1. DATA MANAGEMENT AND CURATION AT TAIR
Margarita Garcia-Hernandez

2. The ‘systems biology’ paradigm
FACT: huge amounts of data
NEED: systematic harvesting & easy accessibility (store, sort, interlink)
PROBLEM: complexity & heterogeneity of data
CHALLENGE: to describe complete biological systems in an integrated way (organizing, defining relationships, defining metadata standards, interpreting, quality control assessment – DATA MANAGEMENT)
With the advent of new techniques in biology we are experiencing an explosion of data similar to that seen in other disciplines, such as astronomy. In order for regular researchers to be able to not only have easy access to the data, but also be able to make use of it in a comprehensive way (use the data to make hypothesis). The problem is the data is not only complex, but also coming from diverse sources. The most accepted solution to deal with this is to use relational databases, where the data is stored in a way where relationships between different data classes can be established and maintained. The challenge is how model and manage the data , is that, how to organize it, define the relationships among them, standardize it and provide quality control assessments. My talk will focus on how we address this at TAIR.

3. Data Management Flow Chart
Data generation
Collection
Selection
Organization of similar data types
Remove redundancy
Correct errors
Data curation
Association of different data types
Establish unambiguous identifiers
Define and validate relationships
This cartoon represents a overview of the series of steps involved in management of data, from data generation to their dissemination to the public. Starting with the collection of the data from the source/s, it may follow a selection, for example to exclude low quality or not relevant data.
Resolve data heterogeneity - standardization
Annotation – add descriptions
Define standard vocabulary
Data modeling
Database population
Data dissemination

4. Quality Control Issues TAIR’s approaches Accuracy of information
Consistency in format and content
Up-to-datedness
Conflicting data
TAIR’s approaches
Personnel training (Ph.D. level biologists)
User input
Source attribution
Checking curation consistency (computationally and manually)
Adopt Standard Operation Procedures
Define and use controlled vocabularies
An important issue in data curation is dealing with quality control issues, and most importantly how is quality recognized. How good is the data and how to recognize god from bad data. There are several issues related to quality control. One is how accurate is the information, second how consistent it is, both in format and, third how up-to date (date changes continuously), and last, but not least, how to deal with conflicting data

5. Many Data Types with Many Sources
Literature
Public Databases
Community
submissions
Computational
analysis
Functional
Genomic
projects
Genes/
Gene Products
Mutant
Phenotypes
Expression
Metabolism
Stocks

6. Two Examples of Data Curation
Sources
Data Types
Literature
Genes/
Gene Products
Public Databases
(SMD)
Mutant
Phenotypes
Community
submissions
Microarray
Expression
Functional
Genomic projects
(AFGC)
Metabolism
Computational
analysis
Stocks

7. Literature Curation PubSearch
A literature curation management system designed to store and manage the available literature for an organism of interest
PubSearch software is freely available at
Generic Model Organism Database (GMOD)
a joint effort by several model organism databases to develop reusable components for creating new biological databases
PubSearch is a literature curation management system designed to store and manage the available literature for an organism of interest. It is one of the GMOD tool kits. GMOD is a joint effort by several model organism databases to develop reusable components for creating new biological databases

8. Literature Curation Step 1: Collection of References
Meetings abstracts
Dissertations
Textbooks
Full text papers
(scanning, online)
Biosis
PubMed
Agricola
(‘Arabidopsis’ in title or abstract)
Arabidopsis References
Remove redundancy
Journal names standardization
PubSearch DB
(21,527)
(curation tool)
TAIR DB
(public db)

9. Literature Curation Step 2: Assigning References to Genes
Arabidopsis References
(PubSearch)
known
gene
names
Scanning references for terms in list
(programmatically)
Term
List
(17,470)
candidate
gene
names
Ref 1
Ref 2 ..
Ref n
Gene X
Reference hit Validation
(by curators using PubSearch)
Validated list of references for each gene

10. Literature Curation Step 3: Extracting information
Gene-centric curation approach
Each curator is assigned 2 genes per day
Papers are read and information extracted (following SOPs and using PubSearch curation tool):
Name validation & add aliases
Add sequence info
Assign locus (mapping to the genome by BLAST)
Merge/split genes
Write summary sentence
Correct errors
Annotation using controlled vocabularies (GO, POC)

11. Controlled Vocabularies
A collection of defined terms (organized in a hierarchy) intended to serve as a standard nomenclature
Provide a common set of terms that users of a single system (or across multiple systems) can share
Allows retrieval of ALL relevant information
Example:
Find all the genes that have transporter activity (regardless of how they are named, or what type they are)
It is very important to assure consistency.

12. Controlled Vocabularies used at TAIR
Gene Ontology (GO)
Goal: to produce a controlled vocabulary for describing genes and proteins that
can be applied to all organisms
Molecular Function
Cellular component
Biological process
Plant Organism Consortium (POC)
Gramene, TAIR, Univ Missouri St Louis, MaizeDB, IRIS, MIPS, Oryzabase & Monsanto & Pioneer as collaborators
Goal: to develop structured controlled vocabularies for plant-specific knowledge domains:
Plant Anatomy (morphology, organs, tissue and cell types)
Temporal stages (plant growth and developmental stages)
Phenotype Ontology (in the works)

13. Qualifying Annotations with supporting evidence
References
Evidence code usage
A set of controlled vocabulary, which provides evidence to support the association between gene products and annotations
IDA: Inferred from Direct Assay
IMP: Inferred from Mutant Phenotype
ISS: Inferred from Sequence Similarity
IEA: Inferred from Electronic Annotation
IEP: Inferred from Expression Pattern ……
Evidence code description
E.g., IPI : Inferred from Physical Interaction
Co-immunoprecipitation
Co-purification
Co-sedimentation ….

14. Gene Annotation Display in TAIR

15. QC of Literature Curation
Weekly annotation meeting
Quality control manager
Use of standardized vocabularies
Random checks of annotations
Annotations are tagged by date and curator
Automatic checks in software
Use SOPs – curation guidelines

16. Curation of Microarray Data
Sources
Data Types
Literature
Genes/
Gene Products
Public Databases
(SMD)
Mutant
Phenotypes
Community
submissions
Microarray
Data
Functional
Genomic projects
(AFGC)
Metabolism
Computational
analysis
Stocks

17. Curation of AFGC Microarray Data Data Collection and Selection
Arabidopsis
Functional
Genomics
Consortium
Stanford
Microarray
Database
sample info
proposal abstracts
protocols
results
array design
minimal descriptions of individual arrays
- All Arabidopsis public arrays
- exclude QC arrays (45)
Selected Arrays
(516)
Metadata
Numeric Results Data
(raw and normalized)
Array
Elements
Protocols
Samples
Experiments

18. Curation of Metadata: Array elements
1. classify, organize, add missing sequences, correct errors
2. mapping to the Arabidopsis genome & association to genes (pipeline)
Samples & Experiments
1. Data extraction from flat files (abstracts, RNA forms), and database (SMD)
e.g., tissue type, treatments, experimental design
2. Organization of data & parsing into tables
3. Develop controlled vocabularies for experiment categorization & treatments
4. Standardization using those vocabularies
5. Data association
grouping arrays  replicate sets  experiments
merging replicate samples to minimize redundancy
linking to other related data (germplasm, clones, publications, people)
6. Annotation
Experiments: GO process, category, experimental variables
Samples: tissues (POC anatomy & temporal) & treatment
Data Submission

19. Curation of Microarray Results Data
Numeric Results Data
-Quality control
Remove poor quality arrays (2)
Exclude spots flagged as bad
Re-normalize using lowess method (minimize spatial bias)
Remove arrays with strong spatial/plate bias (72)(ANOVA)
Exclude array elements with intensity < 350 in both channels
Exclude array elements with null values in 80% of arrays
-Analysis
Calculate log2 ratio [ch2N/ch1N]
Calculate fold change [ch2N/ch1N]
Calculate averages for each array element (array & replicates)
Element
fold change/log2 ratio
std error
per array
Element
fold change/log2 ratio
std error
per replicate arrays

20. Conclusions Requires trained biologists familiar with data
Can be facilitated computationally (repetitive tasks), but is mainly a knowledge-based task that can only be done by humans
Essential for assuring data quality
Adds value to data
Slow process
Can be inconsistent