complete set of metabolites
  produced by a genome at any one time.
• Metabolomics Study of the metabolome

5
Metabolomics
Comprehensive (?) estimation of the low molecular
weight compounds (e.g. sugars) in a biological
sample.

• Example applications
• Basic biology
• Food quality (human/animal)
• Clinical trials
• Examples of samples
• Plant material
• Blood/urine from humans and animals
• Microbial material (yeast/fungi)
• Samples may be taken from
• The organisms themselves or
• Their environment (footprinting)

6
Our projects (1) Metabolome technology for the
profiling of GM and conventionally bred plant
materials

• 3 year project (Dec 2001 to Dec 2004)
• Funded by the UK Food Standards Agency (FSA).
• Collaboration between DCS and IBS at Aberystwyth
  and the Max Planck Institute of Molecular Plant
  Physiology in Golm, Germany.
• Aim To build a technology platform for
  performing metabolomics to assess the differences
  between GM and conventionally bred versions of
  plant material.
• Using Arabidopsis thaliana and potato as model
  plants.
• Our remit To build a robust database for
  metabolome and related data

7
Our projects (2) Hierarchical plant metabolomics
(HiMet)

• 3 year project (Apr 2003 to Apr 2006)
• Funded by the Biotechnology and Biological
  Sciences Research Council (BBSRC)
• Collaboration between DCS and IBS at Aberystwyth,
  the John Innes Centre in Norwich, Rothamsted
  Research in Harpenden and CNAP at York University
• Aim To understand metabolome patterns through
  analysis of the data produced by high-throughput
  chemical analyses of metabolism mutants.
• Using mutant lines of Arabidopsis thaliana.
• Our remit To build a database to store
  information about plant metabolomics experiments
  and their results and facilitate statistical and
  data mining analysis of the datasets.

8
Our projects (3) The UK Centre for Plant and
Microbial Metabolomics (MeT-RO)

• 4 year project (Dec 2004 to Dec 2008)
• Funded by the BBSRC
• Collaboration between Rothamsted Research in
  Harpenden, DCS and IBS at Aberystwyth and the
  Department of Chemistry at the University of
  Manchester.
• Aims
• To establish and operate a high-throughput,
  state-of-the-art global metabolome
  fingerprinting (as distinct from metabolic
  profiling) service - accessible on a user pays
  basis to all members of the biological sciences
  community.
• Carry out research and training in plant and
  microbial metabolomics and associated
  bioinformatics.
• Our remit To build a robust data handling
  platform for storing and querying the datasets
  produced by experiments carried out through the
  centre.

9
Understanding metabolomics data (1) The
different types of omic data

• Reference (standard) data
• Gene sequences
• Gene expressions
• Proteins
• Metabolites
• Experiment data
• Data from particular experiments, e.g. an
  experiment to assess the effect of changes in the
  metabolome of a particular organism during times
  of drought.

10
Understanding metabolomics data (2) Biological
signal versus experimental noise

• Metabolomics experiments aim to measure
  biological signal
• Organisms differ.
• But the process of performing an experiment may
  induce noise
• Unintended side-effects of the experimental
  process.
• Study of the metabolome provides us with a means
  for determining the extent of a biological signal
  BUT the dynamic nature of the metabolome means
  that experimentally induced noise may well
  obscure the signal.
• Proper understanding of experimental context is,
  therefore, key to
• Correct interpretation of experimental results.
• Selection of datasets that may be meaningfully
  compared.
• Experiment reproducibility.

11
Understanding metabolomics data (3) The
metabolomics experiment pipeline
12
Understanding metabolomics data (4) Data from
GC-MS
Output
GC-MS Machine
13
Understanding metabolomics data (5) Processing
the output from GC-MS
Metabolite Area Peak one xxx Peak two xxx Peak
three xxx Peak four xxx
Noise Removal Peak Identification Peak
Deconvolution Peak Quantification
Quantified Peak List
Peak Labelling
Chemical Identity
14
Understanding metabolomics data (6) Metabolome
signatures
FT-IR Output
DI-ESI Output
15
Designing a metabolomics database

• From our understanding of the nature of
  metabolomics data we can define the following
  list of requirements for a metabolomics database
• It should support data from all parts of the
  experiment pipeline.
• It should support a range of experimental
  procedures as used by different laboratories.
• It should support the range of dataset types as
  produced by different chemical analysis
  technologies.
• It should facilitate sophisticated querying for
  datasets and mining over experimental metadata.
• Therefore, our database design must be both
  flexible and extensible.

16
A component-based structure

• Our solution was to specify a component-based
  structure for our data model
• A set of components, each of which aims to
  describe a different part of a plant metabolomics
  experiment.
• Each component has a set of core, required data
  items, relevant to all plant metabolomics
  experiments.
• Each component may have one or more
  sub-components that extend the core data for the
  component to describe particular methodologies
  and/or technologies.
• An experiment is then described by a complete set
  of components each one of which may contain
  either
• The core data for that component or
• The core data for that component plus additional
  data items specified in one of its
  sub-components.

17
ArMet (Architecture for Metabolomics) components
Metabolome Estimate


Genotype
DataPoint
Output
18
Components and sub-components
19
A general Metabolome Estimate component
Metabolome Estimate
Targeted output


Genotype
Metabolite
Output
and/or
Metabolome output
Peak
Output
Fingerprint output
Output
DataPoint
20
Recent Developments
Protocol repository
Protocol log
21
This lecture will include

• A word on the importance of data standards for
  the omics and where our work fits in.
• Discussion of some of the issues we have faced
  whilst implementing our databases.
• A brief look at where this field is going in the
  future.

22
Data standards

• Typically experimentalists will want to be able
  to do one or many or the following
• Transmit datasets to collaborators or other
  external sites.
• Store data with adequate curation.
• Make datasets available for statistical analysis
  and data mining.
• Data standards support these activities by
  defining
• The structure of datasets.
• The types of information that they should contain.

23
Data standards for the -omics

• Transcriptomics
• MIAME (Nat. Genetics., 29365-371) defines the
  data that should be recorded for a transcriptome
  experiment.
• MAGE (Spellman et. al., Genome. Biol., 3, 2002)
  is an associated formal data description.
• Proteomics
• PEDRo (Nat. Biotechnol., 21247-254) is a formal
  data description for proteomics.
• PSI-OM (Proteomics, 4490-491) is the result of
  further development of PEDRo by HUPO PSI
• MIAPE is an associated definition of the data
  that should be recorded for proteomics
  experiments.

24
Data standards for metabolomics

• Until recently there was no equivalent
  metabolomics-specific data standard for defining
  metabolite complements in their experimental
  context. Now we have
• SMRS (http//www.smrsgroup.org/) is a draft
  policy for standardisation of reporting for
  metabolic analysis mainly for pre-clinical drug
  trials
• MIAMET (Trends in Plant Science, 9(9)418-425) is
  a definition of the data that should be recorded
  for a metabolomics experiment
• ArMet (Nat. Biotechnol., 221601-1606) is a
  formal data description for plant metabolomics.

25
Benefits of data standards

• Data standards such as these, that describe
  experiment results in their experimental context,
  enable
• Proper interpretation of results.
• Laboratory interoperability.
• Meaningful comparison of datasets.
• Replication of experiments.
• More options for retrospective analysis of data.

26
Other benefits of data standards

• Data standards, and the formal data descriptions
  that underlie, them
• Encourage consideration and development of best
  practice and Standard Operating Procedures.
• Standardize the reporting of experiments and
  archiving of data associated with publications.
• Enable the development of databases and
  verifiable transmission mechanisms to facilitate
  the storage, collection and dissemination of
  logically correct datasets.

27
Implementation Issues

• Speed and optimisation of querying
• Building acceptable user interfaces for data
  submission
• Standardising terminology

28
Speed of querying

• Metabolomics experiments can produce vast
  quantities of results data.
• For example
• Each experiment within our HiMet project involves
  220 samples.
• There are currently 11 experiments.
• Each sample is subjected to multiple types of
  chemical analysis.
• Each dataset produced from a single chemical
  analysis contains 2000 datapoints.
• All of this results data is, of course, in
  addition to the experiment metadata that provides
  the experimental context for the results.
• Storing all of these points in a single unordered
  database table can lead to query times of hours
  for the data from a single experiment or sample.

29
Optimisation (1) Database tuning

• A key aspect of the development of relational
  databases is normalisation
• Rules for grouping the fields that we want to
  store into tables
• Denormalisation involves reducing the degree of
  normalisation of the tables in a relational
  database to improve performance
• Should not be done when a table is regularly
  updated as normalisation protects us against
  insert/update/delete anomalies that can reduce
  the integrity of a database.

30
Optimisation (2) Data Organisation

• Hash files
• Records in a table are written to disk based on a
  hash function.
• Indexes
• A structure that is separate from, but
  associated with, a table contains an entry for
  each value in the indexed column(s) of the table
  provides direct pointers to the rows in the
  table on the disk.
• B-trees
• Records in a table/index organised into a
  balanced-tree structure.
• Clustered tables
• Groups of one or more tables physically stored
  together because they share common columns and
  are often used together.

31
Optimisation (3) Distribution

• Table partitioning
• Break tables or index files down into smaller
  more manageable chunks that may be distributed
  across disks.
• Parallelising queries
• Breaking down the query into a number of
  functions that are carried out by different
  processes.

32
User Interface Issues

• Experimentalists gather the data about their
  experiments in a number of formats
• Lab books
• Protocol documentation
• Spreadsheets for material tracking
• Instrument-specific data files for chemical
  analysis output.
• Initially we produced web-based, form-filling
  interfaces for metadata submission linked to
  upload facilities for directly uploading dataset
  files, but
• The detail required to capture the sources of
  experimentally induced noise makes this type of
  form-filling interface overwhelming.

33
Standardising Terminology

• Crucial to the production of standard datasets
  that can be compared and interpreted correctly is
  the use of common (standard) terminology to
  describe the concepts involved in an experiment.
• Many biological terms have other meanings when
  used in other situations, e.g. everyday meaning,
  chemical meaning.
• In our solutions so far we have employed the
  notion of extensible controlled vocabularies.
• Certain key fields within the database must take
  their values from a controlled list
• The controlled list entries include information
  on the authority for that term, i.e. the
  standard definition or meaning.

34
The Future

• Omic ontologies
• Pan-omic data models

35
Ontologies and controlled vocabularies

• An ontology is a hierarchical formal
  specification of the concepts within a domain. It
  specifies the vocabulary that may be used to
  refer to those concepts and how the concepts may
  be related.
• Controlled vocabularies are required by
  MIAME/MAGE, PEDRo/PSI-OM and ArMet.
• The development of ontologies will help to
  formalise these controlled vocabularies which are
  currently modelled as values with authorities.
• Both MGED in the microarray community and the PSI
  in the proteomics community have recognised this
  and are working on ontologies to support the
  description of transcriptomics and proteomics
  experiments.

36
Ontology for experiments (EXPO)

• Work carried out by colleagues at Aberystwyth
  (Larissa Soldatova and Prof. Ross King) has
  resulted in an ontology for experiments (EXPO).
• They are now planning to validate the ontology by
  showing that it may be instantiated for
  metabolomics experiments.
• Benefits to ArMet
• Formalisation of the required controlled
  vocabularies
• Alternative modelling approach is helping to
  inform the development of the ArMet data model.
• As EXPO is at a higher conceptual level than any
  of the omics (and could potentially be
  instantiated for transcriptomics and proteomics
  experiments as well) we are also hoping to
  evaluate its usefulness to the process of
  development of common models for sample
  description.

37
A Single Data Model for Sample Description

• Metabolomics analyses will be performed in
  conjunction with those on the transciptome and
  proteome in integrative experiments.
• Need a common data standard for experiment
  descriptions.
• Must contain the metadata required to fully
  evaluate the different omic analyses performed
  on samples from a single experiment
• Others working in this area have produced
• SysBio-OM (Bioinformatics, 20(12), 2004-2015).
• FGE-OM (Bioinformatics, 20(10), 1583-1590).